.

Co-authored-by: Claude Sonnet 4.6 <noreply@anthropic.com>
Co-authored-by: greptile-apps[bot] <165735046+greptile-apps[bot]@users.noreply.github.com>

.cursor/skills/onyx-cli/SKILL.md

@@ -0,0 +1,186 @@
+---
+name: onyx-cli
+description: Query the Onyx knowledge base using the onyx-cli command. Use when the user wants to search company documents, ask questions about internal knowledge, query connected data sources, or look up information stored in Onyx.
+---
+
+# Onyx CLI — Agent Tool
+
+Onyx is an enterprise search and Gen-AI platform that connects to company documents, apps, and people. The `onyx-cli` CLI provides non-interactive commands to query the Onyx knowledge base and list available agents.
+
+## Prerequisites
+
+### 1. Check if installed
+
+```bash
+which onyx-cli
+```
+
+### 2. Install (if needed)
+
+**Primary — pip:**
+
+```bash
+pip install onyx-cli
+```
+
+**From source (Go):**
+
+```bash
+cd cli && go build -o onyx-cli . && sudo mv onyx-cli /usr/local/bin/
+```
+
+### 3. Check if configured
+
+```bash
+onyx-cli validate-config
+```
+
+This checks the config file exists, API key is present, and tests the server connection via `/api/me`. Exit code 0 on success, non-zero with a descriptive error on failure.
+
+If unconfigured, you have two options:
+
+**Option A — Interactive setup (requires user input):**
+
+```bash
+onyx-cli configure
+```
+
+This prompts for the Onyx server URL and API key, tests the connection, and saves config.
+
+**Option B — Environment variables (non-interactive, preferred for agents):**
+
+```bash
+export ONYX_SERVER_URL="https://your-onyx-server.com"  # default: https://cloud.onyx.app
+export ONYX_API_KEY="your-api-key"
+```
+
+Environment variables override the config file. If these are set, no config file is needed.
+
+| Variable | Required | Description |
+|----------|----------|-------------|
+| `ONYX_SERVER_URL` | No | Onyx server base URL (default: `https://cloud.onyx.app`) |
+| `ONYX_API_KEY` | Yes | API key for authentication |
+| `ONYX_PERSONA_ID` | No | Default agent/persona ID |
+
+If neither the config file nor environment variables are set, tell the user that `onyx-cli` needs to be configured and ask them to either:
+- Run `onyx-cli configure` interactively, or
+- Set `ONYX_SERVER_URL` and `ONYX_API_KEY` environment variables
+
+## Commands
+
+### Validate configuration
+
+```bash
+onyx-cli validate-config
+```
+
+Checks config file exists, API key is present, and tests the server connection. Use this before `ask` or `agents` to confirm the CLI is properly set up.
+
+### List available agents
+
+```bash
+onyx-cli agents
+```
+
+Prints a table of agent IDs, names, and descriptions. Use `--json` for structured output:
+
+```bash
+onyx-cli agents --json
+```
+
+Use agent IDs with `ask --agent-id` to query a specific agent.
+
+### Basic query (plain text output)
+
+```bash
+onyx-cli ask "What is our company's PTO policy?"
+```
+
+Streams the answer as plain text to stdout. Exit code 0 on success, non-zero on error.
+
+### JSON output (structured events)
+
+```bash
+onyx-cli ask --json "What authentication methods do we support?"
+```
+
+Outputs JSON-encoded parsed stream events (one object per line). Key event objects include message deltas, stop, errors, search-start, and citation payloads.
+
+Each line is a JSON object with this envelope:
+
+```json
+{"type": "<event_type>", "event": { ... }}
+```
+
+| Event Type | Description |
+|------------|-------------|
+| `message_delta` | Content token — concatenate all `content` fields for the full answer |
+| `stop` | Stream complete |
+| `error` | Error with `error` message field |
+| `search_tool_start` | Onyx started searching documents |
+| `citation_info` | Source citation — see shape below |
+
+`citation_info` event shape:
+
+```json
+{
+  "type": "citation_info",
+  "event": {
+    "citation_number": 1,
+    "document_id": "abc123def456",
+    "placement": {"turn_index": 0, "tab_index": 0, "sub_turn_index": null}
+  }
+}
+```
+
+`placement` is metadata about where in the conversation the citation appeared and can be ignored for most use cases.
+
+### Specify an agent
+
+```bash
+onyx-cli ask --agent-id 5 "Summarize our Q4 roadmap"
+```
+
+Uses a specific Onyx agent/persona instead of the default.
+
+### All flags
+
+| Flag | Type | Description |
+|------|------|-------------|
+| `--agent-id` | int | Agent ID to use (overrides default) |
+| `--json` | bool | Output raw NDJSON events instead of plain text |
+
+## Statelessness
+
+Each `onyx-cli ask` call creates an independent chat session. There is no built-in way to chain context across multiple `ask` invocations — every call starts fresh. If you need multi-turn conversation with memory, use the interactive TUI (`onyx-cli` or `onyx-cli chat`) instead.
+
+## When to Use
+
+Use `onyx-cli ask` when:
+
+- The user asks about company-specific information (policies, docs, processes)
+- You need to search internal knowledge bases or connected data sources
+- The user references Onyx, asks you to "search Onyx", or wants to query their documents
+- You need context from company wikis, Confluence, Google Drive, Slack, or other connected sources
+
+Do NOT use when:
+
+- The question is about general programming knowledge (use your own knowledge)
+- The user is asking about code in the current repository (use grep/read tools)
+- The user hasn't mentioned Onyx and the question doesn't require internal company data
+
+## Examples
+
+```bash
+# Simple question
+onyx-cli ask "What are the steps to deploy to production?"
+
+# Get structured output for parsing
+onyx-cli ask --json "List all active API integrations"
+
+# Use a specialized agent
+onyx-cli ask --agent-id 3 "What were the action items from last week's standup?"
+
+# Pipe the answer into another command
+onyx-cli ask "What is the database schema for users?" | head -20
+```

.github/workflows/deployment.yml

@@ -182,9 +182,53 @@ jobs:
           title: "🚨 Version Tag Check Failed"
           ref-name: ${{ github.ref_name }}
 
-  build-desktop:
+  # Create GitHub release first, before desktop builds start.
+  # This ensures all desktop matrix jobs upload to the same release instead of
+  # racing to create duplicate releases.
+  create-release:
     needs: determine-builds
     if: needs.determine-builds.outputs.build-desktop == 'true'
+    runs-on: ubuntu-slim
+    timeout-minutes: 10
+    permissions:
+      contents: write
+    outputs:
+      release-id: ${{ steps.create-release.outputs.id }}
+    steps:
+      - name: Checkout
+        uses: actions/checkout@de0fac2e4500dabe0009e67214ff5f5447ce83dd # ratchet:actions/checkout@v6
+        with:
+          persist-credentials: false
+
+      - name: Determine release tag
+        id: release-tag
+        env:
+          IS_TEST_RUN: ${{ needs.determine-builds.outputs.is-test-run }}
+          SHORT_SHA: ${{ needs.determine-builds.outputs.short-sha }}
+        run: |
+          if [ "${IS_TEST_RUN}" == "true" ]; then
+            echo "tag=v0.0.0-dev+${SHORT_SHA}" >> "$GITHUB_OUTPUT"
+          else
+            echo "tag=${GITHUB_REF_NAME}" >> "$GITHUB_OUTPUT"
+          fi
+
+      - name: Create GitHub Release
+        id: create-release
+        uses: softprops/action-gh-release@da05d552573ad5aba039eaac05058a918a7bf631 # ratchet:softprops/action-gh-release@v2
+        with:
+          tag_name: ${{ steps.release-tag.outputs.tag }}
+          name: ${{ steps.release-tag.outputs.tag }}
+          body: "See the assets to download this version and install."
+          draft: true
+          prerelease: false
+        env:
+          GITHUB_TOKEN: ${{ secrets.GITHUB_TOKEN }}
+
+  build-desktop:
+    needs:
+      - determine-builds
+      - create-release
+    if: needs.determine-builds.outputs.build-desktop == 'true'
     permissions:
       id-token: write
       contents: write
@@ -208,12 +252,12 @@ jobs:
     steps:
       - uses: actions/checkout@de0fac2e4500dabe0009e67214ff5f5447ce83dd # ratchet:actions/checkout@v6.0.2
         with:
-          # NOTE: persist-credentials is needed for tauri-action to create GitHub releases.
+          # NOTE: persist-credentials is needed for tauri-action to upload assets to GitHub releases.
           persist-credentials: true # zizmor: ignore[artipacked]
 
       - name: Configure AWS credentials
         if: startsWith(matrix.platform, 'macos-')
-        uses: aws-actions/configure-aws-credentials@61815dcd50bd041e203e49132bacad1fd04d2708
+        uses: aws-actions/configure-aws-credentials@8df5847569e6427dd6c4fb1cf565c83acfa8afa7
         with:
           role-to-assume: ${{ secrets.AWS_OIDC_ROLE_ARN }}
           aws-region: us-east-2
@@ -353,11 +397,9 @@ jobs:
           APPLE_SIGNING_IDENTITY: ${{ env.CERT_ID }}
           APPLE_TEAM_ID: ${{ env.APPLE_TEAM_ID }}
         with:
-          tagName: ${{ needs.determine-builds.outputs.is-test-run != 'true' && 'v__VERSION__' || format('v0.0.0-dev+{0}', needs.determine-builds.outputs.short-sha) }}
-          releaseName: ${{ needs.determine-builds.outputs.is-test-run != 'true' && 'v__VERSION__' || format('v0.0.0-dev+{0}', needs.determine-builds.outputs.short-sha) }}
-          releaseBody: "See the assets to download this version and install."
-          releaseDraft: true
-          prerelease: false
+          # Use the release created by the create-release job to avoid race conditions
+          # when multiple matrix jobs try to create/update the same release simultaneously
+          releaseId: ${{ needs.create-release.outputs.release-id }}
           assetNamePattern: "[name]_[arch][ext]"
           args: ${{ matrix.args }}
 
@@ -384,7 +426,7 @@ jobs:
           persist-credentials: false
 
       - name: Configure AWS credentials
-        uses: aws-actions/configure-aws-credentials@61815dcd50bd041e203e49132bacad1fd04d2708
+        uses: aws-actions/configure-aws-credentials@8df5847569e6427dd6c4fb1cf565c83acfa8afa7
         with:
           role-to-assume: ${{ secrets.AWS_OIDC_ROLE_ARN }}
           aws-region: us-east-2
@@ -458,7 +500,7 @@ jobs:
           persist-credentials: false
 
       - name: Configure AWS credentials
-        uses: aws-actions/configure-aws-credentials@61815dcd50bd041e203e49132bacad1fd04d2708
+        uses: aws-actions/configure-aws-credentials@8df5847569e6427dd6c4fb1cf565c83acfa8afa7
         with:
           role-to-assume: ${{ secrets.AWS_OIDC_ROLE_ARN }}
           aws-region: us-east-2
@@ -527,7 +569,7 @@ jobs:
       - uses: runs-on/action@cd2b598b0515d39d78c38a02d529db87d2196d1e # ratchet:runs-on/action@v2
 
       - name: Configure AWS credentials
-        uses: aws-actions/configure-aws-credentials@61815dcd50bd041e203e49132bacad1fd04d2708
+        uses: aws-actions/configure-aws-credentials@8df5847569e6427dd6c4fb1cf565c83acfa8afa7
         with:
           role-to-assume: ${{ secrets.AWS_OIDC_ROLE_ARN }}
           aws-region: us-east-2
@@ -597,7 +639,7 @@ jobs:
           persist-credentials: false
 
       - name: Configure AWS credentials
-        uses: aws-actions/configure-aws-credentials@61815dcd50bd041e203e49132bacad1fd04d2708
+        uses: aws-actions/configure-aws-credentials@8df5847569e6427dd6c4fb1cf565c83acfa8afa7
         with:
           role-to-assume: ${{ secrets.AWS_OIDC_ROLE_ARN }}
           aws-region: us-east-2
@@ -679,7 +721,7 @@ jobs:
           persist-credentials: false
 
       - name: Configure AWS credentials
-        uses: aws-actions/configure-aws-credentials@61815dcd50bd041e203e49132bacad1fd04d2708
+        uses: aws-actions/configure-aws-credentials@8df5847569e6427dd6c4fb1cf565c83acfa8afa7
         with:
           role-to-assume: ${{ secrets.AWS_OIDC_ROLE_ARN }}
           aws-region: us-east-2
@@ -756,7 +798,7 @@ jobs:
       - uses: runs-on/action@cd2b598b0515d39d78c38a02d529db87d2196d1e # ratchet:runs-on/action@v2
 
       - name: Configure AWS credentials
-        uses: aws-actions/configure-aws-credentials@61815dcd50bd041e203e49132bacad1fd04d2708
+        uses: aws-actions/configure-aws-credentials@8df5847569e6427dd6c4fb1cf565c83acfa8afa7
         with:
           role-to-assume: ${{ secrets.AWS_OIDC_ROLE_ARN }}
           aws-region: us-east-2
@@ -823,7 +865,7 @@ jobs:
           persist-credentials: false
 
       - name: Configure AWS credentials
-        uses: aws-actions/configure-aws-credentials@61815dcd50bd041e203e49132bacad1fd04d2708
+        uses: aws-actions/configure-aws-credentials@8df5847569e6427dd6c4fb1cf565c83acfa8afa7
         with:
           role-to-assume: ${{ secrets.AWS_OIDC_ROLE_ARN }}
           aws-region: us-east-2
@@ -896,7 +938,7 @@ jobs:
           persist-credentials: false
 
       - name: Configure AWS credentials
-        uses: aws-actions/configure-aws-credentials@61815dcd50bd041e203e49132bacad1fd04d2708
+        uses: aws-actions/configure-aws-credentials@8df5847569e6427dd6c4fb1cf565c83acfa8afa7
         with:
           role-to-assume: ${{ secrets.AWS_OIDC_ROLE_ARN }}
           aws-region: us-east-2
@@ -964,7 +1006,7 @@ jobs:
       - uses: runs-on/action@cd2b598b0515d39d78c38a02d529db87d2196d1e # ratchet:runs-on/action@v2
 
       - name: Configure AWS credentials
-        uses: aws-actions/configure-aws-credentials@61815dcd50bd041e203e49132bacad1fd04d2708
+        uses: aws-actions/configure-aws-credentials@8df5847569e6427dd6c4fb1cf565c83acfa8afa7
         with:
           role-to-assume: ${{ secrets.AWS_OIDC_ROLE_ARN }}
           aws-region: us-east-2
@@ -1034,7 +1076,7 @@ jobs:
           persist-credentials: false
 
       - name: Configure AWS credentials
-        uses: aws-actions/configure-aws-credentials@61815dcd50bd041e203e49132bacad1fd04d2708
+        uses: aws-actions/configure-aws-credentials@8df5847569e6427dd6c4fb1cf565c83acfa8afa7
         with:
           role-to-assume: ${{ secrets.AWS_OIDC_ROLE_ARN }}
           aws-region: us-east-2
@@ -1107,7 +1149,7 @@ jobs:
           persist-credentials: false
 
       - name: Configure AWS credentials
-        uses: aws-actions/configure-aws-credentials@61815dcd50bd041e203e49132bacad1fd04d2708
+        uses: aws-actions/configure-aws-credentials@8df5847569e6427dd6c4fb1cf565c83acfa8afa7
         with:
           role-to-assume: ${{ secrets.AWS_OIDC_ROLE_ARN }}
           aws-region: us-east-2
@@ -1176,7 +1218,7 @@ jobs:
       - uses: runs-on/action@cd2b598b0515d39d78c38a02d529db87d2196d1e # ratchet:runs-on/action@v2
 
       - name: Configure AWS credentials
-        uses: aws-actions/configure-aws-credentials@61815dcd50bd041e203e49132bacad1fd04d2708
+        uses: aws-actions/configure-aws-credentials@8df5847569e6427dd6c4fb1cf565c83acfa8afa7
         with:
           role-to-assume: ${{ secrets.AWS_OIDC_ROLE_ARN }}
           aws-region: us-east-2
@@ -1246,7 +1288,7 @@ jobs:
           persist-credentials: false
 
       - name: Configure AWS credentials
-        uses: aws-actions/configure-aws-credentials@61815dcd50bd041e203e49132bacad1fd04d2708
+        uses: aws-actions/configure-aws-credentials@8df5847569e6427dd6c4fb1cf565c83acfa8afa7
         with:
           role-to-assume: ${{ secrets.AWS_OIDC_ROLE_ARN }}
           aws-region: us-east-2
@@ -1326,7 +1368,7 @@ jobs:
           persist-credentials: false
 
       - name: Configure AWS credentials
-        uses: aws-actions/configure-aws-credentials@61815dcd50bd041e203e49132bacad1fd04d2708
+        uses: aws-actions/configure-aws-credentials@8df5847569e6427dd6c4fb1cf565c83acfa8afa7
         with:
           role-to-assume: ${{ secrets.AWS_OIDC_ROLE_ARN }}
           aws-region: us-east-2
@@ -1400,7 +1442,7 @@ jobs:
       - uses: runs-on/action@cd2b598b0515d39d78c38a02d529db87d2196d1e # ratchet:runs-on/action@v2
 
       - name: Configure AWS credentials
-        uses: aws-actions/configure-aws-credentials@61815dcd50bd041e203e49132bacad1fd04d2708
+        uses: aws-actions/configure-aws-credentials@8df5847569e6427dd6c4fb1cf565c83acfa8afa7
         with:
           role-to-assume: ${{ secrets.AWS_OIDC_ROLE_ARN }}
           aws-region: us-east-2
@@ -1465,7 +1507,7 @@ jobs:
       - uses: runs-on/action@cd2b598b0515d39d78c38a02d529db87d2196d1e # ratchet:runs-on/action@v2
 
       - name: Configure AWS credentials
-        uses: aws-actions/configure-aws-credentials@61815dcd50bd041e203e49132bacad1fd04d2708
+        uses: aws-actions/configure-aws-credentials@8df5847569e6427dd6c4fb1cf565c83acfa8afa7
         with:
           role-to-assume: ${{ secrets.AWS_OIDC_ROLE_ARN }}
           aws-region: us-east-2
@@ -1520,7 +1562,7 @@ jobs:
       - uses: runs-on/action@cd2b598b0515d39d78c38a02d529db87d2196d1e # ratchet:runs-on/action@v2
 
       - name: Configure AWS credentials
-        uses: aws-actions/configure-aws-credentials@61815dcd50bd041e203e49132bacad1fd04d2708
+        uses: aws-actions/configure-aws-credentials@8df5847569e6427dd6c4fb1cf565c83acfa8afa7
         with:
           role-to-assume: ${{ secrets.AWS_OIDC_ROLE_ARN }}
           aws-region: us-east-2
@@ -1580,7 +1622,7 @@ jobs:
           persist-credentials: false
 
       - name: Configure AWS credentials
-        uses: aws-actions/configure-aws-credentials@61815dcd50bd041e203e49132bacad1fd04d2708
+        uses: aws-actions/configure-aws-credentials@8df5847569e6427dd6c4fb1cf565c83acfa8afa7
         with:
           role-to-assume: ${{ secrets.AWS_OIDC_ROLE_ARN }}
           aws-region: us-east-2
@@ -1637,7 +1679,7 @@ jobs:
       - uses: runs-on/action@cd2b598b0515d39d78c38a02d529db87d2196d1e # ratchet:runs-on/action@v2
 
       - name: Configure AWS credentials
-        uses: aws-actions/configure-aws-credentials@61815dcd50bd041e203e49132bacad1fd04d2708
+        uses: aws-actions/configure-aws-credentials@8df5847569e6427dd6c4fb1cf565c83acfa8afa7
         with:
           role-to-assume: ${{ secrets.AWS_OIDC_ROLE_ARN }}
           aws-region: us-east-2

.github/workflows/nightly-llm-provider-chat.yml

@@ -15,6 +15,8 @@ permissions:
 jobs:
   provider-chat-test:
     uses: ./.github/workflows/reusable-nightly-llm-provider-chat.yml
+    secrets:
+      AWS_OIDC_ROLE_ARN: ${{ secrets.AWS_OIDC_ROLE_ARN }}
     permissions:
       contents: read
       id-token: write

.github/workflows/post-merge-beta-cherry-pick.yml

@@ -6,11 +6,13 @@ on:
       - main
 
 permissions:
-  contents: write
-  pull-requests: write
+  contents: read
 
 jobs:
   cherry-pick-to-latest-release:
+    permissions:
+      contents: write
+      pull-requests: write
     outputs:
       should_cherrypick: ${{ steps.gate.outputs.should_cherrypick }}
       pr_number: ${{ steps.gate.outputs.pr_number }}

.github/workflows/pr-desktop-build.yml

@@ -57,7 +57,7 @@ jobs:
           cache-dependency-path: ./desktop/package-lock.json
 
       - name: Setup Rust
-        uses: dtolnay/rust-toolchain@4be9e76fd7c4901c61fb841f559994984270fce7
+        uses: dtolnay/rust-toolchain@efa25f7f19611383d5b0ccf2d1c8914531636bf9
         with:
           toolchain: stable
           targets: ${{ matrix.target }}

.github/workflows/pr-golang-tests.yml

@@ -0,0 +1,56 @@
+name: Golang Tests
+concurrency:
+  group: Golang-Tests-${{ github.workflow }}-${{ github.head_ref || github.event.workflow_run.head_branch || github.run_id }}
+  cancel-in-progress: true
+
+on:
+  merge_group:
+  pull_request:
+    branches:
+      - main
+      - "release/**"
+  push:
+    tags:
+      - "v*.*.*"
+
+permissions: {}
+
+env:
+  GO_VERSION: "1.26"
+
+jobs:
+  detect-modules:
+    runs-on: ubuntu-latest
+    timeout-minutes: 10
+    outputs:
+      modules: ${{ steps.set-modules.outputs.modules }}
+    steps:
+      - uses: actions/checkout@8e8c483db84b4bee98b60c0593521ed34d9990e8
+        with:
+          persist-credentials: false
+      - id: set-modules
+        run: echo "modules=$(find . -name 'go.mod' -exec dirname {} \; | jq -Rc '[.,inputs]')" >> "$GITHUB_OUTPUT"
+
+  golang:
+    needs: detect-modules
+    runs-on: ubuntu-latest
+    timeout-minutes: 10
+    strategy:
+      matrix:
+        modules: ${{ fromJSON(needs.detect-modules.outputs.modules) }}
+    steps:
+      - uses: actions/checkout@8e8c483db84b4bee98b60c0593521ed34d9990e8 # ratchet:actions/checkout@v6
+        with:
+          persist-credentials: false
+      - uses: actions/setup-go@4dc6199c7b1a012772edbd06daecab0f50c9053c # zizmor: ignore[cache-poisoning]
+        with:
+          go-version: ${{ env.GO_VERSION }}
+          cache-dependency-path: "**/go.sum"
+
+      - run: go mod tidy
+        working-directory: ${{ matrix.modules }}
+      - run: git diff --exit-code go.mod go.sum
+        working-directory: ${{ matrix.modules }}
+
+      - run: go test ./...
+        working-directory: ${{ matrix.modules }}

.github/workflows/pr-helm-chart-testing.yml

@@ -71,7 +71,7 @@ jobs:
 
       - name: Create kind cluster
         if: steps.list-changed.outputs.changed == 'true'
-        uses: helm/kind-action@92086f6be054225fa813e0a4b13787fc9088faab # ratchet:helm/kind-action@v1.13.0
+        uses: helm/kind-action@ef37e7f390d99f746eb8b610417061a60e82a6cc # ratchet:helm/kind-action@v1.14.0
 
       - name: Pre-install cluster status check
         if: steps.list-changed.outputs.changed == 'true'

.github/workflows/pr-integration-tests.yml

@@ -335,7 +335,6 @@ jobs:
           # TODO(Nik): https://linear.app/onyx-app/issue/ENG-1/update-test-infra-to-use-test-license
           LICENSE_ENFORCEMENT_ENABLED=false
           CHECK_TTL_MANAGEMENT_TASK_FREQUENCY_IN_HOURS=0.001
-          USE_LIGHTWEIGHT_BACKGROUND_WORKER=false
           EOF
           fi
 
@@ -471,13 +470,13 @@ jobs:
           path: ${{ github.workspace }}/docker-compose.log
       # ------------------------------------------------------------
 
-  no-vectordb-tests:
+  onyx-lite-tests:
     needs: [build-backend-image, build-integration-image]
     runs-on:
       [
         runs-on,
         runner=4cpu-linux-arm64,
-        "run-id=${{ github.run_id }}-no-vectordb-tests",
+        "run-id=${{ github.run_id }}-onyx-lite-tests",
         "extras=ecr-cache",
       ]
     timeout-minutes: 45
@@ -495,13 +494,12 @@ jobs:
           username: ${{ secrets.DOCKER_USERNAME }}
           password: ${{ secrets.DOCKER_TOKEN }}
 
-      - name: Create .env file for no-vectordb Docker Compose
+      - name: Create .env file for Onyx Lite Docker Compose
         env:
           ECR_CACHE: ${{ env.RUNS_ON_ECR_CACHE }}
           RUN_ID: ${{ github.run_id }}
         run: |
           cat <<EOF > deployment/docker_compose/.env
-          COMPOSE_PROFILES=s3-filestore
           ENABLE_PAID_ENTERPRISE_EDITION_FEATURES=true
           LICENSE_ENFORCEMENT_ENABLED=false
           AUTH_TYPE=basic
@@ -509,28 +507,23 @@ jobs:
           POSTGRES_USE_NULL_POOL=true
           REQUIRE_EMAIL_VERIFICATION=false
           DISABLE_TELEMETRY=true
-          DISABLE_VECTOR_DB=true
           ONYX_BACKEND_IMAGE=${ECR_CACHE}:integration-test-backend-test-${RUN_ID}
           INTEGRATION_TESTS_MODE=true
-          USE_LIGHTWEIGHT_BACKGROUND_WORKER=true
           EOF
 
-      # Start only the services needed for no-vectordb mode (no Vespa, no model servers)
-      - name: Start Docker containers (no-vectordb)
+      # Start only the services needed for Onyx Lite (Postgres + API server)
+      - name: Start Docker containers (onyx-lite)
         run: |
           cd deployment/docker_compose
-          docker compose -f docker-compose.yml -f docker-compose.no-vectordb.yml -f docker-compose.dev.yml up \
+          docker compose -f docker-compose.yml -f docker-compose.onyx-lite.yml -f docker-compose.dev.yml up \
             relational_db \
-            cache \
-            minio \
             api_server \
-            background \
             -d
-        id: start_docker_no_vectordb
+        id: start_docker_onyx_lite
 
       - name: Wait for services to be ready
         run: |
-          echo "Starting wait-for-service script (no-vectordb)..."
+          echo "Starting wait-for-service script (onyx-lite)..."
           start_time=$(date +%s)
           timeout=300
           while true; do
@@ -552,14 +545,14 @@ jobs:
             sleep 5
           done
 
-      - name: Run No-VectorDB Integration Tests
+      - name: Run Onyx Lite Integration Tests
         uses: nick-fields/retry@ce71cc2ab81d554ebbe88c79ab5975992d79ba08 # ratchet:nick-fields/retry@v3
         with:
           timeout_minutes: 20
           max_attempts: 3
           retry_wait_seconds: 10
           command: |
-            echo "Running no-vectordb integration tests..."
+            echo "Running onyx-lite integration tests..."
             docker run --rm --network onyx_default \
               --name test-runner \
               -e POSTGRES_HOST=relational_db \
@@ -570,39 +563,38 @@ jobs:
               -e DB_READONLY_PASSWORD=password \
               -e POSTGRES_POOL_PRE_PING=true \
               -e POSTGRES_USE_NULL_POOL=true \
-              -e REDIS_HOST=cache \
               -e API_SERVER_HOST=api_server \
               -e OPENAI_API_KEY=${OPENAI_API_KEY} \
               -e TEST_WEB_HOSTNAME=test-runner \
               ${{ env.RUNS_ON_ECR_CACHE }}:integration-test-${{ github.run_id }} \
               /app/tests/integration/tests/no_vectordb
 
-      - name: Dump API server logs (no-vectordb)
+      - name: Dump API server logs (onyx-lite)
         if: always()
         run: |
           cd deployment/docker_compose
-          docker compose -f docker-compose.yml -f docker-compose.no-vectordb.yml -f docker-compose.dev.yml \
-            logs --no-color api_server > $GITHUB_WORKSPACE/api_server_no_vectordb.log || true
+          docker compose -f docker-compose.yml -f docker-compose.onyx-lite.yml -f docker-compose.dev.yml \
+            logs --no-color api_server > $GITHUB_WORKSPACE/api_server_onyx_lite.log || true
 
-      - name: Dump all-container logs (no-vectordb)
+      - name: Dump all-container logs (onyx-lite)
         if: always()
         run: |
           cd deployment/docker_compose
-          docker compose -f docker-compose.yml -f docker-compose.no-vectordb.yml -f docker-compose.dev.yml \
-            logs --no-color > $GITHUB_WORKSPACE/docker-compose-no-vectordb.log || true
+          docker compose -f docker-compose.yml -f docker-compose.onyx-lite.yml -f docker-compose.dev.yml \
+            logs --no-color > $GITHUB_WORKSPACE/docker-compose-onyx-lite.log || true
 
-      - name: Upload logs (no-vectordb)
+      - name: Upload logs (onyx-lite)
         if: always()
         uses: actions/upload-artifact@b7c566a772e6b6bfb58ed0dc250532a479d7789f
         with:
-          name: docker-all-logs-no-vectordb
-          path: ${{ github.workspace }}/docker-compose-no-vectordb.log
+          name: docker-all-logs-onyx-lite
+          path: ${{ github.workspace }}/docker-compose-onyx-lite.log
 
-      - name: Stop Docker containers (no-vectordb)
+      - name: Stop Docker containers (onyx-lite)
         if: always()
         run: |
           cd deployment/docker_compose
-          docker compose -f docker-compose.yml -f docker-compose.no-vectordb.yml -f docker-compose.dev.yml down -v
+          docker compose -f docker-compose.yml -f docker-compose.onyx-lite.yml -f docker-compose.dev.yml down -v
 
   multitenant-tests:
     needs:
@@ -744,7 +736,7 @@ jobs:
     # NOTE: Github-hosted runners have about 20s faster queue times and are preferred here.
     runs-on: ubuntu-slim
     timeout-minutes: 45
-    needs: [integration-tests, no-vectordb-tests, multitenant-tests]
+    needs: [integration-tests, onyx-lite-tests, multitenant-tests]
     if: ${{ always() }}
     steps:
       - name: Check job status

.github/workflows/pr-jest-tests.yml

@@ -31,7 +31,7 @@ jobs:
         uses: actions/setup-node@6044e13b5dc448c55e2357c09f80417699197238 # ratchet:actions/setup-node@v4
         with:
           node-version: 22
-          cache: "npm"
+          cache: "npm" # zizmor: ignore[cache-poisoning] test-only workflow; no deploy artifacts
           cache-dependency-path: ./web/package-lock.json
 
       - name: Install node dependencies

.github/workflows/pr-playwright-tests.yml

@@ -268,10 +268,11 @@ jobs:
           persist-credentials: false
 
       - name: Setup node
+        # zizmor: ignore[cache-poisoning] ephemeral runners; no release artifacts
         uses: actions/setup-node@6044e13b5dc448c55e2357c09f80417699197238 # ratchet:actions/setup-node@v4
         with:
           node-version: 22
-          cache: "npm"
+          cache: "npm" # zizmor: ignore[cache-poisoning]
           cache-dependency-path: ./web/package-lock.json
 
       - name: Install node dependencies
@@ -279,6 +280,7 @@ jobs:
         run: npm ci
 
       - name: Cache playwright cache
+        # zizmor: ignore[cache-poisoning] ephemeral runners; no release artifacts
         uses: runs-on/cache@50350ad4242587b6c8c2baa2e740b1bc11285ff4 # ratchet:runs-on/cache@v4
         with:
           path: ~/.cache/ms-playwright
@@ -459,7 +461,7 @@ jobs:
       # --- Visual Regression Diff ---
       - name: Configure AWS credentials
         if: always()
-        uses: aws-actions/configure-aws-credentials@61815dcd50bd041e203e49132bacad1fd04d2708
+        uses: aws-actions/configure-aws-credentials@8df5847569e6427dd6c4fb1cf565c83acfa8afa7
         with:
           role-to-assume: ${{ secrets.AWS_OIDC_ROLE_ARN }}
           aws-region: us-east-2
@@ -590,6 +592,108 @@ jobs:
           name: docker-logs-${{ matrix.project }}-${{ github.run_id }}
           path: ${{ github.workspace }}/docker-compose.log
 
+  playwright-tests-lite:
+    needs: [build-web-image, build-backend-image]
+    name: Playwright Tests (lite)
+    runs-on:
+      - runs-on
+      - runner=4cpu-linux-arm64
+      - "run-id=${{ github.run_id }}-playwright-tests-lite"
+      - "extras=ecr-cache"
+    timeout-minutes: 30
+    steps:
+      - uses: runs-on/action@cd2b598b0515d39d78c38a02d529db87d2196d1e # ratchet:runs-on/action@v2
+
+      - name: Checkout code
+        uses: actions/checkout@de0fac2e4500dabe0009e67214ff5f5447ce83dd # ratchet:actions/checkout@v6
+        with:
+          persist-credentials: false
+
+      - name: Setup node
+        # zizmor: ignore[cache-poisoning] ephemeral runners; no release artifacts
+        uses: actions/setup-node@6044e13b5dc448c55e2357c09f80417699197238 # ratchet:actions/setup-node@v4
+        with:
+          node-version: 22
+          cache: "npm" # zizmor: ignore[cache-poisoning]
+          cache-dependency-path: ./web/package-lock.json
+
+      - name: Install node dependencies
+        working-directory: ./web
+        run: npm ci
+
+      - name: Cache playwright cache
+        # zizmor: ignore[cache-poisoning] ephemeral runners; no release artifacts
+        uses: runs-on/cache@50350ad4242587b6c8c2baa2e740b1bc11285ff4 # ratchet:runs-on/cache@v4
+        with:
+          path: ~/.cache/ms-playwright
+          key: ${{ runner.os }}-playwright-npm-${{ hashFiles('web/package-lock.json') }}
+          restore-keys: |
+            ${{ runner.os }}-playwright-npm-
+
+      - name: Install playwright browsers
+        working-directory: ./web
+        run: npx playwright install --with-deps
+
+      - name: Create .env file for Docker Compose
+        env:
+          OPENAI_API_KEY_VALUE: ${{ env.OPENAI_API_KEY }}
+          ECR_CACHE: ${{ env.RUNS_ON_ECR_CACHE }}
+          RUN_ID: ${{ github.run_id }}
+        run: |
+          cat <<EOF > deployment/docker_compose/.env
+          ENABLE_PAID_ENTERPRISE_EDITION_FEATURES=true
+          LICENSE_ENFORCEMENT_ENABLED=false
+          AUTH_TYPE=basic
+          INTEGRATION_TESTS_MODE=true
+          GEN_AI_API_KEY=${OPENAI_API_KEY_VALUE}
+          MOCK_LLM_RESPONSE=true
+          REQUIRE_EMAIL_VERIFICATION=false
+          DISABLE_TELEMETRY=true
+          ONYX_BACKEND_IMAGE=${ECR_CACHE}:playwright-test-backend-${RUN_ID}
+          ONYX_WEB_SERVER_IMAGE=${ECR_CACHE}:playwright-test-web-${RUN_ID}
+          EOF
+
+      # needed for pulling external images otherwise, we hit the "Unauthenticated users" limit
+      # https://docs.docker.com/docker-hub/usage/
+      - name: Login to Docker Hub
+        uses: docker/login-action@c94ce9fb468520275223c153574b00df6fe4bcc9 # ratchet:docker/login-action@v3
+        with:
+          username: ${{ secrets.DOCKER_USERNAME }}
+          password: ${{ secrets.DOCKER_TOKEN }}
+
+      - name: Start Docker containers (lite)
+        run: |
+          cd deployment/docker_compose
+          docker compose -f docker-compose.yml -f docker-compose.onyx-lite.yml -f docker-compose.dev.yml up -d
+        id: start_docker
+
+      - name: Run Playwright tests (lite)
+        working-directory: ./web
+        run: npx playwright test --project lite
+
+      - uses: actions/upload-artifact@b7c566a772e6b6bfb58ed0dc250532a479d7789f
+        if: always()
+        with:
+          name: playwright-test-results-lite-${{ github.run_id }}
+          path: ./web/output/playwright/
+          retention-days: 30
+
+      - name: Save Docker logs
+        if: success() || failure()
+        env:
+          WORKSPACE: ${{ github.workspace }}
+        run: |
+          cd deployment/docker_compose
+          docker compose logs > docker-compose.log
+          mv docker-compose.log ${WORKSPACE}/docker-compose.log
+
+      - name: Upload logs
+        if: success() || failure()
+        uses: actions/upload-artifact@b7c566a772e6b6bfb58ed0dc250532a479d7789f
+        with:
+          name: docker-logs-lite-${{ github.run_id }}
+          path: ${{ github.workspace }}/docker-compose.log
+
   # Post a single combined visual regression comment after all matrix jobs finish
   visual-regression-comment:
     needs: [playwright-tests]
@@ -686,7 +790,7 @@ jobs:
     # NOTE: Github-hosted runners have about 20s faster queue times and are preferred here.
     runs-on: ubuntu-slim
     timeout-minutes: 45
-    needs: [playwright-tests]
+    needs: [playwright-tests, playwright-tests-lite]
     if: ${{ always() }}
     steps:
       - name: Check job status

.github/workflows/pr-quality-checks.yml

@@ -38,9 +38,9 @@ jobs:
       - name: Install node dependencies
         working-directory: ./web
         run: npm ci
-      - uses: j178/prek-action@9d6a3097e0c1865ecce00cfb89fe80f2ee91b547 # ratchet:j178/prek-action@v1
+      - uses: j178/prek-action@0bb87d7f00b0c99306c8bcb8b8beba1eb581c037 # ratchet:j178/prek-action@v1
         with:
-          prek-version: '0.2.21'
+          prek-version: '0.3.4'
           extra-args: ${{ github.event_name == 'pull_request' && format('--from-ref {0} --to-ref {1}', github.event.pull_request.base.sha, github.event.pull_request.head.sha) || github.event_name == 'merge_group' && format('--from-ref {0} --to-ref {1}', github.event.merge_group.base_sha, github.event.merge_group.head_sha) || github.ref_name == 'main' && '--all-files' || '' }}
       - name: Check Actions
         uses: giner/check-actions@28d366c7cbbe235f9624a88aa31a628167eee28c # ratchet:giner/check-actions@v1.0.1

.github/workflows/release-cli.yml

@@ -0,0 +1,39 @@
+name: Release CLI
+
+on:
+  push:
+    tags:
+      - "cli/v*.*.*"
+
+jobs:
+  pypi:
+    runs-on: ubuntu-latest
+    environment:
+      name: release-cli
+    permissions:
+      id-token: write
+    timeout-minutes: 10
+    strategy:
+      matrix:
+        os-arch:
+          - { goos: "linux", goarch: "amd64" }
+          - { goos: "linux", goarch: "arm64" }
+          - { goos: "windows", goarch: "amd64" }
+          - { goos: "windows", goarch: "arm64" }
+          - { goos: "darwin", goarch: "amd64" }
+          - { goos: "darwin", goarch: "arm64" }
+    steps:
+      - uses: actions/checkout@de0fac2e4500dabe0009e67214ff5f5447ce83dd # ratchet:actions/checkout@v6
+        with:
+          persist-credentials: false
+      - uses: astral-sh/setup-uv@61cb8a9741eeb8a550a1b8544337180c0fc8476b # ratchet:astral-sh/setup-uv@v7
+        with:
+          enable-cache: false
+          version: "0.9.9"
+      - run: |
+          GOOS="${{ matrix.os-arch.goos }}" \
+          GOARCH="${{ matrix.os-arch.goarch }}" \
+          uv build --wheel
+        working-directory: cli
+      - run: uv publish
+        working-directory: cli

.github/workflows/release-devtools.yml

@@ -22,12 +22,10 @@ jobs:
           - { goos: "windows", goarch: "arm64" }
           - { goos: "darwin", goarch: "amd64" }
           - { goos: "darwin", goarch: "arm64" }
-          - { goos: "", goarch: "" }
     steps:
       - uses: actions/checkout@de0fac2e4500dabe0009e67214ff5f5447ce83dd # ratchet:actions/checkout@v6
         with:
           persist-credentials: false
-          fetch-depth: 0
       - uses: astral-sh/setup-uv@61cb8a9741eeb8a550a1b8544337180c0fc8476b # ratchet:astral-sh/setup-uv@v7
         with:
           enable-cache: false

.github/workflows/reusable-nightly-llm-provider-chat.yml

@@ -48,6 +48,10 @@ on:
         required: false
         default: true
         type: boolean
+    secrets:
+      AWS_OIDC_ROLE_ARN:
+        description: "AWS role ARN for OIDC auth"
+        required: true
 
 permissions:
   contents: read
@@ -73,7 +77,7 @@ jobs:
           persist-credentials: false
 
       - name: Configure AWS credentials
-        uses: aws-actions/configure-aws-credentials@61815dcd50bd041e203e49132bacad1fd04d2708
+        uses: aws-actions/configure-aws-credentials@8df5847569e6427dd6c4fb1cf565c83acfa8afa7
         with:
           role-to-assume: ${{ secrets.AWS_OIDC_ROLE_ARN }}
           aws-region: us-east-2
@@ -116,7 +120,7 @@ jobs:
           persist-credentials: false
 
       - name: Configure AWS credentials
-        uses: aws-actions/configure-aws-credentials@61815dcd50bd041e203e49132bacad1fd04d2708
+        uses: aws-actions/configure-aws-credentials@8df5847569e6427dd6c4fb1cf565c83acfa8afa7
         with:
           role-to-assume: ${{ secrets.AWS_OIDC_ROLE_ARN }}
           aws-region: us-east-2
@@ -158,7 +162,7 @@ jobs:
           persist-credentials: false
 
       - name: Configure AWS credentials
-        uses: aws-actions/configure-aws-credentials@61815dcd50bd041e203e49132bacad1fd04d2708
+        uses: aws-actions/configure-aws-credentials@8df5847569e6427dd6c4fb1cf565c83acfa8afa7
         with:
           role-to-assume: ${{ secrets.AWS_OIDC_ROLE_ARN }}
           aws-region: us-east-2
@@ -264,7 +268,7 @@ jobs:
           persist-credentials: false
 
       - name: Configure AWS credentials
-        uses: aws-actions/configure-aws-credentials@61815dcd50bd041e203e49132bacad1fd04d2708
+        uses: aws-actions/configure-aws-credentials@8df5847569e6427dd6c4fb1cf565c83acfa8afa7
         with:
           role-to-assume: ${{ secrets.AWS_OIDC_ROLE_ARN }}
           aws-region: us-east-2

.github/workflows/sandbox-deployment.yml

@@ -110,7 +110,7 @@ jobs:
           persist-credentials: false
 
       - name: Configure AWS credentials
-        uses: aws-actions/configure-aws-credentials@61815dcd50bd041e203e49132bacad1fd04d2708
+        uses: aws-actions/configure-aws-credentials@8df5847569e6427dd6c4fb1cf565c83acfa8afa7
         with:
           role-to-assume: ${{ secrets.AWS_OIDC_ROLE_ARN }}
           aws-region: us-east-2
@@ -180,7 +180,7 @@ jobs:
           persist-credentials: false
 
       - name: Configure AWS credentials
-        uses: aws-actions/configure-aws-credentials@61815dcd50bd041e203e49132bacad1fd04d2708
+        uses: aws-actions/configure-aws-credentials@8df5847569e6427dd6c4fb1cf565c83acfa8afa7
         with:
           role-to-assume: ${{ secrets.AWS_OIDC_ROLE_ARN }}
           aws-region: us-east-2
@@ -244,7 +244,7 @@ jobs:
       - uses: runs-on/action@cd2b598b0515d39d78c38a02d529db87d2196d1e # ratchet:runs-on/action@v2
 
       - name: Configure AWS credentials
-        uses: aws-actions/configure-aws-credentials@61815dcd50bd041e203e49132bacad1fd04d2708
+        uses: aws-actions/configure-aws-credentials@8df5847569e6427dd6c4fb1cf565c83acfa8afa7
         with:
           role-to-assume: ${{ secrets.AWS_OIDC_ROLE_ARN }}
           aws-region: us-east-2

.pre-commit-config.yaml

@@ -119,10 +119,11 @@ repos:
           ]
 
   - repo: https://github.com/golangci/golangci-lint
-    rev: 9f61b0f53f80672872fced07b6874397c3ed197b # frozen: v2.7.2
+    rev: 5d1e709b7be35cb2025444e19de266b056b7b7ee # frozen: v2.10.1
     hooks:
       - id: golangci-lint
-        entry: bash -c "find tools/ -name go.mod -print0 | xargs -0 -I{} bash -c 'cd \"$(dirname {})\" && golangci-lint run ./...'"
+        language_version: "1.26.0"
+        entry: bash -c "find . -name go.mod -not -path './.venv/*' -print0 | xargs -0 -I{} bash -c 'cd \"$(dirname {})\" && golangci-lint run ./...'"
 
   - repo: https://github.com/astral-sh/ruff-pre-commit
     # Ruff version.

.vscode/launch.json

@@ -40,19 +40,7 @@
       }
     },
     {
-      "name": "Celery (lightweight mode)",
-      "configurations": [
-        "Celery primary",
-        "Celery background",
-        "Celery beat"
-      ],
-      "presentation": {
-        "group": "1"
-      },
-      "stopAll": true
-    },
-    {
-      "name": "Celery (standard mode)",
+      "name": "Celery",
       "configurations": [
         "Celery primary",
         "Celery light",
@@ -253,35 +241,6 @@
       },
       "consoleTitle": "Celery light Console"
     },
-    {
-      "name": "Celery background",
-      "type": "debugpy",
-      "request": "launch",
-      "module": "celery",
-      "cwd": "${workspaceFolder}/backend",
-      "envFile": "${workspaceFolder}/.vscode/.env",
-      "env": {
-        "LOG_LEVEL": "INFO",
-        "PYTHONUNBUFFERED": "1",
-        "PYTHONPATH": "."
-      },
-      "args": [
-        "-A",
-        "onyx.background.celery.versioned_apps.background",
-        "worker",
-        "--pool=threads",
-        "--concurrency=20",
-        "--prefetch-multiplier=4",
-        "--loglevel=INFO",
-        "--hostname=background@%n",
-        "-Q",
-        "vespa_metadata_sync,connector_deletion,doc_permissions_upsert,checkpoint_cleanup,index_attempt_cleanup,docprocessing,connector_doc_fetching,connector_pruning,connector_doc_permissions_sync,connector_external_group_sync,csv_generation,kg_processing,monitoring,user_file_processing,user_file_project_sync,user_file_delete,opensearch_migration"
-      ],
-      "presentation": {
-        "group": "2"
-      },
-      "consoleTitle": "Celery background Console"
-    },
     {
       "name": "Celery heavy",
       "type": "debugpy",
@@ -526,21 +485,6 @@
         "group": "3"
       }
     },
-    {
-      "name": "Clear and Restart OpenSearch Container",
-      // Generic debugger type, required arg but has no bearing on bash.
-      "type": "node",
-      "request": "launch",
-      "runtimeExecutable": "bash",
-      "runtimeArgs": [
-        "${workspaceFolder}/backend/scripts/restart_opensearch_container.sh"
-      ],
-      "cwd": "${workspaceFolder}",
-      "console": "integratedTerminal",
-      "presentation": {
-        "group": "3"
-      }
-    },
     {
       "name": "Eval CLI",
       "type": "debugpy",

AGENTS.md

@@ -86,37 +86,6 @@ Onyx uses Celery for asynchronous task processing with multiple specialized work
      - Monitoring tasks (every 5 minutes)
      - Cleanup tasks (hourly)
 
-#### Worker Deployment Modes
-
-Onyx supports two deployment modes for background workers, controlled by the `USE_LIGHTWEIGHT_BACKGROUND_WORKER` environment variable:
-
-**Lightweight Mode** (default, `USE_LIGHTWEIGHT_BACKGROUND_WORKER=true`):
-
-- Runs a single consolidated `background` worker that handles all background tasks:
-  - Light worker tasks (Vespa operations, permissions sync, deletion)
-  - Document processing (indexing pipeline)
-  - Document fetching (connector data retrieval)
-  - Pruning operations (from `heavy` worker)
-  - Knowledge graph processing (from `kg_processing` worker)
-  - Monitoring tasks (from `monitoring` worker)
-  - User file processing (from `user_file_processing` worker)
-- Lower resource footprint (fewer worker processes)
-- Suitable for smaller deployments or development environments
-- Default concurrency: 20 threads (increased to handle combined workload)
-
-**Standard Mode** (`USE_LIGHTWEIGHT_BACKGROUND_WORKER=false`):
-
-- Runs separate specialized workers as documented above (light, docprocessing, docfetching, heavy, kg_processing, monitoring, user_file_processing)
-- Better isolation and scalability
-- Can scale individual workers independently based on workload
-- Suitable for production deployments with higher load
-
-The deployment mode affects:
-
-- **Backend**: Worker processes spawned by supervisord or dev scripts
-- **Helm**: Which Kubernetes deployments are created
-- **Dev Environment**: Which workers `dev_run_background_jobs.py` spawns
-
 #### Key Features
 
 - **Thread-based Workers**: All workers use thread pools (not processes) for stability
@@ -135,6 +104,10 @@ The deployment mode affects:
 
 - Always use `@shared_task` rather than `@celery_app`
 - Put tasks under `background/celery/tasks/` or `ee/background/celery/tasks`
+- Never enqueue a task without an expiration. Always supply `expires=` when
+  sending tasks, either from the beat schedule or directly from another task. It
+  should never be acceptable to submit code which enqueues tasks without an
+  expiration, as doing so can lead to unbounded task queue growth.
 
 **Defining APIs**:
 When creating new FastAPI APIs, do NOT use the `response_model` field. Instead, just type the
@@ -617,6 +590,45 @@ Keep it high level. You can reference certain files or functions though.
 
 Before writing your plan, make sure to do research. Explore the relevant sections in the codebase.
 
+## Error Handling
+
+**Always raise `OnyxError` from `onyx.error_handling.exceptions` instead of `HTTPException`.
+Never hardcode status codes or use `starlette.status` / `fastapi.status` constants directly.**
+
+A global FastAPI exception handler converts `OnyxError` into a JSON response with the standard
+`{"error_code": "...", "message": "..."}` shape. This eliminates boilerplate and keeps error
+handling consistent across the entire backend.
+
+```python
+from onyx.error_handling.error_codes import OnyxErrorCode
+from onyx.error_handling.exceptions import OnyxError
+
+# ✅ Good
+raise OnyxError(OnyxErrorCode.NOT_FOUND, "Session not found")
+
+# ✅ Good — no extra message needed
+raise OnyxError(OnyxErrorCode.UNAUTHENTICATED)
+
+# ✅ Good — upstream service with dynamic status code
+raise OnyxError(OnyxErrorCode.BAD_GATEWAY, detail, status_code_override=upstream_status)
+
+# ❌ Bad — using HTTPException directly
+raise HTTPException(status_code=404, detail="Session not found")
+
+# ❌ Bad — starlette constant
+raise HTTPException(status_code=status.HTTP_403_FORBIDDEN, detail="Access denied")
+```
+
+Available error codes are defined in `backend/onyx/error_handling/error_codes.py`. If a new error
+category is needed, add it there first — do not invent ad-hoc codes.
+
+**Upstream service errors:** When forwarding errors from an upstream service where the HTTP
+status code is dynamic (comes from the upstream response), use `status_code_override`:
+
+```python
+raise OnyxError(OnyxErrorCode.BAD_GATEWAY, detail, status_code_override=e.response.status_code)
+```
+
 ## Best Practices
 
 In addition to the other content in this file, best practices for contributing

backend/alembic/versions/2664261bfaab_add_cache_store_table.py

@@ -0,0 +1,37 @@
+"""add cache_store table
+
+Revision ID: 2664261bfaab
+Revises: 4a1e4b1c89d2
+Create Date: 2026-02-27 00:00:00.000000
+
+"""
+
+from alembic import op
+import sqlalchemy as sa
+
+# revision identifiers, used by Alembic.
+revision = "2664261bfaab"
+down_revision = "4a1e4b1c89d2"
+branch_labels: None = None
+depends_on: None = None
+
+
+def upgrade() -> None:
+    op.create_table(
+        "cache_store",
+        sa.Column("key", sa.String(), nullable=False),
+        sa.Column("value", sa.LargeBinary(), nullable=True),
+        sa.Column("expires_at", sa.DateTime(timezone=True), nullable=True),
+        sa.PrimaryKeyConstraint("key"),
+    )
+    op.create_index(
+        "ix_cache_store_expires",
+        "cache_store",
+        ["expires_at"],
+        postgresql_where=sa.text("expires_at IS NOT NULL"),
+    )
+
+
+def downgrade() -> None:
+    op.drop_index("ix_cache_store_expires", table_name="cache_store")
+    op.drop_table("cache_store")

backend/alembic/versions/a3b8d9e2f1c4_make_scim_external_id_nullable.py

@@ -0,0 +1,34 @@
+"""make scim_user_mapping.external_id nullable
+
+Revision ID: a3b8d9e2f1c4
+Revises: 2664261bfaab
+Create Date: 2026-03-02
+
+"""
+
+from alembic import op
+
+
+# revision identifiers, used by Alembic.
+revision = "a3b8d9e2f1c4"
+down_revision = "2664261bfaab"
+branch_labels = None
+depends_on = None
+
+
+def upgrade() -> None:
+    op.alter_column(
+        "scim_user_mapping",
+        "external_id",
+        nullable=True,
+    )
+
+
+def downgrade() -> None:
+    # Delete any rows where external_id is NULL before re-applying NOT NULL
+    op.execute("DELETE FROM scim_user_mapping WHERE external_id IS NULL")
+    op.alter_column(
+        "scim_user_mapping",
+        "external_id",
+        nullable=False,
+    )

backend/ee/onyx/background/celery/apps/background.py

@@ -1,15 +0,0 @@
-from onyx.background.celery.apps import app_base
-from onyx.background.celery.apps.background import celery_app
-
-
-celery_app.autodiscover_tasks(
-    app_base.filter_task_modules(
-        [
-            "ee.onyx.background.celery.tasks.doc_permission_syncing",
-            "ee.onyx.background.celery.tasks.external_group_syncing",
-            "ee.onyx.background.celery.tasks.cleanup",
-            "ee.onyx.background.celery.tasks.tenant_provisioning",
-            "ee.onyx.background.celery.tasks.query_history",
-        ]
-    )
-)

backend/ee/onyx/db/license.py

@@ -11,11 +11,10 @@ from ee.onyx.server.license.models import LicenseMetadata
 from ee.onyx.server.license.models import LicensePayload
 from ee.onyx.server.license.models import LicenseSource
 from onyx.auth.schemas import UserRole
+from onyx.cache.factory import get_cache_backend
 from onyx.configs.constants import ANONYMOUS_USER_EMAIL
 from onyx.db.models import License
 from onyx.db.models import User
-from onyx.redis.redis_pool import get_redis_client
-from onyx.redis.redis_pool import get_redis_replica_client
 from onyx.utils.logger import setup_logger
 from shared_configs.configs import MULTI_TENANT
 from shared_configs.contextvars import get_current_tenant_id
@@ -142,7 +141,7 @@ def get_used_seats(tenant_id: str | None = None) -> int:
 
 def get_cached_license_metadata(tenant_id: str | None = None) -> LicenseMetadata | None:
     """
-    Get license metadata from Redis cache.
+    Get license metadata from cache.
 
     Args:
         tenant_id: Tenant ID (for multi-tenant deployments)
@@ -150,38 +149,34 @@ def get_cached_license_metadata(tenant_id: str | None = None) -> LicenseMetadata
     Returns:
         LicenseMetadata if cached, None otherwise
     """
-    tenant = tenant_id or get_current_tenant_id()
-    redis_client = get_redis_replica_client(tenant_id=tenant)
+    cache = get_cache_backend(tenant_id=tenant_id)
+    cached = cache.get(LICENSE_METADATA_KEY)
+    if not cached:
+        return None
 
-    cached = redis_client.get(LICENSE_METADATA_KEY)
-    if cached:
-        try:
-            cached_str: str
-            if isinstance(cached, bytes):
-                cached_str = cached.decode("utf-8")
-            else:
-                cached_str = str(cached)
-            return LicenseMetadata.model_validate_json(cached_str)
-        except Exception as e:
-            logger.warning(f"Failed to parse cached license metadata: {e}")
-            return None
-    return None
+    try:
+        cached_str = (
+            cached.decode("utf-8") if isinstance(cached, bytes) else str(cached)
+        )
+        return LicenseMetadata.model_validate_json(cached_str)
+    except Exception as e:
+        logger.warning(f"Failed to parse cached license metadata: {e}")
+        return None
 
 
 def invalidate_license_cache(tenant_id: str | None = None) -> None:
     """
     Invalidate the license metadata cache (not the license itself).
 
-    This deletes the cached LicenseMetadata from Redis. The actual license
-    in the database is not affected. Redis delete is idempotent - if the
-    key doesn't exist, this is a no-op.
+    Deletes the cached LicenseMetadata. The actual license in the database
+    is not affected. Delete is idempotent — if the key doesn't exist, this
+    is a no-op.
 
     Args:
         tenant_id: Tenant ID (for multi-tenant deployments)
     """
-    tenant = tenant_id or get_current_tenant_id()
-    redis_client = get_redis_client(tenant_id=tenant)
-    redis_client.delete(LICENSE_METADATA_KEY)
+    cache = get_cache_backend(tenant_id=tenant_id)
+    cache.delete(LICENSE_METADATA_KEY)
     logger.info("License cache invalidated")
 
 
@@ -192,7 +187,7 @@ def update_license_cache(
     tenant_id: str | None = None,
 ) -> LicenseMetadata:
     """
-    Update the Redis cache with license metadata.
+    Update the cache with license metadata.
 
     We cache all license statuses (ACTIVE, GRACE_PERIOD, GATED_ACCESS) because:
     1. Frontend needs status to show appropriate UI/banners
@@ -211,7 +206,7 @@ def update_license_cache(
     from ee.onyx.utils.license import get_license_status
 
     tenant = tenant_id or get_current_tenant_id()
-    redis_client = get_redis_client(tenant_id=tenant)
+    cache = get_cache_backend(tenant_id=tenant_id)
 
     used_seats = get_used_seats(tenant)
     status = get_license_status(payload, grace_period_end)
@@ -230,7 +225,7 @@ def update_license_cache(
         stripe_subscription_id=payload.stripe_subscription_id,
     )
 
-    redis_client.set(
+    cache.set(
         LICENSE_METADATA_KEY,
         metadata.model_dump_json(),
         ex=LICENSE_CACHE_TTL_SECONDS,

backend/ee/onyx/db/scim.py

@@ -126,12 +126,16 @@ class ScimDAL(DAL):
 
     def create_user_mapping(
         self,
-        external_id: str,
+        external_id: str | None,
         user_id: UUID,
         scim_username: str | None = None,
         fields: ScimMappingFields | None = None,
     ) -> ScimUserMapping:
-        """Create a mapping between a SCIM externalId and an Onyx user."""
+        """Create a SCIM mapping for a user.
+
+        ``external_id`` may be ``None`` when the IdP omits it (RFC 7643
+        allows this). The mapping still marks the user as SCIM-managed.
+        """
         f = fields or ScimMappingFields()
         mapping = ScimUserMapping(
             external_id=external_id,
@@ -270,8 +274,13 @@ class ScimDAL(DAL):
         Raises:
             ValueError: If the filter uses an unsupported attribute.
         """
-        query = select(User).where(
-            User.role.notin_([UserRole.SLACK_USER, UserRole.EXT_PERM_USER])
+        # Inner-join with ScimUserMapping so only SCIM-managed users appear.
+        # Pre-existing system accounts (anonymous, admin, etc.) are excluded
+        # unless they were explicitly linked via SCIM provisioning.
+        query = (
+            select(User)
+            .join(ScimUserMapping, ScimUserMapping.user_id == User.id)
+            .where(User.role.notin_([UserRole.SLACK_USER, UserRole.EXT_PERM_USER]))
         )
 
         if scim_filter:
@@ -321,34 +330,37 @@ class ScimDAL(DAL):
         scim_username: str | None = None,
         fields: ScimMappingFields | None = None,
     ) -> None:
-        """Create, update, or delete the external ID mapping for a user.
+        """Sync the SCIM mapping for a user.
+
+        If a mapping already exists, its fields are updated (including
+        setting ``external_id`` to ``None`` when the IdP omits it).
+        If no mapping exists and ``new_external_id`` is provided, a new
+        mapping is created.  A mapping is never deleted here — SCIM-managed
+        users must retain their mapping to remain visible in ``GET /Users``.
 
         When *fields* is provided, all mapping fields are written
         unconditionally — including ``None`` values — so that a caller can
         clear a previously-set field (e.g. removing a department).
         """
         mapping = self.get_user_mapping_by_user_id(user_id)
-        if new_external_id:
-            if mapping:
-                if mapping.external_id != new_external_id:
-                    mapping.external_id = new_external_id
-                if scim_username is not None:
-                    mapping.scim_username = scim_username
-                if fields is not None:
-                    mapping.department = fields.department
-                    mapping.manager = fields.manager
-                    mapping.given_name = fields.given_name
-                    mapping.family_name = fields.family_name
-                    mapping.scim_emails_json = fields.scim_emails_json
-            else:
-                self.create_user_mapping(
-                    external_id=new_external_id,
-                    user_id=user_id,
-                    scim_username=scim_username,
-                    fields=fields,
-                )
-        elif mapping:
-            self.delete_user_mapping(mapping.id)
+        if mapping:
+            if mapping.external_id != new_external_id:
+                mapping.external_id = new_external_id
+            if scim_username is not None:
+                mapping.scim_username = scim_username
+            if fields is not None:
+                mapping.department = fields.department
+                mapping.manager = fields.manager
+                mapping.given_name = fields.given_name
+                mapping.family_name = fields.family_name
+                mapping.scim_emails_json = fields.scim_emails_json
+        elif new_external_id:
+            self.create_user_mapping(
+                external_id=new_external_id,
+                user_id=user_id,
+                scim_username=scim_username,
+                fields=fields,
+            )
 
     def _get_user_mappings_batch(
         self, user_ids: list[UUID]

backend/ee/onyx/db/user_group.py

@@ -15,6 +15,7 @@ from sqlalchemy.orm import Session
 from ee.onyx.server.user_group.models import SetCuratorRequest
 from ee.onyx.server.user_group.models import UserGroupCreate
 from ee.onyx.server.user_group.models import UserGroupUpdate
+from onyx.configs.app_configs import DISABLE_VECTOR_DB
 from onyx.db.connector_credential_pair import get_connector_credential_pair_from_id
 from onyx.db.enums import AccessType
 from onyx.db.enums import ConnectorCredentialPairStatus
@@ -471,7 +472,9 @@ def _add_user_group__cc_pair_relationships__no_commit(
 
 def insert_user_group(db_session: Session, user_group: UserGroupCreate) -> UserGroup:
     db_user_group = UserGroup(
-        name=user_group.name, time_last_modified_by_user=func.now()
+        name=user_group.name,
+        time_last_modified_by_user=func.now(),
+        is_up_to_date=DISABLE_VECTOR_DB,
     )
     db_session.add(db_user_group)
     db_session.flush()  # give the group an ID
@@ -774,8 +777,7 @@ def update_user_group(
             cc_pair_ids=user_group_update.cc_pair_ids,
         )
 
-    # only needs to sync with Vespa if the cc_pairs have been updated
-    if cc_pairs_updated:
+    if cc_pairs_updated and not DISABLE_VECTOR_DB:
         db_user_group.is_up_to_date = False
 
     removed_users = db_session.scalars(

backend/ee/onyx/main.py

@@ -4,7 +4,6 @@ from contextlib import asynccontextmanager
 from fastapi import FastAPI
 from httpx_oauth.clients.google import GoogleOAuth2
 
-from ee.onyx.configs.app_configs import LICENSE_ENFORCEMENT_ENABLED
 from ee.onyx.server.analytics.api import router as analytics_router
 from ee.onyx.server.auth_check import check_ee_router_auth
 from ee.onyx.server.billing.api import router as billing_router
@@ -153,12 +152,9 @@ def get_application() -> FastAPI:
     # License management
     include_router_with_global_prefix_prepended(application, license_router)
 
-    # Unified billing API - available when license system is enabled
-    # Works for both self-hosted and cloud deployments
-    # TODO(ENG-3533): Once frontend migrates to /admin/billing/*, this becomes the
-    # primary billing API and /tenants/* billing endpoints can be removed
-    if LICENSE_ENFORCEMENT_ENABLED:
-        include_router_with_global_prefix_prepended(application, billing_router)
+    # Unified billing API - always registered in EE.
+    # Each endpoint is protected by the `current_admin_user` dependency (admin auth).
+    include_router_with_global_prefix_prepended(application, billing_router)
 
     if MULTI_TENANT:
         # Tenant management

backend/ee/onyx/server/billing/api.py

@@ -26,7 +26,6 @@ import asyncio
 import httpx
 from fastapi import APIRouter
 from fastapi import Depends
-from fastapi import HTTPException
 from pydantic import BaseModel
 from sqlalchemy.orm import Session
 
@@ -42,7 +41,6 @@ from ee.onyx.server.billing.models import SeatUpdateRequest
 from ee.onyx.server.billing.models import SeatUpdateResponse
 from ee.onyx.server.billing.models import StripePublishableKeyResponse
 from ee.onyx.server.billing.models import SubscriptionStatusResponse
-from ee.onyx.server.billing.service import BillingServiceError
 from ee.onyx.server.billing.service import (
     create_checkout_session as create_checkout_service,
 )
@@ -58,6 +56,8 @@ from onyx.configs.app_configs import STRIPE_PUBLISHABLE_KEY_OVERRIDE
 from onyx.configs.app_configs import STRIPE_PUBLISHABLE_KEY_URL
 from onyx.configs.app_configs import WEB_DOMAIN
 from onyx.db.engine.sql_engine import get_session
+from onyx.error_handling.error_codes import OnyxErrorCode
+from onyx.error_handling.exceptions import OnyxError
 from onyx.redis.redis_pool import get_shared_redis_client
 from onyx.utils.logger import setup_logger
 from shared_configs.configs import MULTI_TENANT
@@ -169,26 +169,23 @@ async def create_checkout_session(
     if seats is not None:
         used_seats = get_used_seats(tenant_id)
         if seats < used_seats:
-            raise HTTPException(
-                status_code=400,
-                detail=f"Cannot subscribe with fewer seats than current usage. "
+            raise OnyxError(
+                OnyxErrorCode.VALIDATION_ERROR,
+                f"Cannot subscribe with fewer seats than current usage. "
                 f"You have {used_seats} active users/integrations but requested {seats} seats.",
             )
 
     # Build redirect URL for after checkout completion
     redirect_url = f"{WEB_DOMAIN}/admin/billing?checkout=success"
 
-    try:
-        return await create_checkout_service(
-            billing_period=billing_period,
-            seats=seats,
-            email=email,
-            license_data=license_data,
-            redirect_url=redirect_url,
-            tenant_id=tenant_id,
-        )
-    except BillingServiceError as e:
-        raise HTTPException(status_code=e.status_code, detail=e.message)
+    return await create_checkout_service(
+        billing_period=billing_period,
+        seats=seats,
+        email=email,
+        license_data=license_data,
+        redirect_url=redirect_url,
+        tenant_id=tenant_id,
+    )
 
 
 @router.post("/create-customer-portal-session")
@@ -206,18 +203,15 @@ async def create_customer_portal_session(
 
     # Self-hosted requires license
     if not MULTI_TENANT and not license_data:
-        raise HTTPException(status_code=400, detail="No license found")
+        raise OnyxError(OnyxErrorCode.VALIDATION_ERROR, "No license found")
 
     return_url = request.return_url if request else f"{WEB_DOMAIN}/admin/billing"
 
-    try:
-        return await create_portal_service(
-            license_data=license_data,
-            return_url=return_url,
-            tenant_id=tenant_id,
-        )
-    except BillingServiceError as e:
-        raise HTTPException(status_code=e.status_code, detail=e.message)
+    return await create_portal_service(
+        license_data=license_data,
+        return_url=return_url,
+        tenant_id=tenant_id,
+    )
 
 
 @router.get("/billing-information")
@@ -240,9 +234,9 @@ async def get_billing_information(
 
     # Check circuit breaker (self-hosted only)
     if _is_billing_circuit_open():
-        raise HTTPException(
-            status_code=503,
-            detail="Stripe connection temporarily disabled. Click 'Connect to Stripe' to retry.",
+        raise OnyxError(
+            OnyxErrorCode.SERVICE_UNAVAILABLE,
+            "Stripe connection temporarily disabled. Click 'Connect to Stripe' to retry.",
         )
 
     try:
@@ -250,11 +244,15 @@ async def get_billing_information(
             license_data=license_data,
             tenant_id=tenant_id,
         )
-    except BillingServiceError as e:
+    except OnyxError as e:
         # Open circuit breaker on connection failures (self-hosted only)
-        if e.status_code in (502, 503, 504):
+        if e.status_code in (
+            OnyxErrorCode.BAD_GATEWAY.status_code,
+            OnyxErrorCode.SERVICE_UNAVAILABLE.status_code,
+            OnyxErrorCode.GATEWAY_TIMEOUT.status_code,
+        ):
             _open_billing_circuit()
-        raise HTTPException(status_code=e.status_code, detail=e.message)
+        raise
 
 
 @router.post("/seats/update")
@@ -274,31 +272,25 @@ async def update_seats(
 
     # Self-hosted requires license
     if not MULTI_TENANT and not license_data:
-        raise HTTPException(status_code=400, detail="No license found")
+        raise OnyxError(OnyxErrorCode.VALIDATION_ERROR, "No license found")
 
     # Validate that new seat count is not less than current used seats
     used_seats = get_used_seats(tenant_id)
     if request.new_seat_count < used_seats:
-        raise HTTPException(
-            status_code=400,
-            detail=f"Cannot reduce seats below current usage. "
+        raise OnyxError(
+            OnyxErrorCode.VALIDATION_ERROR,
+            f"Cannot reduce seats below current usage. "
             f"You have {used_seats} active users/integrations but requested {request.new_seat_count} seats.",
         )
 
-    try:
-        result = await update_seat_service(
-            new_seat_count=request.new_seat_count,
-            license_data=license_data,
-            tenant_id=tenant_id,
-        )
-
-        # Note: Don't store license here - the control plane may still be processing
-        # the subscription update. The frontend should call /license/claim after a
-        # short delay to get the freshly generated license.
-
-        return result
-    except BillingServiceError as e:
-        raise HTTPException(status_code=e.status_code, detail=e.message)
+    # Note: Don't store license here - the control plane may still be processing
+    # the subscription update. The frontend should call /license/claim after a
+    # short delay to get the freshly generated license.
+    return await update_seat_service(
+        new_seat_count=request.new_seat_count,
+        license_data=license_data,
+        tenant_id=tenant_id,
+    )
 
 
 @router.get("/stripe-publishable-key")
@@ -329,18 +321,18 @@ async def get_stripe_publishable_key() -> StripePublishableKeyResponse:
         if STRIPE_PUBLISHABLE_KEY_OVERRIDE:
             key = STRIPE_PUBLISHABLE_KEY_OVERRIDE.strip()
             if not key.startswith("pk_"):
-                raise HTTPException(
-                    status_code=500,
-                    detail="Invalid Stripe publishable key format",
+                raise OnyxError(
+                    OnyxErrorCode.INTERNAL_ERROR,
+                    "Invalid Stripe publishable key format",
                 )
             _stripe_publishable_key_cache = key
             return StripePublishableKeyResponse(publishable_key=key)
 
         # Fall back to S3 bucket
         if not STRIPE_PUBLISHABLE_KEY_URL:
-            raise HTTPException(
-                status_code=500,
-                detail="Stripe publishable key is not configured",
+            raise OnyxError(
+                OnyxErrorCode.INTERNAL_ERROR,
+                "Stripe publishable key is not configured",
             )
 
         try:
@@ -351,17 +343,17 @@ async def get_stripe_publishable_key() -> StripePublishableKeyResponse:
 
                 # Validate key format
                 if not key.startswith("pk_"):
-                    raise HTTPException(
-                        status_code=500,
-                        detail="Invalid Stripe publishable key format",
+                    raise OnyxError(
+                        OnyxErrorCode.INTERNAL_ERROR,
+                        "Invalid Stripe publishable key format",
                     )
 
                 _stripe_publishable_key_cache = key
                 return StripePublishableKeyResponse(publishable_key=key)
         except httpx.HTTPError:
-            raise HTTPException(
-                status_code=500,
-                detail="Failed to fetch Stripe publishable key",
+            raise OnyxError(
+                OnyxErrorCode.INTERNAL_ERROR,
+                "Failed to fetch Stripe publishable key",
             )
 
 

backend/ee/onyx/server/billing/service.py

@@ -22,6 +22,8 @@ from ee.onyx.server.billing.models import SeatUpdateResponse
 from ee.onyx.server.billing.models import SubscriptionStatusResponse
 from ee.onyx.server.tenants.access import generate_data_plane_token
 from onyx.configs.app_configs import CONTROL_PLANE_API_BASE_URL
+from onyx.error_handling.error_codes import OnyxErrorCode
+from onyx.error_handling.exceptions import OnyxError
 from onyx.utils.logger import setup_logger
 from shared_configs.configs import MULTI_TENANT
 
@@ -31,15 +33,6 @@ logger = setup_logger()
 _REQUEST_TIMEOUT = 30.0
 
 
-class BillingServiceError(Exception):
-    """Exception raised for billing service errors."""
-
-    def __init__(self, message: str, status_code: int = 500):
-        self.message = message
-        self.status_code = status_code
-        super().__init__(self.message)
-
-
 def _get_proxy_headers(license_data: str | None) -> dict[str, str]:
     """Build headers for proxy requests (self-hosted).
 
@@ -101,7 +94,7 @@ async def _make_billing_request(
         Response JSON as dict
 
     Raises:
-        BillingServiceError: If request fails
+        OnyxError: If request fails
     """
 
     base_url = _get_base_url()
@@ -128,11 +121,17 @@ async def _make_billing_request(
         except Exception:
             pass
         logger.error(f"{error_message}: {e.response.status_code} - {detail}")
-        raise BillingServiceError(detail, e.response.status_code)
+        raise OnyxError(
+            OnyxErrorCode.BAD_GATEWAY,
+            detail,
+            status_code_override=e.response.status_code,
+        )
 
     except httpx.RequestError:
         logger.exception("Failed to connect to billing service")
-        raise BillingServiceError("Failed to connect to billing service", 502)
+        raise OnyxError(
+            OnyxErrorCode.BAD_GATEWAY, "Failed to connect to billing service"
+        )
 
 
 async def create_checkout_session(

backend/ee/onyx/server/enterprise_settings/api.py

@@ -223,6 +223,15 @@ def get_active_scim_token(
     token = dal.get_active_token()
     if not token:
         raise HTTPException(status_code=404, detail="No active SCIM token")
+
+    # Derive the IdP domain from the first synced user as a heuristic.
+    idp_domain: str | None = None
+    mappings, _total = dal.list_user_mappings(start_index=1, count=1)
+    if mappings:
+        user = dal.get_user(mappings[0].user_id)
+        if user and "@" in user.email:
+            idp_domain = user.email.rsplit("@", 1)[1]
+
     return ScimTokenResponse(
         id=token.id,
         name=token.name,
@@ -230,6 +239,7 @@ def get_active_scim_token(
         is_active=token.is_active,
         created_at=token.created_at,
         last_used_at=token.last_used_at,
+        idp_domain=idp_domain,
     )
 
 

backend/ee/onyx/server/license/api.py

@@ -14,7 +14,6 @@ import requests
 from fastapi import APIRouter
 from fastapi import Depends
 from fastapi import File
-from fastapi import HTTPException
 from fastapi import UploadFile
 from sqlalchemy.orm import Session
 
@@ -35,6 +34,8 @@ from ee.onyx.server.license.models import SeatUsageResponse
 from ee.onyx.utils.license import verify_license_signature
 from onyx.auth.users import User
 from onyx.db.engine.sql_engine import get_session
+from onyx.error_handling.error_codes import OnyxErrorCode
+from onyx.error_handling.exceptions import OnyxError
 from onyx.utils.logger import setup_logger
 from shared_configs.configs import MULTI_TENANT
 
@@ -127,9 +128,9 @@ async def claim_license(
     2. Without session_id: Re-claim using existing license for auth
     """
     if MULTI_TENANT:
-        raise HTTPException(
-            status_code=400,
-            detail="License claiming is only available for self-hosted deployments",
+        raise OnyxError(
+            OnyxErrorCode.VALIDATION_ERROR,
+            "License claiming is only available for self-hosted deployments",
         )
 
     try:
@@ -146,15 +147,16 @@ async def claim_license(
             # Re-claim using existing license for auth
             metadata = get_license_metadata(db_session)
             if not metadata or not metadata.tenant_id:
-                raise HTTPException(
-                    status_code=400,
-                    detail="No license found. Provide session_id after checkout.",
+                raise OnyxError(
+                    OnyxErrorCode.VALIDATION_ERROR,
+                    "No license found. Provide session_id after checkout.",
                 )
 
             license_row = get_license(db_session)
             if not license_row or not license_row.license_data:
-                raise HTTPException(
-                    status_code=400, detail="No license found in database"
+                raise OnyxError(
+                    OnyxErrorCode.VALIDATION_ERROR,
+                    "No license found in database",
                 )
 
             url = f"{CLOUD_DATA_PLANE_URL}/proxy/license/{metadata.tenant_id}"
@@ -173,7 +175,7 @@ async def claim_license(
         license_data = data.get("license")
 
         if not license_data:
-            raise HTTPException(status_code=404, detail="No license in response")
+            raise OnyxError(OnyxErrorCode.NOT_FOUND, "No license in response")
 
         # Verify signature before persisting
         payload = verify_license_signature(license_data)
@@ -199,12 +201,14 @@ async def claim_license(
             detail = error_data.get("detail", detail)
         except Exception:
             pass
-        raise HTTPException(status_code=status_code, detail=detail)
+        raise OnyxError(
+            OnyxErrorCode.BAD_GATEWAY, detail, status_code_override=status_code
+        )
     except ValueError as e:
-        raise HTTPException(status_code=400, detail=str(e))
+        raise OnyxError(OnyxErrorCode.VALIDATION_ERROR, str(e))
     except requests.RequestException:
-        raise HTTPException(
-            status_code=502, detail="Failed to connect to license server"
+        raise OnyxError(
+            OnyxErrorCode.BAD_GATEWAY, "Failed to connect to license server"
         )
 
 
@@ -221,9 +225,9 @@ async def upload_license(
     The license file must be cryptographically signed by Onyx.
     """
     if MULTI_TENANT:
-        raise HTTPException(
-            status_code=400,
-            detail="License upload is only available for self-hosted deployments",
+        raise OnyxError(
+            OnyxErrorCode.VALIDATION_ERROR,
+            "License upload is only available for self-hosted deployments",
         )
 
     try:
@@ -234,14 +238,14 @@ async def upload_license(
         # Remove any stray whitespace/newlines from user input
         license_data = license_data.strip()
     except UnicodeDecodeError:
-        raise HTTPException(status_code=400, detail="Invalid license file format")
+        raise OnyxError(OnyxErrorCode.INVALID_INPUT, "Invalid license file format")
 
     # Verify cryptographic signature - this is the only validation needed
     # The license's tenant_id identifies the customer in control plane, not locally
     try:
         payload = verify_license_signature(license_data)
     except ValueError as e:
-        raise HTTPException(status_code=400, detail=str(e))
+        raise OnyxError(OnyxErrorCode.VALIDATION_ERROR, str(e))
 
     # Persist to DB and update cache
     upsert_license(db_session, license_data)
@@ -297,9 +301,9 @@ async def delete_license(
     Admin only - removes license from database and invalidates cache.
     """
     if MULTI_TENANT:
-        raise HTTPException(
-            status_code=400,
-            detail="License deletion is only available for self-hosted deployments",
+        raise OnyxError(
+            OnyxErrorCode.VALIDATION_ERROR,
+            "License deletion is only available for self-hosted deployments",
         )
 
     try:

backend/ee/onyx/server/middleware/license_enforcement.py

@@ -46,7 +46,6 @@ from fastapi import FastAPI
 from fastapi import Request
 from fastapi import Response
 from fastapi.responses import JSONResponse
-from redis.exceptions import RedisError
 from sqlalchemy.exc import SQLAlchemyError
 
 from ee.onyx.configs.app_configs import LICENSE_ENFORCEMENT_ENABLED
@@ -56,6 +55,7 @@ from ee.onyx.configs.license_enforcement_config import (
 )
 from ee.onyx.db.license import get_cached_license_metadata
 from ee.onyx.db.license import refresh_license_cache
+from onyx.cache.interface import CACHE_TRANSIENT_ERRORS
 from onyx.db.engine.sql_engine import get_session_with_current_tenant
 from onyx.server.settings.models import ApplicationStatus
 from shared_configs.contextvars import get_current_tenant_id
@@ -164,9 +164,9 @@ def add_license_enforcement_middleware(
                     "[license_enforcement] No license, allowing community features"
                 )
                 is_gated = False
-        except RedisError as e:
+        except CACHE_TRANSIENT_ERRORS as e:
             logger.warning(f"Failed to check license metadata: {e}")
-            # Fail open - don't block users due to Redis connectivity issues
+            # Fail open - don't block users due to cache connectivity issues
             is_gated = False
 
         if is_gated:

backend/ee/onyx/server/scim/api.py

@@ -423,15 +423,63 @@ def create_user(
 
     email = user_resource.userName.strip()
 
-    # Enforce seat limit
+    # Check for existing user — if they exist but aren't SCIM-managed yet,
+    # link them to the IdP rather than rejecting with 409.
+    external_id: str | None = user_resource.externalId
+    scim_username: str = user_resource.userName.strip()
+    fields: ScimMappingFields = _fields_from_resource(user_resource)
+
+    existing_user = dal.get_user_by_email(email)
+    if existing_user:
+        existing_mapping = dal.get_user_mapping_by_user_id(existing_user.id)
+        if existing_mapping:
+            return _scim_error_response(409, f"User with email {email} already exists")
+
+        # Adopt pre-existing user into SCIM management.
+        # Reactivating a deactivated user consumes a seat, so enforce the
+        # seat limit the same way replace_user does.
+        if user_resource.active and not existing_user.is_active:
+            seat_error = _check_seat_availability(dal)
+            if seat_error:
+                return _scim_error_response(403, seat_error)
+
+        personal_name = _scim_name_to_str(user_resource.name)
+        dal.update_user(
+            existing_user,
+            is_active=user_resource.active,
+            **({"personal_name": personal_name} if personal_name else {}),
+        )
+
+        try:
+            dal.create_user_mapping(
+                external_id=external_id,
+                user_id=existing_user.id,
+                scim_username=scim_username,
+                fields=fields,
+            )
+            dal.commit()
+        except IntegrityError:
+            dal.rollback()
+            return _scim_error_response(
+                409, f"User with email {email} already has a SCIM mapping"
+            )
+
+        return _scim_resource_response(
+            provider.build_user_resource(
+                existing_user,
+                external_id,
+                scim_username=scim_username,
+                fields=fields,
+            ),
+            status_code=201,
+        )
+
+    # Only enforce seat limit for net-new users — adopting a pre-existing
+    # user doesn't consume a new seat.
     seat_error = _check_seat_availability(dal)
     if seat_error:
         return _scim_error_response(403, seat_error)
 
-    # Check for existing user
-    if dal.get_user_by_email(email):
-        return _scim_error_response(409, f"User with email {email} already exists")
-
     # Create user with a random password (SCIM users authenticate via IdP)
     personal_name = _scim_name_to_str(user_resource.name)
     user = User(
@@ -449,21 +497,21 @@ def create_user(
         dal.rollback()
         return _scim_error_response(409, f"User with email {email} already exists")
 
-    # Create SCIM mapping when externalId is provided — this is how the IdP
-    # correlates this user on subsequent requests.  Per RFC 7643, externalId
-    # is optional and assigned by the provisioning client.
-    external_id = user_resource.externalId
-    scim_username = user_resource.userName.strip()
-    fields = _fields_from_resource(user_resource)
-    if external_id:
+    # Always create a SCIM mapping so that the user is marked as
+    # SCIM-managed. externalId may be None (RFC 7643 says it's optional).
+    try:
         dal.create_user_mapping(
             external_id=external_id,
             user_id=user.id,
             scim_username=scim_username,
             fields=fields,
         )
-
-    dal.commit()
+        dal.commit()
+    except IntegrityError:
+        dal.rollback()
+        return _scim_error_response(
+            409, f"User with email {email} already has a SCIM mapping"
+        )
 
     return _scim_resource_response(
         provider.build_user_resource(

backend/ee/onyx/server/scim/models.py

@@ -365,6 +365,7 @@ class ScimTokenResponse(BaseModel):
     is_active: bool
     created_at: datetime
     last_used_at: datetime | None = None
+    idp_domain: str | None = None
 
 
 class ScimTokenCreatedResponse(ScimTokenResponse):

backend/ee/onyx/server/scim/providers/base.py

@@ -170,7 +170,10 @@ class ScimProvider(ABC):
                 formatted=user.personal_name or "",
             )
         if not user.personal_name:
-            return ScimName(givenName="", familyName="", formatted="")
+            # Derive a reasonable name from the email so that SCIM spec tests
+            # see non-empty givenName / familyName for every user resource.
+            local = user.email.split("@")[0] if user.email else ""
+            return ScimName(givenName=local, familyName="", formatted=local)
         parts = user.personal_name.split(" ", 1)
         return ScimName(
             givenName=parts[0],

backend/ee/onyx/server/settings/api.py

@@ -6,6 +6,7 @@ from sqlalchemy.exc import SQLAlchemyError
 from ee.onyx.configs.app_configs import LICENSE_ENFORCEMENT_ENABLED
 from ee.onyx.db.license import get_cached_license_metadata
 from ee.onyx.db.license import refresh_license_cache
+from onyx.cache.interface import CACHE_TRANSIENT_ERRORS
 from onyx.configs.app_configs import ENTERPRISE_EDITION_ENABLED
 from onyx.db.engine.sql_engine import get_session_with_current_tenant
 from onyx.server.settings.models import ApplicationStatus
@@ -125,7 +126,7 @@ def apply_license_status_to_settings(settings: Settings) -> Settings:
                 # syncing) means indexed data may need protection.
                 settings.application_status = _BLOCKING_STATUS
             settings.ee_features_enabled = False
-    except RedisError as e:
+    except CACHE_TRANSIENT_ERRORS as e:
         logger.warning(f"Failed to check license metadata for settings: {e}")
         # Fail closed - disable EE features if we can't verify license
         settings.ee_features_enabled = False

backend/ee/onyx/server/tenants/billing_api.py

@@ -21,7 +21,6 @@ import asyncio
 import httpx
 from fastapi import APIRouter
 from fastapi import Depends
-from fastapi import HTTPException
 
 from ee.onyx.auth.users import current_admin_user
 from ee.onyx.server.tenants.access import control_plane_dep
@@ -43,6 +42,8 @@ from onyx.auth.users import User
 from onyx.configs.app_configs import STRIPE_PUBLISHABLE_KEY_OVERRIDE
 from onyx.configs.app_configs import STRIPE_PUBLISHABLE_KEY_URL
 from onyx.configs.app_configs import WEB_DOMAIN
+from onyx.error_handling.error_codes import OnyxErrorCode
+from onyx.error_handling.exceptions import OnyxError
 from onyx.utils.logger import setup_logger
 from shared_configs.contextvars import CURRENT_TENANT_ID_CONTEXTVAR
 from shared_configs.contextvars import get_current_tenant_id
@@ -116,9 +117,14 @@ async def create_customer_portal_session(
     try:
         portal_url = fetch_customer_portal_session(tenant_id, return_url)
         return {"stripe_customer_portal_url": portal_url}
-    except Exception as e:
+    except OnyxError:
+        raise
+    except Exception:
         logger.exception("Failed to create customer portal session")
-        raise HTTPException(status_code=500, detail=str(e))
+        raise OnyxError(
+            OnyxErrorCode.INTERNAL_ERROR,
+            "Failed to create customer portal session",
+        )
 
 
 @router.post("/create-checkout-session")
@@ -134,9 +140,14 @@ async def create_checkout_session(
     try:
         checkout_url = fetch_stripe_checkout_session(tenant_id, billing_period, seats)
         return {"stripe_checkout_url": checkout_url}
-    except Exception as e:
+    except OnyxError:
+        raise
+    except Exception:
         logger.exception("Failed to create checkout session")
-        raise HTTPException(status_code=500, detail=str(e))
+        raise OnyxError(
+            OnyxErrorCode.INTERNAL_ERROR,
+            "Failed to create checkout session",
+        )
 
 
 @router.post("/create-subscription-session")
@@ -147,15 +158,20 @@ async def create_subscription_session(
     try:
         tenant_id = CURRENT_TENANT_ID_CONTEXTVAR.get()
         if not tenant_id:
-            raise HTTPException(status_code=400, detail="Tenant ID not found")
+            raise OnyxError(OnyxErrorCode.VALIDATION_ERROR, "Tenant ID not found")
 
         billing_period = request.billing_period if request else "monthly"
         session_id = fetch_stripe_checkout_session(tenant_id, billing_period)
         return SubscriptionSessionResponse(sessionId=session_id)
 
-    except Exception as e:
+    except OnyxError:
+        raise
+    except Exception:
         logger.exception("Failed to create subscription session")
-        raise HTTPException(status_code=500, detail=str(e))
+        raise OnyxError(
+            OnyxErrorCode.INTERNAL_ERROR,
+            "Failed to create subscription session",
+        )
 
 
 @router.get("/stripe-publishable-key")
@@ -186,18 +202,18 @@ async def get_stripe_publishable_key() -> StripePublishableKeyResponse:
         if STRIPE_PUBLISHABLE_KEY_OVERRIDE:
             key = STRIPE_PUBLISHABLE_KEY_OVERRIDE.strip()
             if not key.startswith("pk_"):
-                raise HTTPException(
-                    status_code=500,
-                    detail="Invalid Stripe publishable key format",
+                raise OnyxError(
+                    OnyxErrorCode.INTERNAL_ERROR,
+                    "Invalid Stripe publishable key format",
                 )
             _stripe_publishable_key_cache = key
             return StripePublishableKeyResponse(publishable_key=key)
 
         # Fall back to S3 bucket
         if not STRIPE_PUBLISHABLE_KEY_URL:
-            raise HTTPException(
-                status_code=500,
-                detail="Stripe publishable key is not configured",
+            raise OnyxError(
+                OnyxErrorCode.INTERNAL_ERROR,
+                "Stripe publishable key is not configured",
             )
 
         try:
@@ -208,15 +224,15 @@ async def get_stripe_publishable_key() -> StripePublishableKeyResponse:
 
                 # Validate key format
                 if not key.startswith("pk_"):
-                    raise HTTPException(
-                        status_code=500,
-                        detail="Invalid Stripe publishable key format",
+                    raise OnyxError(
+                        OnyxErrorCode.INTERNAL_ERROR,
+                        "Invalid Stripe publishable key format",
                     )
 
                 _stripe_publishable_key_cache = key
                 return StripePublishableKeyResponse(publishable_key=key)
         except httpx.HTTPError:
-            raise HTTPException(
-                status_code=500,
-                detail="Failed to fetch Stripe publishable key",
+            raise OnyxError(
+                OnyxErrorCode.INTERNAL_ERROR,
+                "Failed to fetch Stripe publishable key",
             )

backend/ee/onyx/server/user_group/api.py

@@ -5,6 +5,8 @@ from sqlalchemy.exc import IntegrityError
 from sqlalchemy.orm import Session
 
 from ee.onyx.db.user_group import add_users_to_user_group
+from ee.onyx.db.user_group import delete_user_group as db_delete_user_group
+from ee.onyx.db.user_group import fetch_user_group
 from ee.onyx.db.user_group import fetch_user_groups
 from ee.onyx.db.user_group import fetch_user_groups_for_user
 from ee.onyx.db.user_group import insert_user_group
@@ -20,6 +22,7 @@ from ee.onyx.server.user_group.models import UserGroupUpdate
 from onyx.auth.users import current_admin_user
 from onyx.auth.users import current_curator_or_admin_user
 from onyx.auth.users import current_user
+from onyx.configs.app_configs import DISABLE_VECTOR_DB
 from onyx.configs.constants import PUBLIC_API_TAGS
 from onyx.db.engine.sql_engine import get_session
 from onyx.db.models import User
@@ -153,3 +156,8 @@ def delete_user_group(
         prepare_user_group_for_deletion(db_session, user_group_id)
     except ValueError as e:
         raise HTTPException(status_code=404, detail=str(e))
+
+    if DISABLE_VECTOR_DB:
+        user_group = fetch_user_group(db_session, user_group_id)
+        if user_group:
+            db_delete_user_group(db_session, user_group)

backend/onyx/auth/users.py

@@ -120,7 +120,6 @@ from onyx.db.models import User
 from onyx.db.pat import fetch_user_for_pat
 from onyx.db.users import get_user_by_email
 from onyx.redis.redis_pool import get_async_redis_connection
-from onyx.redis.redis_pool import get_redis_client
 from onyx.server.settings.store import load_settings
 from onyx.server.utils import BasicAuthenticationError
 from onyx.utils.logger import setup_logger
@@ -201,13 +200,14 @@ def user_needs_to_be_verified() -> bool:
 
 
 def anonymous_user_enabled(*, tenant_id: str | None = None) -> bool:
-    redis_client = get_redis_client(tenant_id=tenant_id)
-    value = redis_client.get(OnyxRedisLocks.ANONYMOUS_USER_ENABLED)
+    from onyx.cache.factory import get_cache_backend
+
+    cache = get_cache_backend(tenant_id=tenant_id)
+    value = cache.get(OnyxRedisLocks.ANONYMOUS_USER_ENABLED)
 
     if value is None:
         return False
 
-    assert isinstance(value, bytes)
     return int(value.decode("utf-8")) == 1
 
 

backend/onyx/background/celery/apps/background.py

@@ -1,142 +0,0 @@
-from typing import Any
-from typing import cast
-
-from celery import Celery
-from celery import signals
-from celery import Task
-from celery.apps.worker import Worker
-from celery.signals import celeryd_init
-from celery.signals import worker_init
-from celery.signals import worker_process_init
-from celery.signals import worker_ready
-from celery.signals import worker_shutdown
-
-import onyx.background.celery.apps.app_base as app_base
-from onyx.background.celery.celery_utils import httpx_init_vespa_pool
-from onyx.configs.app_configs import MANAGED_VESPA
-from onyx.configs.app_configs import VESPA_CLOUD_CERT_PATH
-from onyx.configs.app_configs import VESPA_CLOUD_KEY_PATH
-from onyx.configs.constants import POSTGRES_CELERY_WORKER_BACKGROUND_APP_NAME
-from onyx.db.engine.sql_engine import SqlEngine
-from onyx.utils.logger import setup_logger
-from shared_configs.configs import MULTI_TENANT
-
-
-logger = setup_logger()
-
-celery_app = Celery(__name__)
-celery_app.config_from_object("onyx.background.celery.configs.background")
-celery_app.Task = app_base.TenantAwareTask  # type: ignore [misc]
-
-
-@signals.task_prerun.connect
-def on_task_prerun(
-    sender: Any | None = None,
-    task_id: str | None = None,
-    task: Task | None = None,
-    args: tuple | None = None,
-    kwargs: dict | None = None,
-    **kwds: Any,
-) -> None:
-    app_base.on_task_prerun(sender, task_id, task, args, kwargs, **kwds)
-
-
-@signals.task_postrun.connect
-def on_task_postrun(
-    sender: Any | None = None,
-    task_id: str | None = None,
-    task: Task | None = None,
-    args: tuple | None = None,
-    kwargs: dict | None = None,
-    retval: Any | None = None,
-    state: str | None = None,
-    **kwds: Any,
-) -> None:
-    app_base.on_task_postrun(sender, task_id, task, args, kwargs, retval, state, **kwds)
-
-
-@celeryd_init.connect
-def on_celeryd_init(sender: str, conf: Any = None, **kwargs: Any) -> None:
-    app_base.on_celeryd_init(sender, conf, **kwargs)
-
-
-@worker_init.connect
-def on_worker_init(sender: Worker, **kwargs: Any) -> None:
-    EXTRA_CONCURRENCY = 8  # small extra fudge factor for connection limits
-
-    logger.info("worker_init signal received for consolidated background worker.")
-
-    SqlEngine.set_app_name(POSTGRES_CELERY_WORKER_BACKGROUND_APP_NAME)
-    pool_size = cast(int, sender.concurrency)  # type: ignore
-    SqlEngine.init_engine(pool_size=pool_size, max_overflow=EXTRA_CONCURRENCY)
-
-    # Initialize Vespa httpx pool (needed for light worker tasks)
-    if MANAGED_VESPA:
-        httpx_init_vespa_pool(
-            sender.concurrency + EXTRA_CONCURRENCY,  # type: ignore
-            ssl_cert=VESPA_CLOUD_CERT_PATH,
-            ssl_key=VESPA_CLOUD_KEY_PATH,
-        )
-    else:
-        httpx_init_vespa_pool(sender.concurrency + EXTRA_CONCURRENCY)  # type: ignore
-
-    app_base.wait_for_redis(sender, **kwargs)
-    app_base.wait_for_db(sender, **kwargs)
-    app_base.wait_for_vespa_or_shutdown(sender, **kwargs)
-
-    # Less startup checks in multi-tenant case
-    if MULTI_TENANT:
-        return
-
-    app_base.on_secondary_worker_init(sender, **kwargs)
-
-
-@worker_ready.connect
-def on_worker_ready(sender: Any, **kwargs: Any) -> None:
-    app_base.on_worker_ready(sender, **kwargs)
-
-
-@worker_shutdown.connect
-def on_worker_shutdown(sender: Any, **kwargs: Any) -> None:
-    app_base.on_worker_shutdown(sender, **kwargs)
-
-
-@worker_process_init.connect
-def init_worker(**kwargs: Any) -> None:  # noqa: ARG001
-    SqlEngine.reset_engine()
-
-
-@signals.setup_logging.connect
-def on_setup_logging(
-    loglevel: Any, logfile: Any, format: Any, colorize: Any, **kwargs: Any
-) -> None:
-    app_base.on_setup_logging(loglevel, logfile, format, colorize, **kwargs)
-
-
-base_bootsteps = app_base.get_bootsteps()
-for bootstep in base_bootsteps:
-    celery_app.steps["worker"].add(bootstep)
-
-celery_app.autodiscover_tasks(
-    app_base.filter_task_modules(
-        [
-            # Original background worker tasks
-            "onyx.background.celery.tasks.pruning",
-            "onyx.background.celery.tasks.monitoring",
-            "onyx.background.celery.tasks.user_file_processing",
-            "onyx.background.celery.tasks.llm_model_update",
-            # Light worker tasks
-            "onyx.background.celery.tasks.shared",
-            "onyx.background.celery.tasks.vespa",
-            "onyx.background.celery.tasks.connector_deletion",
-            "onyx.background.celery.tasks.doc_permission_syncing",
-            "onyx.background.celery.tasks.opensearch_migration",
-            # Docprocessing worker tasks
-            "onyx.background.celery.tasks.docprocessing",
-            # Docfetching worker tasks
-            "onyx.background.celery.tasks.docfetching",
-            # Sandbox cleanup tasks (isolated in build feature)
-            "onyx.server.features.build.sandbox.tasks",
-        ]
-    )
-)

backend/onyx/background/celery/celery_utils.py

@@ -39,9 +39,13 @@ CT = TypeVar("CT", bound=ConnectorCheckpoint)
 
 
 class SlimConnectorExtractionResult(BaseModel):
-    """Result of extracting document IDs and hierarchy nodes from a connector."""
+    """Result of extracting document IDs and hierarchy nodes from a connector.
 
-    doc_ids: set[str]
+    raw_id_to_parent maps document ID → parent_hierarchy_raw_node_id (or None).
+    Use raw_id_to_parent.keys() wherever the old set of IDs was needed.
+    """
+
+    raw_id_to_parent: dict[str, str | None]
     hierarchy_nodes: list[HierarchyNode]
 
 
@@ -93,30 +97,37 @@ def _get_failure_id(failure: ConnectorFailure) -> str | None:
     return None
 
 
+class BatchResult(BaseModel):
+    raw_id_to_parent: dict[str, str | None]
+    hierarchy_nodes: list[HierarchyNode]
+
+
 def _extract_from_batch(
     doc_list: Sequence[Document | SlimDocument | HierarchyNode | ConnectorFailure],
-) -> tuple[set[str], list[HierarchyNode]]:
-    """Separate a batch into document IDs and hierarchy nodes.
+) -> BatchResult:
+    """Separate a batch into document IDs (with parent mapping) and hierarchy nodes.
 
     ConnectorFailure items have their failed document/entity IDs added to the
-    ID set so that failed-to-retrieve documents are not accidentally pruned.
+    ID dict so that failed-to-retrieve documents are not accidentally pruned.
     """
-    ids: set[str] = set()
+    ids: dict[str, str | None] = {}
     hierarchy_nodes: list[HierarchyNode] = []
     for item in doc_list:
         if isinstance(item, HierarchyNode):
             hierarchy_nodes.append(item)
-            ids.add(item.raw_node_id)
+            if item.raw_node_id not in ids:
+                ids[item.raw_node_id] = None
         elif isinstance(item, ConnectorFailure):
             failed_id = _get_failure_id(item)
             if failed_id:
-                ids.add(failed_id)
+                ids[failed_id] = None
             logger.warning(
                 f"Failed to retrieve document {failed_id}: " f"{item.failure_message}"
             )
         else:
-            ids.add(item.id)
-    return ids, hierarchy_nodes
+            parent_raw = getattr(item, "parent_hierarchy_raw_node_id", None)
+            ids[item.id] = parent_raw
+    return BatchResult(raw_id_to_parent=ids, hierarchy_nodes=hierarchy_nodes)
 
 
 def extract_ids_from_runnable_connector(
@@ -132,7 +143,7 @@ def extract_ids_from_runnable_connector(
 
     Optionally, a callback can be passed to handle the length of each document batch.
     """
-    all_connector_doc_ids: set[str] = set()
+    all_raw_id_to_parent: dict[str, str | None] = {}
     all_hierarchy_nodes: list[HierarchyNode] = []
 
     # Sequence (covariant) lets all the specific list[...] iterator types unify here
@@ -177,15 +188,20 @@ def extract_ids_from_runnable_connector(
                 "extract_ids_from_runnable_connector: Stop signal detected"
             )
 
-        batch_ids, batch_nodes = _extract_from_batch(doc_list)
-        all_connector_doc_ids.update(doc_batch_processing_func(batch_ids))
+        batch_result = _extract_from_batch(doc_list)
+        batch_ids = batch_result.raw_id_to_parent
+        batch_nodes = batch_result.hierarchy_nodes
+        doc_batch_processing_func(batch_ids)
+        for k, v in batch_ids.items():
+            if v is not None or k not in all_raw_id_to_parent:
+                all_raw_id_to_parent[k] = v
         all_hierarchy_nodes.extend(batch_nodes)
 
         if callback:
             callback.progress("extract_ids_from_runnable_connector", len(batch_ids))
 
     return SlimConnectorExtractionResult(
-        doc_ids=all_connector_doc_ids,
+        raw_id_to_parent=all_raw_id_to_parent,
         hierarchy_nodes=all_hierarchy_nodes,
     )
 

backend/onyx/background/celery/configs/background.py

@@ -1,23 +0,0 @@
-import onyx.background.celery.configs.base as shared_config
-from onyx.configs.app_configs import CELERY_WORKER_BACKGROUND_CONCURRENCY
-
-broker_url = shared_config.broker_url
-broker_connection_retry_on_startup = shared_config.broker_connection_retry_on_startup
-broker_pool_limit = shared_config.broker_pool_limit
-broker_transport_options = shared_config.broker_transport_options
-
-redis_socket_keepalive = shared_config.redis_socket_keepalive
-redis_retry_on_timeout = shared_config.redis_retry_on_timeout
-redis_backend_health_check_interval = shared_config.redis_backend_health_check_interval
-
-result_backend = shared_config.result_backend
-result_expires = shared_config.result_expires  # 86400 seconds is the default
-
-task_default_priority = shared_config.task_default_priority
-task_acks_late = shared_config.task_acks_late
-
-worker_concurrency = CELERY_WORKER_BACKGROUND_CONCURRENCY
-worker_pool = "threads"
-# Increased from 1 to 4 to handle fast light worker tasks more efficiently
-# This allows the worker to prefetch multiple tasks per thread
-worker_prefetch_multiplier = 4

backend/onyx/background/celery/tasks/opensearch_migration/tasks.py

@@ -30,6 +30,7 @@ from onyx.background.celery.tasks.opensearch_migration.transformer import (
     transform_vespa_chunks_to_opensearch_chunks,
 )
 from onyx.configs.app_configs import ENABLE_OPENSEARCH_INDEXING_FOR_ONYX
+from onyx.configs.app_configs import VESPA_MIGRATION_REQUEST_TIMEOUT_S
 from onyx.configs.constants import OnyxCeleryTask
 from onyx.configs.constants import OnyxRedisLocks
 from onyx.db.engine.sql_engine import get_session_with_current_tenant
@@ -47,6 +48,7 @@ from onyx.document_index.interfaces_new import TenantState
 from onyx.document_index.opensearch.opensearch_document_index import (
     OpenSearchDocumentIndex,
 )
+from onyx.document_index.vespa.shared_utils.utils import get_vespa_http_client
 from onyx.document_index.vespa.vespa_document_index import VespaDocumentIndex
 from onyx.indexing.models import IndexingSetting
 from onyx.redis.redis_pool import get_redis_client
@@ -146,7 +148,12 @@ def migrate_chunks_from_vespa_to_opensearch_task(
             task_logger.error(err_str)
             return False
 
-        with get_session_with_current_tenant() as db_session:
+        with (
+            get_session_with_current_tenant() as db_session,
+            get_vespa_http_client(
+                timeout=VESPA_MIGRATION_REQUEST_TIMEOUT_S
+            ) as vespa_client,
+        ):
             try_insert_opensearch_tenant_migration_record_with_commit(db_session)
             search_settings = get_current_search_settings(db_session)
             tenant_state = TenantState(tenant_id=tenant_id, multitenant=MULTI_TENANT)
@@ -161,6 +168,7 @@ def migrate_chunks_from_vespa_to_opensearch_task(
                 index_name=search_settings.index_name,
                 tenant_state=tenant_state,
                 large_chunks_enabled=False,
+                httpx_client=vespa_client,
             )
 
             sanitized_doc_start_time = time.monotonic()

backend/onyx/background/celery/tasks/pruning/tasks.py

@@ -29,6 +29,7 @@ from onyx.configs.constants import CELERY_GENERIC_BEAT_LOCK_TIMEOUT
 from onyx.configs.constants import CELERY_PRUNING_LOCK_TIMEOUT
 from onyx.configs.constants import CELERY_TASK_WAIT_FOR_FENCE_TIMEOUT
 from onyx.configs.constants import DANSWER_REDIS_FUNCTION_LOCK_PREFIX
+from onyx.configs.constants import DocumentSource
 from onyx.configs.constants import OnyxCeleryPriority
 from onyx.configs.constants import OnyxCeleryQueues
 from onyx.configs.constants import OnyxCeleryTask
@@ -47,6 +48,8 @@ from onyx.db.enums import AccessType
 from onyx.db.enums import ConnectorCredentialPairStatus
 from onyx.db.enums import SyncStatus
 from onyx.db.enums import SyncType
+from onyx.db.hierarchy import link_hierarchy_nodes_to_documents
+from onyx.db.hierarchy import update_document_parent_hierarchy_nodes
 from onyx.db.hierarchy import upsert_hierarchy_nodes_batch
 from onyx.db.models import ConnectorCredentialPair
 from onyx.db.sync_record import insert_sync_record
@@ -57,6 +60,8 @@ from onyx.redis.redis_connector_prune import RedisConnectorPrune
 from onyx.redis.redis_connector_prune import RedisConnectorPrunePayload
 from onyx.redis.redis_hierarchy import cache_hierarchy_nodes_batch
 from onyx.redis.redis_hierarchy import ensure_source_node_exists
+from onyx.redis.redis_hierarchy import get_node_id_from_raw_id
+from onyx.redis.redis_hierarchy import get_source_node_id_from_cache
 from onyx.redis.redis_hierarchy import HierarchyNodeCacheEntry
 from onyx.redis.redis_pool import get_redis_client
 from onyx.redis.redis_pool import get_redis_replica_client
@@ -113,6 +118,38 @@ class PruneCallback(IndexingCallbackBase):
         super().progress(tag, amount)
 
 
+def _resolve_and_update_document_parents(
+    db_session: Session,
+    redis_client: Redis,
+    source: DocumentSource,
+    raw_id_to_parent: dict[str, str | None],
+) -> None:
+    """Resolve parent_hierarchy_raw_node_id → parent_hierarchy_node_id for
+    each document and bulk-update the DB. Mirrors the resolution logic in
+    run_docfetching.py."""
+    source_node_id = get_source_node_id_from_cache(redis_client, db_session, source)
+
+    resolved: dict[str, int | None] = {}
+    for doc_id, raw_parent_id in raw_id_to_parent.items():
+        if raw_parent_id is None:
+            continue
+        node_id, found = get_node_id_from_raw_id(redis_client, source, raw_parent_id)
+        resolved[doc_id] = node_id if found else source_node_id
+
+    if not resolved:
+        return
+
+    update_document_parent_hierarchy_nodes(
+        db_session=db_session,
+        doc_parent_map=resolved,
+        commit=True,
+    )
+    task_logger.info(
+        f"Pruning: resolved and updated parent hierarchy for "
+        f"{len(resolved)} documents (source={source.value})"
+    )
+
+
 """Jobs / utils for kicking off pruning tasks."""
 
 
@@ -535,22 +572,22 @@ def connector_pruning_generator_task(
             extraction_result = extract_ids_from_runnable_connector(
                 runnable_connector, callback
             )
-            all_connector_doc_ids = extraction_result.doc_ids
+            all_connector_doc_ids = extraction_result.raw_id_to_parent
 
             # Process hierarchy nodes (same as docfetching):
             # upsert to Postgres and cache in Redis
+            source = cc_pair.connector.source
+            redis_client = get_redis_client(tenant_id=tenant_id)
+
             if extraction_result.hierarchy_nodes:
                 is_connector_public = cc_pair.access_type == AccessType.PUBLIC
 
-                redis_client = get_redis_client(tenant_id=tenant_id)
-                ensure_source_node_exists(
-                    redis_client, db_session, cc_pair.connector.source
-                )
+                ensure_source_node_exists(redis_client, db_session, source)
 
                 upserted_nodes = upsert_hierarchy_nodes_batch(
                     db_session=db_session,
                     nodes=extraction_result.hierarchy_nodes,
-                    source=cc_pair.connector.source,
+                    source=source,
                     commit=True,
                     is_connector_public=is_connector_public,
                 )
@@ -561,7 +598,7 @@ def connector_pruning_generator_task(
                 ]
                 cache_hierarchy_nodes_batch(
                     redis_client=redis_client,
-                    source=cc_pair.connector.source,
+                    source=source,
                     entries=cache_entries,
                 )
 
@@ -570,6 +607,26 @@ def connector_pruning_generator_task(
                     f"hierarchy nodes for cc_pair={cc_pair_id}"
                 )
 
+            ensure_source_node_exists(redis_client, db_session, source)
+            # Resolve parent_hierarchy_raw_node_id → parent_hierarchy_node_id
+            # and bulk-update documents, mirroring the docfetching resolution
+            _resolve_and_update_document_parents(
+                db_session=db_session,
+                redis_client=redis_client,
+                source=source,
+                raw_id_to_parent=all_connector_doc_ids,
+            )
+
+            # Link hierarchy nodes to documents for sources where pages can be
+            # both hierarchy nodes AND documents (e.g. Notion, Confluence)
+            all_doc_id_list = list(all_connector_doc_ids.keys())
+            link_hierarchy_nodes_to_documents(
+                db_session=db_session,
+                document_ids=all_doc_id_list,
+                source=source,
+                commit=True,
+            )
+
             # a list of docs in our local index
             all_indexed_document_ids = {
                 doc.id
@@ -581,7 +638,9 @@ def connector_pruning_generator_task(
             }
 
             # generate list of docs to remove (no longer in the source)
-            doc_ids_to_remove = list(all_indexed_document_ids - all_connector_doc_ids)
+            doc_ids_to_remove = list(
+                all_indexed_document_ids - all_connector_doc_ids.keys()
+            )
 
             task_logger.info(
                 "Pruning set collected: "

backend/onyx/background/celery/tasks/user_file_processing/tasks.py

@@ -520,6 +520,7 @@ def process_user_file_impl(
         task_logger.exception(
             f"process_user_file_impl - Error processing file id={user_file_id} - {e.__class__.__name__}"
         )
+        raise
     finally:
         if file_lock is not None and file_lock.owned():
             file_lock.release()
@@ -675,6 +676,7 @@ def delete_user_file_impl(
         task_logger.exception(
             f"delete_user_file_impl - Error processing file id={user_file_id} - {e.__class__.__name__}"
         )
+        raise
     finally:
         if file_lock is not None and file_lock.owned():
             file_lock.release()
@@ -849,6 +851,7 @@ def project_sync_user_file_impl(
         task_logger.exception(
             f"project_sync_user_file_impl - Error syncing project for file id={user_file_id} - {e.__class__.__name__}"
         )
+        raise
     finally:
         if file_lock is not None and file_lock.owned():
             file_lock.release()

backend/onyx/background/celery/versioned_apps/background.py

@@ -1,10 +0,0 @@
-from celery import Celery
-
-from onyx.utils.variable_functionality import fetch_versioned_implementation
-from onyx.utils.variable_functionality import set_is_ee_based_on_env_variable
-
-set_is_ee_based_on_env_variable()
-app: Celery = fetch_versioned_implementation(
-    "onyx.background.celery.apps.background",
-    "celery_app",
-)

backend/onyx/background/indexing/run_docfetching.py

@@ -58,8 +58,6 @@ from onyx.file_store.document_batch_storage import DocumentBatchStorage
 from onyx.file_store.document_batch_storage import get_document_batch_storage
 from onyx.indexing.indexing_heartbeat import IndexingHeartbeatInterface
 from onyx.indexing.indexing_pipeline import index_doc_batch_prepare
-from onyx.indexing.postgres_sanitization import sanitize_document_for_postgres
-from onyx.indexing.postgres_sanitization import sanitize_hierarchy_nodes_for_postgres
 from onyx.redis.redis_hierarchy import cache_hierarchy_nodes_batch
 from onyx.redis.redis_hierarchy import ensure_source_node_exists
 from onyx.redis.redis_hierarchy import get_node_id_from_raw_id
@@ -71,6 +69,8 @@ from onyx.server.features.build.indexing.persistent_document_writer import (
 )
 from onyx.utils.logger import setup_logger
 from onyx.utils.middleware import make_randomized_onyx_request_id
+from onyx.utils.postgres_sanitization import sanitize_document_for_postgres
+from onyx.utils.postgres_sanitization import sanitize_hierarchy_nodes_for_postgres
 from onyx.utils.variable_functionality import global_version
 from shared_configs.configs import MULTI_TENANT
 from shared_configs.contextvars import INDEX_ATTEMPT_INFO_CONTEXTVAR

backend/onyx/background/periodic_poller.py

@@ -59,6 +59,12 @@ def _run_auto_llm_update() -> None:
         sync_llm_models_from_github(db_session)
 
 
+def _run_cache_cleanup() -> None:
+    from onyx.cache.postgres_backend import cleanup_expired_cache_entries
+
+    cleanup_expired_cache_entries()
+
+
 def _run_scheduled_eval() -> None:
     from onyx.configs.app_configs import BRAINTRUST_API_KEY
     from onyx.configs.app_configs import SCHEDULED_EVAL_DATASET_NAMES
@@ -100,12 +106,26 @@ def _run_scheduled_eval() -> None:
             )
 
 
+_CACHE_CLEANUP_INTERVAL_SECONDS = 300
+
+
 def _build_periodic_tasks() -> list[_PeriodicTaskDef]:
+    from onyx.cache.interface import CacheBackendType
     from onyx.configs.app_configs import AUTO_LLM_CONFIG_URL
     from onyx.configs.app_configs import AUTO_LLM_UPDATE_INTERVAL_SECONDS
+    from onyx.configs.app_configs import CACHE_BACKEND
     from onyx.configs.app_configs import SCHEDULED_EVAL_DATASET_NAMES
 
     tasks: list[_PeriodicTaskDef] = []
+    if CACHE_BACKEND == CacheBackendType.POSTGRES:
+        tasks.append(
+            _PeriodicTaskDef(
+                name="cache-cleanup",
+                interval_seconds=_CACHE_CLEANUP_INTERVAL_SECONDS,
+                lock_id=PERIODIC_TASK_LOCK_BASE + 2,
+                run_fn=_run_cache_cleanup,
+            )
+        )
     if AUTO_LLM_CONFIG_URL:
         tasks.append(
             _PeriodicTaskDef(

backend/onyx/background/task_utils.py

@@ -75,31 +75,41 @@ def _claim_next_processing_file(db_session: Session) -> UUID | None:
     return file_id
 
 
-def _claim_next_deleting_file(db_session: Session) -> UUID | None:
+def _claim_next_deleting_file(
+    db_session: Session,
+    exclude_ids: set[UUID] | None = None,
+) -> UUID | None:
     """Claim the next DELETING file.
 
     No status transition needed — the impl deletes the row on success.
     The short-lived FOR UPDATE lock prevents concurrent claims.
+    *exclude_ids* prevents re-processing the same file if the impl fails.
     """
-    file_id = db_session.execute(
+    stmt = (
         select(UserFile.id)
         .where(UserFile.status == UserFileStatus.DELETING)
         .order_by(UserFile.created_at)
         .limit(1)
         .with_for_update(skip_locked=True)
-    ).scalar_one_or_none()
-    # Commit to release the row lock promptly.
+    )
+    if exclude_ids:
+        stmt = stmt.where(UserFile.id.notin_(exclude_ids))
+    file_id = db_session.execute(stmt).scalar_one_or_none()
     db_session.commit()
     return file_id
 
 
-def _claim_next_sync_file(db_session: Session) -> UUID | None:
+def _claim_next_sync_file(
+    db_session: Session,
+    exclude_ids: set[UUID] | None = None,
+) -> UUID | None:
     """Claim the next file needing project/persona sync.
 
     No status transition needed — the impl clears the sync flags on
     success.  The short-lived FOR UPDATE lock prevents concurrent claims.
+    *exclude_ids* prevents re-processing the same file if the impl fails.
     """
-    file_id = db_session.execute(
+    stmt = (
         select(UserFile.id)
         .where(
             sa.and_(
@@ -113,7 +123,10 @@ def _claim_next_sync_file(db_session: Session) -> UUID | None:
         .order_by(UserFile.created_at)
         .limit(1)
         .with_for_update(skip_locked=True)
-    ).scalar_one_or_none()
+    )
+    if exclude_ids:
+        stmt = stmt.where(UserFile.id.notin_(exclude_ids))
+    file_id = db_session.execute(stmt).scalar_one_or_none()
     db_session.commit()
     return file_id
 
@@ -135,11 +148,14 @@ def drain_processing_loop(tenant_id: str) -> None:
             file_id = _claim_next_processing_file(session)
         if file_id is None:
             break
-        process_user_file_impl(
-            user_file_id=str(file_id),
-            tenant_id=tenant_id,
-            redis_locking=False,
-        )
+        try:
+            process_user_file_impl(
+                user_file_id=str(file_id),
+                tenant_id=tenant_id,
+                redis_locking=False,
+            )
+        except Exception:
+            logger.exception(f"Failed to process user file {file_id}")
 
 
 def drain_delete_loop(tenant_id: str) -> None:
@@ -149,16 +165,21 @@ def drain_delete_loop(tenant_id: str) -> None:
     )
     from onyx.db.engine.sql_engine import get_session_with_current_tenant
 
+    failed: set[UUID] = set()
     while True:
         with get_session_with_current_tenant() as session:
-            file_id = _claim_next_deleting_file(session)
+            file_id = _claim_next_deleting_file(session, exclude_ids=failed)
         if file_id is None:
             break
-        delete_user_file_impl(
-            user_file_id=str(file_id),
-            tenant_id=tenant_id,
-            redis_locking=False,
-        )
+        try:
+            delete_user_file_impl(
+                user_file_id=str(file_id),
+                tenant_id=tenant_id,
+                redis_locking=False,
+            )
+        except Exception:
+            logger.exception(f"Failed to delete user file {file_id}")
+            failed.add(file_id)
 
 
 def drain_project_sync_loop(tenant_id: str) -> None:
@@ -168,13 +189,18 @@ def drain_project_sync_loop(tenant_id: str) -> None:
     )
     from onyx.db.engine.sql_engine import get_session_with_current_tenant
 
+    failed: set[UUID] = set()
     while True:
         with get_session_with_current_tenant() as session:
-            file_id = _claim_next_sync_file(session)
+            file_id = _claim_next_sync_file(session, exclude_ids=failed)
         if file_id is None:
             break
-        project_sync_user_file_impl(
-            user_file_id=str(file_id),
-            tenant_id=tenant_id,
-            redis_locking=False,
-        )
+        try:
+            project_sync_user_file_impl(
+                user_file_id=str(file_id),
+                tenant_id=tenant_id,
+                redis_locking=False,
+            )
+        except Exception:
+            logger.exception(f"Failed to sync user file {file_id}")
+            failed.add(file_id)

backend/onyx/cache/factory.py

@@ -12,9 +12,15 @@ def _build_redis_backend(tenant_id: str) -> CacheBackend:
     return RedisCacheBackend(redis_pool.get_client(tenant_id))
 
 
+def _build_postgres_backend(tenant_id: str) -> CacheBackend:
+    from onyx.cache.postgres_backend import PostgresCacheBackend
+
+    return PostgresCacheBackend(tenant_id)
+
+
 _BACKEND_BUILDERS: dict[CacheBackendType, Callable[[str], CacheBackend]] = {
     CacheBackendType.REDIS: _build_redis_backend,
-    # CacheBackendType.POSTGRES will be added in a follow-up PR.
+    CacheBackendType.POSTGRES: _build_postgres_backend,
 }
 
 

backend/onyx/cache/interface.py

@@ -1,6 +1,20 @@
 import abc
 from enum import Enum
 
+from redis.exceptions import RedisError
+from sqlalchemy.exc import SQLAlchemyError
+
+TTL_KEY_NOT_FOUND = -2
+TTL_NO_EXPIRY = -1
+
+CACHE_TRANSIENT_ERRORS: tuple[type[Exception], ...] = (RedisError, SQLAlchemyError)
+"""Exception types that represent transient cache connectivity / operational
+failures.  Callers that want to fail-open (or fail-closed) on cache errors
+should catch this tuple instead of bare ``Exception``.
+
+When adding a new ``CacheBackend`` implementation, add its transient error
+base class(es) here so all call-sites pick it up automatically."""
+
 
 class CacheBackendType(str, Enum):
     REDIS = "redis"
@@ -26,6 +40,14 @@ class CacheLock(abc.ABC):
     def owned(self) -> bool:
         raise NotImplementedError
 
+    def __enter__(self) -> "CacheLock":
+        if not self.acquire():
+            raise RuntimeError("Failed to acquire lock")
+        return self
+
+    def __exit__(self, *args: object) -> None:
+        self.release()
+
 
 class CacheBackend(abc.ABC):
     """Thin abstraction over a key-value cache with TTL, locks, and blocking lists.
@@ -65,7 +87,11 @@ class CacheBackend(abc.ABC):
 
     @abc.abstractmethod
     def ttl(self, key: str) -> int:
-        """Return remaining TTL in seconds. -1 if no expiry, -2 if key missing."""
+        """Return remaining TTL in seconds.
+
+        Returns ``TTL_NO_EXPIRY`` (-1) if key exists without expiry,
+        ``TTL_KEY_NOT_FOUND`` (-2) if key is missing or expired.
+        """
         raise NotImplementedError
 
     # -- distributed lock --------------------------------------------------

backend/onyx/cache/postgres_backend.py

@@ -0,0 +1,323 @@
+"""PostgreSQL-backed ``CacheBackend`` for NO_VECTOR_DB deployments.
+
+Uses the ``cache_store`` table for key-value storage, PostgreSQL advisory locks
+for distributed locking, and a polling loop for the BLPOP pattern.
+"""
+
+import hashlib
+import struct
+import time
+import uuid
+from contextlib import AbstractContextManager
+from datetime import datetime
+from datetime import timedelta
+from datetime import timezone
+
+from sqlalchemy import delete
+from sqlalchemy import func
+from sqlalchemy import or_
+from sqlalchemy import select
+from sqlalchemy import update
+from sqlalchemy.dialects.postgresql import insert as pg_insert
+from sqlalchemy.orm import Session
+
+from onyx.cache.interface import CacheBackend
+from onyx.cache.interface import CacheLock
+from onyx.cache.interface import TTL_KEY_NOT_FOUND
+from onyx.cache.interface import TTL_NO_EXPIRY
+from onyx.db.models import CacheStore
+
+_LIST_KEY_PREFIX = "_q:"
+# ASCII: ':' (0x3A) < ';' (0x3B). Upper bound for range queries so [prefix+, prefix;)
+# captures all list-item keys (e.g. _q:mylist:123:uuid) without including other
+# lists whose names share a prefix (e.g. _q:mylist2:...).
+_LIST_KEY_RANGE_TERMINATOR = ";"
+_LIST_ITEM_TTL_SECONDS = 3600
+_LOCK_POLL_INTERVAL = 0.1
+_BLPOP_POLL_INTERVAL = 0.25
+
+
+def _list_item_key(key: str) -> str:
+    """Unique key for a list item. Timestamp for FIFO ordering; UUID prevents
+    collision when concurrent rpush calls occur within the same nanosecond.
+    """
+    return f"{_LIST_KEY_PREFIX}{key}:{time.time_ns()}:{uuid.uuid4().hex}"
+
+
+def _to_bytes(value: str | bytes | int | float) -> bytes:
+    if isinstance(value, bytes):
+        return value
+    return str(value).encode()
+
+
+# ------------------------------------------------------------------
+# Lock
+# ------------------------------------------------------------------
+
+
+class PostgresCacheLock(CacheLock):
+    """Advisory-lock-based distributed lock.
+
+    Uses ``get_session_with_tenant`` for connection lifecycle.  The lock is tied
+    to the session's connection; releasing or closing the session frees it.
+
+    NOTE: Unlike Redis locks, advisory locks do not auto-expire after
+    ``timeout`` seconds.  They are released when ``release()`` is
+    called or when the session is closed.
+    """
+
+    def __init__(self, lock_id: int, timeout: float | None, tenant_id: str) -> None:
+        self._lock_id = lock_id
+        self._timeout = timeout
+        self._tenant_id = tenant_id
+        self._session_cm: AbstractContextManager[Session] | None = None
+        self._session: Session | None = None
+        self._acquired = False
+
+    def acquire(
+        self,
+        blocking: bool = True,
+        blocking_timeout: float | None = None,
+    ) -> bool:
+        from onyx.db.engine.sql_engine import get_session_with_tenant
+
+        self._session_cm = get_session_with_tenant(tenant_id=self._tenant_id)
+        self._session = self._session_cm.__enter__()
+        try:
+            if not blocking:
+                return self._try_lock()
+
+            effective_timeout = blocking_timeout or self._timeout
+            deadline = (
+                (time.monotonic() + effective_timeout) if effective_timeout else None
+            )
+            while True:
+                if self._try_lock():
+                    return True
+                if deadline is not None and time.monotonic() >= deadline:
+                    return False
+                time.sleep(_LOCK_POLL_INTERVAL)
+        finally:
+            if not self._acquired:
+                self._close_session()
+
+    def release(self) -> None:
+        if not self._acquired or self._session is None:
+            return
+        try:
+            self._session.execute(select(func.pg_advisory_unlock(self._lock_id)))
+        finally:
+            self._acquired = False
+            self._close_session()
+
+    def owned(self) -> bool:
+        return self._acquired
+
+    def _close_session(self) -> None:
+        if self._session_cm is not None:
+            try:
+                self._session_cm.__exit__(None, None, None)
+            finally:
+                self._session_cm = None
+                self._session = None
+
+    def _try_lock(self) -> bool:
+        assert self._session is not None
+        result = self._session.execute(
+            select(func.pg_try_advisory_lock(self._lock_id))
+        ).scalar()
+        if result:
+            self._acquired = True
+            return True
+        return False
+
+
+# ------------------------------------------------------------------
+# Backend
+# ------------------------------------------------------------------
+
+
+class PostgresCacheBackend(CacheBackend):
+    """``CacheBackend`` backed by the ``cache_store`` table in PostgreSQL.
+
+    Each operation opens and closes its own database session so the backend
+    is safe to share across threads.  Tenant isolation is handled by
+    SQLAlchemy's ``schema_translate_map`` (set by ``get_session_with_tenant``).
+    """
+
+    def __init__(self, tenant_id: str) -> None:
+        self._tenant_id = tenant_id
+
+    # -- basic key/value ---------------------------------------------------
+
+    def get(self, key: str) -> bytes | None:
+        from onyx.db.engine.sql_engine import get_session_with_tenant
+
+        stmt = select(CacheStore.value).where(
+            CacheStore.key == key,
+            or_(CacheStore.expires_at.is_(None), CacheStore.expires_at > func.now()),
+        )
+        with get_session_with_tenant(tenant_id=self._tenant_id) as session:
+            value = session.execute(stmt).scalar_one_or_none()
+        if value is None:
+            return None
+        return bytes(value)
+
+    def set(
+        self,
+        key: str,
+        value: str | bytes | int | float,
+        ex: int | None = None,
+    ) -> None:
+        from onyx.db.engine.sql_engine import get_session_with_tenant
+
+        value_bytes = _to_bytes(value)
+        expires_at = (
+            datetime.now(timezone.utc) + timedelta(seconds=ex)
+            if ex is not None
+            else None
+        )
+        stmt = (
+            pg_insert(CacheStore)
+            .values(key=key, value=value_bytes, expires_at=expires_at)
+            .on_conflict_do_update(
+                index_elements=[CacheStore.key],
+                set_={"value": value_bytes, "expires_at": expires_at},
+            )
+        )
+        with get_session_with_tenant(tenant_id=self._tenant_id) as session:
+            session.execute(stmt)
+            session.commit()
+
+    def delete(self, key: str) -> None:
+        from onyx.db.engine.sql_engine import get_session_with_tenant
+
+        with get_session_with_tenant(tenant_id=self._tenant_id) as session:
+            session.execute(delete(CacheStore).where(CacheStore.key == key))
+            session.commit()
+
+    def exists(self, key: str) -> bool:
+        from onyx.db.engine.sql_engine import get_session_with_tenant
+
+        stmt = (
+            select(CacheStore.key)
+            .where(
+                CacheStore.key == key,
+                or_(
+                    CacheStore.expires_at.is_(None),
+                    CacheStore.expires_at > func.now(),
+                ),
+            )
+            .limit(1)
+        )
+        with get_session_with_tenant(tenant_id=self._tenant_id) as session:
+            return session.execute(stmt).first() is not None
+
+    # -- TTL ---------------------------------------------------------------
+
+    def expire(self, key: str, seconds: int) -> None:
+        from onyx.db.engine.sql_engine import get_session_with_tenant
+
+        new_exp = datetime.now(timezone.utc) + timedelta(seconds=seconds)
+        stmt = (
+            update(CacheStore).where(CacheStore.key == key).values(expires_at=new_exp)
+        )
+        with get_session_with_tenant(tenant_id=self._tenant_id) as session:
+            session.execute(stmt)
+            session.commit()
+
+    def ttl(self, key: str) -> int:
+        from onyx.db.engine.sql_engine import get_session_with_tenant
+
+        stmt = select(CacheStore.expires_at).where(CacheStore.key == key)
+        with get_session_with_tenant(tenant_id=self._tenant_id) as session:
+            result = session.execute(stmt).first()
+        if result is None:
+            return TTL_KEY_NOT_FOUND
+        expires_at: datetime | None = result[0]
+        if expires_at is None:
+            return TTL_NO_EXPIRY
+        remaining = (expires_at - datetime.now(timezone.utc)).total_seconds()
+        if remaining <= 0:
+            return TTL_KEY_NOT_FOUND
+        return int(remaining)
+
+    # -- distributed lock --------------------------------------------------
+
+    def lock(self, name: str, timeout: float | None = None) -> CacheLock:
+        return PostgresCacheLock(
+            self._lock_id_for(name), timeout, tenant_id=self._tenant_id
+        )
+
+    # -- blocking list (MCP OAuth BLPOP pattern) ---------------------------
+
+    def rpush(self, key: str, value: str | bytes) -> None:
+        self.set(_list_item_key(key), value, ex=_LIST_ITEM_TTL_SECONDS)
+
+    def blpop(self, keys: list[str], timeout: int = 0) -> tuple[bytes, bytes] | None:
+        if timeout <= 0:
+            raise ValueError(
+                "PostgresCacheBackend.blpop requires timeout > 0. "
+                "timeout=0 would block the calling thread indefinitely "
+                "with no way to interrupt short of process termination."
+            )
+        from onyx.db.engine.sql_engine import get_session_with_tenant
+
+        deadline = time.monotonic() + timeout
+        while True:
+            for key in keys:
+                lower = f"{_LIST_KEY_PREFIX}{key}:"
+                upper = f"{_LIST_KEY_PREFIX}{key}{_LIST_KEY_RANGE_TERMINATOR}"
+                stmt = (
+                    select(CacheStore)
+                    .where(
+                        CacheStore.key >= lower,
+                        CacheStore.key < upper,
+                        or_(
+                            CacheStore.expires_at.is_(None),
+                            CacheStore.expires_at > func.now(),
+                        ),
+                    )
+                    .order_by(CacheStore.key)
+                    .limit(1)
+                    .with_for_update(skip_locked=True)
+                )
+                with get_session_with_tenant(tenant_id=self._tenant_id) as session:
+                    row = session.execute(stmt).scalars().first()
+                    if row is not None:
+                        value = bytes(row.value) if row.value else b""
+                        session.delete(row)
+                        session.commit()
+                        return (key.encode(), value)
+            if time.monotonic() >= deadline:
+                return None
+            time.sleep(_BLPOP_POLL_INTERVAL)
+
+    # -- helpers -----------------------------------------------------------
+
+    def _lock_id_for(self, name: str) -> int:
+        """Map *name* to a 64-bit signed int for ``pg_advisory_lock``."""
+        h = hashlib.md5(f"{self._tenant_id}:{name}".encode()).digest()
+        return struct.unpack("q", h[:8])[0]
+
+
+# ------------------------------------------------------------------
+# Periodic cleanup
+# ------------------------------------------------------------------
+
+
+def cleanup_expired_cache_entries() -> None:
+    """Delete rows whose ``expires_at`` is in the past.
+
+    Called by the periodic poller every 5 minutes.
+    """
+    from onyx.db.engine.sql_engine import get_session_with_current_tenant
+
+    with get_session_with_current_tenant() as session:
+        session.execute(
+            delete(CacheStore).where(
+                CacheStore.expires_at.is_not(None),
+                CacheStore.expires_at < func.now(),
+            )
+        )
+        session.commit()

backend/onyx/chat/chat_processing_checker.py

@@ -1,57 +1,52 @@
 from uuid import UUID
 
-from redis.client import Redis
+from onyx.cache.interface import CacheBackend
 
-# Redis key prefixes for chat message processing
 PREFIX = "chatprocessing"
 FENCE_PREFIX = f"{PREFIX}_fence"
 FENCE_TTL = 30 * 60  # 30 minutes
 
 
 def _get_fence_key(chat_session_id: UUID) -> str:
-    """
-    Generate the Redis key for a chat session processing a message.
+    """Generate the cache key for a chat session processing fence.
 
     Args:
         chat_session_id: The UUID of the chat session
 
     Returns:
-        The fence key string (tenant_id is automatically added by the Redis client)
+        The fence key string. Tenant isolation is handled automatically
+        by the cache backend (Redis key-prefixing or Postgres schema routing).
     """
     return f"{FENCE_PREFIX}_{chat_session_id}"
 
 
 def set_processing_status(
-    chat_session_id: UUID, redis_client: Redis, value: bool
+    chat_session_id: UUID, cache: CacheBackend, value: bool
 ) -> None:
-    """
-    Set or clear the fence for a chat session processing a message.
+    """Set or clear the fence for a chat session processing a message.
 
-    If the key exists, we are processing a message. If the key does not exist, we are not processing a message.
+    If the key exists, a message is being processed.
 
     Args:
         chat_session_id: The UUID of the chat session
-        redis_client: The Redis client to use
+        cache: Tenant-aware cache backend
         value: True to set the fence, False to clear it
     """
     fence_key = _get_fence_key(chat_session_id)
-
     if value:
-        redis_client.set(fence_key, 0, ex=FENCE_TTL)
+        cache.set(fence_key, 0, ex=FENCE_TTL)
     else:
-        redis_client.delete(fence_key)
+        cache.delete(fence_key)
 
 
-def is_chat_session_processing(chat_session_id: UUID, redis_client: Redis) -> bool:
-    """
-    Check if the chat session is processing a message.
+def is_chat_session_processing(chat_session_id: UUID, cache: CacheBackend) -> bool:
+    """Check if the chat session is processing a message.
 
     Args:
         chat_session_id: The UUID of the chat session
-        redis_client: The Redis client to use
+        cache: Tenant-aware cache backend
 
     Returns:
         True if the chat session is processing a message, False otherwise
     """
-    fence_key = _get_fence_key(chat_session_id)
-    return bool(redis_client.exists(fence_key))
+    return cache.exists(_get_fence_key(chat_session_id))

backend/onyx/chat/llm_loop.py

@@ -36,7 +36,6 @@ from onyx.db.memory import add_memory
 from onyx.db.memory import update_memory_at_index
 from onyx.db.memory import UserMemoryContext
 from onyx.db.models import Persona
-from onyx.llm.constants import LlmProviderNames
 from onyx.llm.interfaces import LLM
 from onyx.llm.interfaces import LLMUserIdentity
 from onyx.llm.interfaces import ToolChoiceOptions
@@ -52,6 +51,7 @@ from onyx.tools.built_in_tools import STOPPING_TOOLS_NAMES
 from onyx.tools.interface import Tool
 from onyx.tools.models import ChatFile
 from onyx.tools.models import MemoryToolResponseSnapshot
+from onyx.tools.models import PythonToolRichResponse
 from onyx.tools.models import ToolCallInfo
 from onyx.tools.models import ToolCallKickoff
 from onyx.tools.models import ToolResponse
@@ -83,28 +83,6 @@ def _looks_like_xml_tool_call_payload(text: str | None) -> bool:
     )
 
 
-def _should_keep_bedrock_tool_definitions(
-    llm: object, simple_chat_history: list[ChatMessageSimple]
-) -> bool:
-    """Bedrock requires tool config when history includes toolUse/toolResult blocks."""
-    model_provider = getattr(getattr(llm, "config", None), "model_provider", None)
-    if model_provider not in {
-        LlmProviderNames.BEDROCK,
-        LlmProviderNames.BEDROCK_CONVERSE,
-    }:
-        return False
-
-    return any(
-        (
-            msg.message_type == MessageType.ASSISTANT
-            and msg.tool_calls
-            and len(msg.tool_calls) > 0
-        )
-        or msg.message_type == MessageType.TOOL_CALL_RESPONSE
-        for msg in simple_chat_history
-    )
-
-
 def _try_fallback_tool_extraction(
     llm_step_result: LlmStepResult,
     tool_choice: ToolChoiceOptions,
@@ -685,12 +663,7 @@ def run_llm_loop(
             elif out_of_cycles or ran_image_gen:
                 # Last cycle, no tools allowed, just answer!
                 tool_choice = ToolChoiceOptions.NONE
-                # Bedrock requires tool config in requests that include toolUse/toolResult history.
-                final_tools = (
-                    tools
-                    if _should_keep_bedrock_tool_definitions(llm, simple_chat_history)
-                    else []
-                )
+                final_tools = []
             else:
                 tool_choice = ToolChoiceOptions.AUTO
                 final_tools = tools
@@ -966,6 +939,13 @@ def run_llm_loop(
                 ):
                     generated_images = tool_response.rich_response.generated_images
 
+                # Extract generated_files if this is a code interpreter response
+                generated_files = None
+                if isinstance(tool_response.rich_response, PythonToolRichResponse):
+                    generated_files = (
+                        tool_response.rich_response.generated_files or None
+                    )
+
                 # Persist memory if this is a memory tool response
                 memory_snapshot: MemoryToolResponseSnapshot | None = None
                 if isinstance(tool_response.rich_response, MemoryToolResponse):
@@ -1017,6 +997,7 @@ def run_llm_loop(
                     tool_call_response=saved_response,
                     search_docs=displayed_docs or search_docs,
                     generated_images=generated_images,
+                    generated_files=generated_files,
                 )
                 # Add to state container for partial save support
                 state_container.add_tool_call(tool_call_info)

backend/onyx/chat/llm_step.py

@@ -15,6 +15,7 @@ from onyx.chat.citation_processor import DynamicCitationProcessor
 from onyx.chat.emitter import Emitter
 from onyx.chat.models import ChatMessageSimple
 from onyx.chat.models import LlmStepResult
+from onyx.chat.tool_call_args_streaming import maybe_emit_argument_delta
 from onyx.configs.app_configs import LOG_ONYX_MODEL_INTERACTIONS
 from onyx.configs.app_configs import PROMPT_CACHE_CHAT_HISTORY
 from onyx.configs.constants import MessageType
@@ -54,7 +55,9 @@ from onyx.server.query_and_chat.streaming_models import ReasoningStart
 from onyx.tools.models import ToolCallKickoff
 from onyx.tracing.framework.create import generation_span
 from onyx.utils.b64 import get_image_type_from_bytes
+from onyx.utils.jsonriver import Parser
 from onyx.utils.logger import setup_logger
+from onyx.utils.postgres_sanitization import sanitize_string
 from onyx.utils.text_processing import find_all_json_objects
 
 logger = setup_logger()
@@ -166,15 +169,6 @@ def _find_function_calls_open_marker(text_lower: str) -> int:
         search_from = idx + 1
 
 
-def _sanitize_llm_output(value: str) -> str:
-    """Remove characters that PostgreSQL's text/JSONB types cannot store.
-
-    - NULL bytes (\x00): Not allowed in PostgreSQL text types
-    - UTF-16 surrogates (\ud800-\udfff): Invalid in UTF-8 encoding
-    """
-    return "".join(c for c in value if c != "\x00" and not ("\ud800" <= c <= "\udfff"))
-
-
 def _try_parse_json_string(value: Any) -> Any:
     """Attempt to parse a JSON string value into its Python equivalent.
 
@@ -222,9 +216,7 @@ def _parse_tool_args_to_dict(raw_args: Any) -> dict[str, Any]:
     if isinstance(raw_args, dict):
         # Parse any string values that look like JSON arrays/objects
         return {
-            k: _try_parse_json_string(
-                _sanitize_llm_output(v) if isinstance(v, str) else v
-            )
+            k: _try_parse_json_string(sanitize_string(v) if isinstance(v, str) else v)
             for k, v in raw_args.items()
         }
 
@@ -232,7 +224,7 @@ def _parse_tool_args_to_dict(raw_args: Any) -> dict[str, Any]:
         return {}
 
     # Sanitize before parsing to remove NULL bytes and surrogates
-    raw_args = _sanitize_llm_output(raw_args)
+    raw_args = sanitize_string(raw_args)
 
     try:
         parsed1: Any = json.loads(raw_args)
@@ -545,12 +537,12 @@ def _extract_xml_attribute(attrs: str, attr_name: str) -> str | None:
     )
     if not attr_match:
         return None
-    return _sanitize_llm_output(unescape(attr_match.group(2).strip()))
+    return sanitize_string(unescape(attr_match.group(2).strip()))
 
 
 def _parse_xml_parameter_value(raw_value: str, string_attr: str | None) -> Any:
     """Parse a parameter value from XML-style tool call payloads."""
-    value = _sanitize_llm_output(unescape(raw_value).strip())
+    value = sanitize_string(unescape(raw_value).strip())
 
     if string_attr and string_attr.lower() == "true":
         return value
@@ -569,6 +561,7 @@ def _resolve_tool_arguments(obj: dict[str, Any]) -> dict[str, Any] | None:
     """
     arguments = obj.get("arguments", obj.get("parameters", {}))
     if isinstance(arguments, str):
+        arguments = sanitize_string(arguments)
         try:
             arguments = json.loads(arguments)
         except json.JSONDecodeError:
@@ -1018,6 +1011,7 @@ def run_llm_step_pkt_generator(
         )
 
     id_to_tool_call_map: dict[int, dict[str, Any]] = {}
+    arg_parsers: dict[int, Parser] = {}
     reasoning_start = False
     answer_start = False
     accumulated_reasoning = ""
@@ -1224,7 +1218,14 @@ def run_llm_step_pkt_generator(
                 yield from _close_reasoning_if_active()
 
                 for tool_call_delta in delta.tool_calls:
+                    # maybe_emit depends and update being called first and attaching the delta
                     _update_tool_call_with_delta(id_to_tool_call_map, tool_call_delta)
+                    yield from maybe_emit_argument_delta(
+                        tool_calls_in_progress=id_to_tool_call_map,
+                        tool_call_delta=tool_call_delta,
+                        placement=_current_placement(),
+                        parsers=arg_parsers,
+                    )
 
         # Flush any tail text buffered while checking for split "<function_calls" markers.
         filtered_content_tail = xml_tool_call_content_filter.flush()

backend/onyx/chat/process_message.py

@@ -11,9 +11,10 @@ from contextvars import Token
 from uuid import UUID
 
 from pydantic import BaseModel
-from redis.client import Redis
 from sqlalchemy.orm import Session
 
+from onyx.cache.factory import get_cache_backend
+from onyx.cache.interface import CacheBackend
 from onyx.chat.chat_processing_checker import set_processing_status
 from onyx.chat.chat_state import ChatStateContainer
 from onyx.chat.chat_state import run_chat_loop_with_state_containers
@@ -79,7 +80,6 @@ from onyx.llm.request_context import reset_llm_mock_response
 from onyx.llm.request_context import set_llm_mock_response
 from onyx.llm.utils import litellm_exception_to_error_msg
 from onyx.onyxbot.slack.models import SlackContext
-from onyx.redis.redis_pool import get_redis_client
 from onyx.server.query_and_chat.models import AUTO_PLACE_AFTER_LATEST_MESSAGE
 from onyx.server.query_and_chat.models import MessageResponseIDInfo
 from onyx.server.query_and_chat.models import SendMessageRequest
@@ -448,7 +448,7 @@ def handle_stream_message_objects(
 
     llm: LLM | None = None
     chat_session: ChatSession | None = None
-    redis_client: Redis | None = None
+    cache: CacheBackend | None = None
 
     user_id = user.id
     if user.is_anonymous:
@@ -809,19 +809,19 @@ def handle_stream_message_objects(
             )
             simple_chat_history.insert(0, summary_simple)
 
-        redis_client = get_redis_client()
+        cache = get_cache_backend()
 
         reset_cancel_status(
             chat_session.id,
-            redis_client,
+            cache,
         )
 
         def check_is_connected() -> bool:
-            return check_stop_signal(chat_session.id, redis_client)
+            return check_stop_signal(chat_session.id, cache)
 
         set_processing_status(
             chat_session_id=chat_session.id,
-            redis_client=redis_client,
+            cache=cache,
             value=True,
         )
 
@@ -968,10 +968,10 @@ def handle_stream_message_objects(
             reset_llm_mock_response(mock_response_token)
 
         try:
-            if redis_client is not None and chat_session is not None:
+            if cache is not None and chat_session is not None:
                 set_processing_status(
                     chat_session_id=chat_session.id,
-                    redis_client=redis_client,
+                    cache=cache,
                     value=False,
                 )
         except Exception:

backend/onyx/chat/save_chat.py

@@ -1,4 +1,5 @@
 import json
+import mimetypes
 
 from sqlalchemy.orm import Session
 
@@ -12,14 +13,42 @@ from onyx.db.chat import create_db_search_doc
 from onyx.db.models import ChatMessage
 from onyx.db.models import ToolCall
 from onyx.db.tools import create_tool_call_no_commit
+from onyx.file_store.models import FileDescriptor
 from onyx.natural_language_processing.utils import BaseTokenizer
 from onyx.natural_language_processing.utils import get_tokenizer
+from onyx.server.query_and_chat.chat_utils import mime_type_to_chat_file_type
 from onyx.tools.models import ToolCallInfo
 from onyx.utils.logger import setup_logger
+from onyx.utils.postgres_sanitization import sanitize_string
 
 logger = setup_logger()
 
 
+def _extract_referenced_file_descriptors(
+    tool_calls: list[ToolCallInfo],
+    message_text: str,
+) -> list[FileDescriptor]:
+    """Extract FileDescriptors for code interpreter files referenced in the message text."""
+    descriptors: list[FileDescriptor] = []
+    for tool_call_info in tool_calls:
+        if not tool_call_info.generated_files:
+            continue
+        for gen_file in tool_call_info.generated_files:
+            file_id = (
+                gen_file.file_link.rsplit("/", 1)[-1] if gen_file.file_link else ""
+            )
+            if file_id and file_id in message_text:
+                mime_type, _ = mimetypes.guess_type(gen_file.filename)
+                descriptors.append(
+                    FileDescriptor(
+                        id=file_id,
+                        type=mime_type_to_chat_file_type(mime_type),
+                        name=gen_file.filename,
+                    )
+                )
+    return descriptors
+
+
 def _create_and_link_tool_calls(
     tool_calls: list[ToolCallInfo],
     assistant_message: ChatMessage,
@@ -173,8 +202,13 @@ def save_chat_turn(
         pre_answer_processing_time: Duration of processing before answer starts (in seconds)
     """
     # 1. Update ChatMessage with message content, reasoning tokens, and token count
-    assistant_message.message = message_text
-    assistant_message.reasoning_tokens = reasoning_tokens
+    sanitized_message_text = (
+        sanitize_string(message_text) if message_text else message_text
+    )
+    assistant_message.message = sanitized_message_text
+    assistant_message.reasoning_tokens = (
+        sanitize_string(reasoning_tokens) if reasoning_tokens else reasoning_tokens
+    )
     assistant_message.is_clarification = is_clarification
 
     # Use pre-answer processing time (captured when MESSAGE_START was emitted)
@@ -184,8 +218,10 @@ def save_chat_turn(
     # Calculate token count using default tokenizer, when storing, this should not use the LLM
     # specific one so we use a system default tokenizer here.
     default_tokenizer = get_tokenizer(None, None)
-    if message_text:
-        assistant_message.token_count = len(default_tokenizer.encode(message_text))
+    if sanitized_message_text:
+        assistant_message.token_count = len(
+            default_tokenizer.encode(sanitized_message_text)
+        )
     else:
         assistant_message.token_count = 0
 
@@ -297,5 +333,16 @@ def save_chat_turn(
         citation_number_to_search_doc_id if citation_number_to_search_doc_id else None
     )
 
+    # 8. Attach code interpreter generated files that the assistant actually
+    # referenced in its response, so they are available via load_all_chat_files
+    # on subsequent turns. Files not mentioned are intermediate artifacts.
+    if sanitized_message_text:
+        referenced = _extract_referenced_file_descriptors(
+            tool_calls, sanitized_message_text
+        )
+        if referenced:
+            existing_files = assistant_message.files or []
+            assistant_message.files = existing_files + referenced
+
     # Finally save the messages, tool calls, and docs
     db_session.commit()

backend/onyx/chat/stop_signal_checker.py

@@ -1,65 +1,58 @@
 from uuid import UUID
 
-from redis.client import Redis
+from onyx.cache.interface import CacheBackend
 
-# Redis key prefixes for chat session stop signals
 PREFIX = "chatsessionstop"
 FENCE_PREFIX = f"{PREFIX}_fence"
-FENCE_TTL = 10 * 60  # 10 minutes - defensive TTL to prevent memory leaks
+FENCE_TTL = 10 * 60  # 10 minutes
 
 
 def _get_fence_key(chat_session_id: UUID) -> str:
-    """
-    Generate the Redis key for a chat session stop signal fence.
+    """Generate the cache key for a chat session stop signal fence.
 
     Args:
         chat_session_id: The UUID of the chat session
 
     Returns:
-        The fence key string (tenant_id is automatically added by the Redis client)
+        The fence key string. Tenant isolation is handled automatically
+        by the cache backend (Redis key-prefixing or Postgres schema routing).
     """
     return f"{FENCE_PREFIX}_{chat_session_id}"
 
 
-def set_fence(chat_session_id: UUID, redis_client: Redis, value: bool) -> None:
-    """
-    Set or clear the stop signal fence for a chat session.
+def set_fence(chat_session_id: UUID, cache: CacheBackend, value: bool) -> None:
+    """Set or clear the stop signal fence for a chat session.
 
     Args:
         chat_session_id: The UUID of the chat session
-        redis_client: Redis client to use (tenant-aware client that auto-prefixes keys)
+        cache: Tenant-aware cache backend
         value: True to set the fence (stop signal), False to clear it
     """
     fence_key = _get_fence_key(chat_session_id)
     if not value:
-        redis_client.delete(fence_key)
+        cache.delete(fence_key)
         return
-
-    redis_client.set(fence_key, 0, ex=FENCE_TTL)
+    cache.set(fence_key, 0, ex=FENCE_TTL)
 
 
-def is_connected(chat_session_id: UUID, redis_client: Redis) -> bool:
-    """
-    Check if the chat session should continue (not stopped).
+def is_connected(chat_session_id: UUID, cache: CacheBackend) -> bool:
+    """Check if the chat session should continue (not stopped).
 
     Args:
         chat_session_id: The UUID of the chat session to check
-        redis_client: Redis client to use for checking the stop signal (tenant-aware client that auto-prefixes keys)
+        cache: Tenant-aware cache backend
 
     Returns:
         True if the session should continue, False if it should stop
     """
-    fence_key = _get_fence_key(chat_session_id)
-    return not bool(redis_client.exists(fence_key))
+    return not cache.exists(_get_fence_key(chat_session_id))
 
 
-def reset_cancel_status(chat_session_id: UUID, redis_client: Redis) -> None:
-    """
-    Clear the stop signal for a chat session.
+def reset_cancel_status(chat_session_id: UUID, cache: CacheBackend) -> None:
+    """Clear the stop signal for a chat session.
 
     Args:
         chat_session_id: The UUID of the chat session
-        redis_client: Redis client to use (tenant-aware client that auto-prefixes keys)
+        cache: Tenant-aware cache backend
     """
-    fence_key = _get_fence_key(chat_session_id)
-    redis_client.delete(fence_key)
+    cache.delete(_get_fence_key(chat_session_id))

backend/onyx/chat/tool_call_args_streaming.py

@@ -0,0 +1,77 @@
+from collections.abc import Generator
+from collections.abc import Mapping
+from typing import Any
+from typing import Type
+
+from onyx.llm.model_response import ChatCompletionDeltaToolCall
+from onyx.server.query_and_chat.placement import Placement
+from onyx.server.query_and_chat.streaming_models import Packet
+from onyx.server.query_and_chat.streaming_models import ToolCallArgumentDelta
+from onyx.tools.built_in_tools import TOOL_NAME_TO_CLASS
+from onyx.tools.interface import Tool
+from onyx.utils.jsonriver import Parser
+
+
+def _get_tool_class(
+    tool_calls_in_progress: Mapping[int, Mapping[str, Any]],
+    tool_call_delta: ChatCompletionDeltaToolCall,
+) -> Type[Tool] | None:
+    """Look up the Tool subclass for a streaming tool call delta."""
+    tool_name = tool_calls_in_progress.get(tool_call_delta.index, {}).get("name")
+    if not tool_name:
+        return None
+    return TOOL_NAME_TO_CLASS.get(tool_name)
+
+
+def maybe_emit_argument_delta(
+    tool_calls_in_progress: Mapping[int, Mapping[str, Any]],
+    tool_call_delta: ChatCompletionDeltaToolCall,
+    placement: Placement,
+    parsers: dict[int, Parser],
+) -> Generator[Packet, None, None]:
+    """Emit decoded tool-call argument deltas to the frontend.
+
+    Uses a ``jsonriver.Parser`` per tool-call index to incrementally parse
+    the JSON argument string and extract only the newly-appended content
+    for each string-valued argument.
+
+    NOTE: Non-string arguments (numbers, booleans, null, arrays, objects)
+    are skipped — they are available in the final tool-call kickoff packet.
+
+    ``parsers`` is a mutable dict keyed by tool-call index. A new
+    ``Parser`` is created automatically for each new index.
+    """
+    tool_cls = _get_tool_class(tool_calls_in_progress, tool_call_delta)
+    if not tool_cls or not tool_cls.should_emit_argument_deltas():
+        return
+
+    fn = tool_call_delta.function
+    delta_fragment = fn.arguments if fn else None
+    if not delta_fragment:
+        return
+
+    idx = tool_call_delta.index
+    if idx not in parsers:
+        parsers[idx] = Parser()
+    parser = parsers[idx]
+
+    deltas = parser.feed(delta_fragment)
+
+    argument_deltas: dict[str, str] = {}
+    for delta in deltas:
+        if isinstance(delta, dict):
+            for key, value in delta.items():
+                if isinstance(value, str):
+                    argument_deltas[key] = argument_deltas.get(key, "") + value
+
+    if not argument_deltas:
+        return
+
+    tc_data = tool_calls_in_progress[tool_call_delta.index]
+    yield Packet(
+        placement=placement,
+        obj=ToolCallArgumentDelta(
+            tool_type=tc_data.get("name", ""),
+            argument_deltas=argument_deltas,
+        ),
+    )

backend/onyx/configs/app_configs.py

@@ -495,14 +495,7 @@ CELERY_WORKER_PRIMARY_POOL_OVERFLOW = int(
     os.environ.get("CELERY_WORKER_PRIMARY_POOL_OVERFLOW") or 4
 )
 
-# Consolidated background worker (light, docprocessing, docfetching, heavy, monitoring, user_file_processing)
-# separate workers' defaults: light=24, docprocessing=6, docfetching=1, heavy=4, kg=2, monitoring=1, user_file=2
-# Total would be 40, but we use a more conservative default of 20 for the consolidated worker
-CELERY_WORKER_BACKGROUND_CONCURRENCY = int(
-    os.environ.get("CELERY_WORKER_BACKGROUND_CONCURRENCY") or 20
-)
-
-# Individual worker concurrency settings (used when USE_LIGHTWEIGHT_BACKGROUND_WORKER is False or on Kuberenetes deployments)
+# Individual worker concurrency settings
 CELERY_WORKER_HEAVY_CONCURRENCY = int(
     os.environ.get("CELERY_WORKER_HEAVY_CONCURRENCY") or 4
 )
@@ -819,7 +812,9 @@ RERANK_COUNT = int(os.environ.get("RERANK_COUNT") or 1000)
 # Tool Configs
 #####
 # Code Interpreter Service Configuration
-CODE_INTERPRETER_BASE_URL = os.environ.get("CODE_INTERPRETER_BASE_URL")
+CODE_INTERPRETER_BASE_URL = os.environ.get(
+    "CODE_INTERPRETER_BASE_URL", "http://localhost:8000"
+)
 
 CODE_INTERPRETER_DEFAULT_TIMEOUT_MS = int(
     os.environ.get("CODE_INTERPRETER_DEFAULT_TIMEOUT_MS") or 60_000
@@ -900,6 +895,9 @@ CUSTOM_ANSWER_VALIDITY_CONDITIONS = json.loads(
 )
 
 VESPA_REQUEST_TIMEOUT = int(os.environ.get("VESPA_REQUEST_TIMEOUT") or "15")
+VESPA_MIGRATION_REQUEST_TIMEOUT_S = int(
+    os.environ.get("VESPA_MIGRATION_REQUEST_TIMEOUT_S") or "120"
+)
 
 SYSTEM_RECURSION_LIMIT = int(os.environ.get("SYSTEM_RECURSION_LIMIT") or "1000")
 

backend/onyx/configs/constants.py

@@ -84,7 +84,6 @@ POSTGRES_CELERY_WORKER_LIGHT_APP_NAME = "celery_worker_light"
 POSTGRES_CELERY_WORKER_DOCPROCESSING_APP_NAME = "celery_worker_docprocessing"
 POSTGRES_CELERY_WORKER_DOCFETCHING_APP_NAME = "celery_worker_docfetching"
 POSTGRES_CELERY_WORKER_INDEXING_CHILD_APP_NAME = "celery_worker_indexing_child"
-POSTGRES_CELERY_WORKER_BACKGROUND_APP_NAME = "celery_worker_background"
 POSTGRES_CELERY_WORKER_HEAVY_APP_NAME = "celery_worker_heavy"
 POSTGRES_CELERY_WORKER_MONITORING_APP_NAME = "celery_worker_monitoring"
 POSTGRES_CELERY_WORKER_USER_FILE_PROCESSING_APP_NAME = (

backend/onyx/connectors/confluence/connector.py

@@ -943,6 +943,9 @@ class ConfluenceConnector(
                         if include_permissions
                         else None
                     ),
+                    parent_hierarchy_raw_node_id=self._get_parent_hierarchy_raw_id(
+                        page
+                    ),
                 )
             )
 
@@ -992,6 +995,7 @@ class ConfluenceConnector(
                             if include_permissions
                             else None
                         ),
+                        parent_hierarchy_raw_node_id=page_id,
                     )
                 )
 

backend/onyx/connectors/google_drive/doc_conversion.py

@@ -781,4 +781,5 @@ def build_slim_document(
     return SlimDocument(
         id=onyx_document_id_from_drive_file(file),
         external_access=external_access,
+        parent_hierarchy_raw_node_id=(file.get("parents") or [None])[0],
     )

backend/onyx/connectors/jira/connector.py

@@ -902,6 +902,11 @@ class JiraConnector(
                         external_access=self._get_project_permissions(
                             project_key, add_prefix=False
                         ),
+                        parent_hierarchy_raw_node_id=(
+                            self._get_parent_hierarchy_raw_node_id(issue, project_key)
+                            if project_key
+                            else None
+                        ),
                     )
                 )
                 current_offset += 1

backend/onyx/connectors/models.py

@@ -385,6 +385,7 @@ class IndexingDocument(Document):
 class SlimDocument(BaseModel):
     id: str
     external_access: ExternalAccess | None = None
+    parent_hierarchy_raw_node_id: str | None = None
 
 
 class HierarchyNode(BaseModel):

backend/onyx/connectors/sharepoint/connector.py

@@ -772,6 +772,7 @@ def _convert_driveitem_to_slim_document(
     drive_name: str,
     ctx: ClientContext,
     graph_client: GraphClient,
+    parent_hierarchy_raw_node_id: str | None = None,
 ) -> SlimDocument:
     if driveitem.id is None:
         raise ValueError("DriveItem ID is required")
@@ -787,11 +788,15 @@ def _convert_driveitem_to_slim_document(
     return SlimDocument(
         id=driveitem.id,
         external_access=external_access,
+        parent_hierarchy_raw_node_id=parent_hierarchy_raw_node_id,
     )
 
 
 def _convert_sitepage_to_slim_document(
-    site_page: dict[str, Any], ctx: ClientContext | None, graph_client: GraphClient
+    site_page: dict[str, Any],
+    ctx: ClientContext | None,
+    graph_client: GraphClient,
+    parent_hierarchy_raw_node_id: str | None = None,
 ) -> SlimDocument:
     """Convert a SharePoint site page to a SlimDocument object."""
     if site_page.get("id") is None:
@@ -808,6 +813,7 @@ def _convert_sitepage_to_slim_document(
     return SlimDocument(
         id=id,
         external_access=external_access,
+        parent_hierarchy_raw_node_id=parent_hierarchy_raw_node_id,
     )
 
 
@@ -1594,12 +1600,22 @@ class SharepointConnector(
                             )
                         )
 
+                    parent_hierarchy_url: str | None = None
+                    if drive_web_url:
+                        parent_hierarchy_url = self._get_parent_hierarchy_url(
+                            site_url, drive_web_url, drive_name, driveitem
+                        )
+
                     try:
                         logger.debug(f"Processing: {driveitem.web_url}")
                         ctx = self._create_rest_client_context(site_descriptor.url)
                         doc_batch.append(
                             _convert_driveitem_to_slim_document(
-                                driveitem, drive_name, ctx, self.graph_client
+                                driveitem,
+                                drive_name,
+                                ctx,
+                                self.graph_client,
+                                parent_hierarchy_raw_node_id=parent_hierarchy_url,
                             )
                         )
                     except Exception as e:
@@ -1619,7 +1635,10 @@ class SharepointConnector(
                     ctx = self._create_rest_client_context(site_descriptor.url)
                     doc_batch.append(
                         _convert_sitepage_to_slim_document(
-                            site_page, ctx, self.graph_client
+                            site_page,
+                            ctx,
+                            self.graph_client,
+                            parent_hierarchy_raw_node_id=site_descriptor.url,
                         )
                     )
                     if len(doc_batch) >= SLIM_BATCH_SIZE:

backend/onyx/connectors/slack/connector.py

@@ -565,6 +565,7 @@ def _get_all_doc_ids(
                             channel_id=channel_id, thread_ts=message["ts"]
                         ),
                         external_access=external_access,
+                        parent_hierarchy_raw_node_id=channel_id,
                     )
                 )
 

backend/onyx/db/chat.py

@@ -38,6 +38,7 @@ from onyx.llm.override_models import LLMOverride
 from onyx.llm.override_models import PromptOverride
 from onyx.server.query_and_chat.models import ChatMessageDetail
 from onyx.utils.logger import setup_logger
+from onyx.utils.postgres_sanitization import sanitize_string
 
 
 logger = setup_logger()
@@ -675,58 +676,43 @@ def set_as_latest_chat_message(
     db_session.commit()
 
 
-def _sanitize_for_postgres(value: str) -> str:
-    """Remove NUL (0x00) characters from strings as PostgreSQL doesn't allow them."""
-    sanitized = value.replace("\x00", "")
-    if value and not sanitized:
-        logger.warning("Sanitization removed all characters from string")
-    return sanitized
-
-
-def _sanitize_list_for_postgres(values: list[str]) -> list[str]:
-    """Remove NUL (0x00) characters from all strings in a list."""
-    return [_sanitize_for_postgres(v) for v in values]
-
-
 def create_db_search_doc(
     server_search_doc: ServerSearchDoc,
     db_session: Session,
     commit: bool = True,
 ) -> DBSearchDoc:
-    # Sanitize string fields to remove NUL characters (PostgreSQL doesn't allow them)
     db_search_doc = DBSearchDoc(
-        document_id=_sanitize_for_postgres(server_search_doc.document_id),
+        document_id=sanitize_string(server_search_doc.document_id),
         chunk_ind=server_search_doc.chunk_ind,
-        semantic_id=_sanitize_for_postgres(server_search_doc.semantic_identifier),
+        semantic_id=sanitize_string(server_search_doc.semantic_identifier),
         link=(
-            _sanitize_for_postgres(server_search_doc.link)
+            sanitize_string(server_search_doc.link)
             if server_search_doc.link is not None
             else None
         ),
-        blurb=_sanitize_for_postgres(server_search_doc.blurb),
+        blurb=sanitize_string(server_search_doc.blurb),
         source_type=server_search_doc.source_type,
         boost=server_search_doc.boost,
         hidden=server_search_doc.hidden,
         doc_metadata=server_search_doc.metadata,
         is_relevant=server_search_doc.is_relevant,
         relevance_explanation=(
-            _sanitize_for_postgres(server_search_doc.relevance_explanation)
+            sanitize_string(server_search_doc.relevance_explanation)
             if server_search_doc.relevance_explanation is not None
             else None
         ),
-        # For docs further down that aren't reranked, we can't use the retrieval score
         score=server_search_doc.score or 0.0,
-        match_highlights=_sanitize_list_for_postgres(
-            server_search_doc.match_highlights
-        ),
+        match_highlights=[
+            sanitize_string(h) for h in server_search_doc.match_highlights
+        ],
         updated_at=server_search_doc.updated_at,
         primary_owners=(
-            _sanitize_list_for_postgres(server_search_doc.primary_owners)
+            [sanitize_string(o) for o in server_search_doc.primary_owners]
             if server_search_doc.primary_owners is not None
             else None
         ),
         secondary_owners=(
-            _sanitize_list_for_postgres(server_search_doc.secondary_owners)
+            [sanitize_string(o) for o in server_search_doc.secondary_owners]
             if server_search_doc.secondary_owners is not None
             else None
         ),

backend/onyx/db/document_set.py

@@ -13,6 +13,7 @@ from sqlalchemy.orm import aliased
 from sqlalchemy.orm import selectinload
 from sqlalchemy.orm import Session
 
+from onyx.configs.app_configs import DISABLE_VECTOR_DB
 from onyx.db.connector_credential_pair import get_cc_pair_groups_for_ids
 from onyx.db.connector_credential_pair import get_connector_credential_pairs
 from onyx.db.enums import AccessType
@@ -246,6 +247,7 @@ def insert_document_set(
             description=document_set_creation_request.description,
             user_id=user_id,
             is_public=document_set_creation_request.is_public,
+            is_up_to_date=DISABLE_VECTOR_DB,
             time_last_modified_by_user=func.now(),
         )
         db_session.add(new_document_set_row)
@@ -336,7 +338,8 @@ def update_document_set(
             )
 
         document_set_row.description = document_set_update_request.description
-        document_set_row.is_up_to_date = False
+        if not DISABLE_VECTOR_DB:
+            document_set_row.is_up_to_date = False
         document_set_row.is_public = document_set_update_request.is_public
         document_set_row.time_last_modified_by_user = func.now()
         versioned_private_doc_set_fn = fetch_versioned_implementation(

backend/onyx/db/hierarchy.py

@@ -1,5 +1,7 @@
 """CRUD operations for HierarchyNode."""
 
+from collections import defaultdict
+
 from sqlalchemy import select
 from sqlalchemy.orm import Session
 
@@ -525,6 +527,53 @@ def get_document_parent_hierarchy_node_ids(
     return {doc_id: parent_id for doc_id, parent_id in results}
 
 
+def update_document_parent_hierarchy_nodes(
+    db_session: Session,
+    doc_parent_map: dict[str, int | None],
+    commit: bool = True,
+) -> int:
+    """Bulk-update Document.parent_hierarchy_node_id for multiple documents.
+
+    Only updates rows whose current value differs from the desired value to
+    avoid unnecessary writes.
+
+    Args:
+        db_session: SQLAlchemy session
+        doc_parent_map: Mapping of document_id → desired parent_hierarchy_node_id
+        commit: Whether to commit the transaction
+
+    Returns:
+        Number of documents actually updated
+    """
+    if not doc_parent_map:
+        return 0
+
+    doc_ids = list(doc_parent_map.keys())
+    existing = get_document_parent_hierarchy_node_ids(db_session, doc_ids)
+
+    by_parent: dict[int | None, list[str]] = defaultdict(list)
+    for doc_id, desired_parent_id in doc_parent_map.items():
+        current = existing.get(doc_id)
+        if current == desired_parent_id or doc_id not in existing:
+            continue
+        by_parent[desired_parent_id].append(doc_id)
+
+    updated = 0
+    for desired_parent_id, ids in by_parent.items():
+        db_session.query(Document).filter(Document.id.in_(ids)).update(
+            {Document.parent_hierarchy_node_id: desired_parent_id},
+            synchronize_session=False,
+        )
+        updated += len(ids)
+
+    if commit:
+        db_session.commit()
+    elif updated:
+        db_session.flush()
+
+    return updated
+
+
 def update_hierarchy_node_permissions(
     db_session: Session,
     raw_node_id: str,

backend/onyx/db/llm.py

@@ -25,8 +25,11 @@ from onyx.server.manage.embedding.models import CloudEmbeddingProvider
 from onyx.server.manage.embedding.models import CloudEmbeddingProviderCreationRequest
 from onyx.server.manage.llm.models import LLMProviderUpsertRequest
 from onyx.server.manage.llm.models import LLMProviderView
+from onyx.utils.logger import setup_logger
 from shared_configs.enums import EmbeddingProvider
 
+logger = setup_logger()
+
 
 def update_group_llm_provider_relationships__no_commit(
     llm_provider_id: int,
@@ -532,6 +535,7 @@ def fetch_default_model(
 ) -> ModelConfiguration | None:
     model_config = db_session.scalar(
         select(ModelConfiguration)
+        .options(selectinload(ModelConfiguration.llm_provider))
         .join(LLMModelFlow)
         .where(
             ModelConfiguration.is_visible == True,  # noqa: E712
@@ -811,6 +815,43 @@ def sync_auto_mode_models(
             changes += 1
 
     db_session.commit()
+
+    # Update the default if this provider currently holds the global CHAT default
+    recommended_default = llm_recommendations.get_default_model(provider.provider)
+    if recommended_default:
+        current_default_name = db_session.scalar(
+            select(ModelConfiguration.name)
+            .join(
+                LLMModelFlow,
+                LLMModelFlow.model_configuration_id == ModelConfiguration.id,
+            )
+            .where(
+                ModelConfiguration.llm_provider_id == provider.id,
+                LLMModelFlow.llm_model_flow_type == LLMModelFlowType.CHAT,
+                LLMModelFlow.is_default == True,  # noqa: E712
+            )
+        )
+
+        if (
+            current_default_name is not None
+            and current_default_name != recommended_default.name
+        ):
+            try:
+                _update_default_model(
+                    db_session=db_session,
+                    provider_id=provider.id,
+                    model=recommended_default.name,
+                    flow_type=LLMModelFlowType.CHAT,
+                )
+                changes += 1
+            except ValueError:
+                logger.warning(
+                    "Recommended default model '%s' not found "
+                    "for provider_id=%s; skipping default update.",
+                    recommended_default.name,
+                    provider.id,
+                )
+
     return changes
 
 

backend/onyx/db/models.py

@@ -2822,8 +2822,17 @@ class LLMProvider(Base):
         postgresql.JSONB(), nullable=True
     )
 
+    # Deprecated: use LLMModelFlow with CHAT flow type instead
+    default_model_name: Mapped[str | None] = mapped_column(String, nullable=True)
+
     deployment_name: Mapped[str | None] = mapped_column(String, nullable=True)
 
+    # Deprecated: use LLMModelFlow.is_default with CHAT flow type instead
+    is_default_provider: Mapped[bool | None] = mapped_column(Boolean, nullable=True)
+    # Deprecated: use LLMModelFlow.is_default with VISION flow type instead
+    is_default_vision_provider: Mapped[bool | None] = mapped_column(Boolean)
+    # Deprecated: use LLMModelFlow with VISION flow type instead
+    default_vision_model: Mapped[str | None] = mapped_column(String, nullable=True)
     # EE only
     is_public: Mapped[bool] = mapped_column(Boolean, nullable=False, default=True)
     # Auto mode: models, visibility, and defaults are managed by GitHub config
@@ -4917,7 +4926,9 @@ class ScimUserMapping(Base):
     __tablename__ = "scim_user_mapping"
 
     id: Mapped[int] = mapped_column(Integer, primary_key=True)
-    external_id: Mapped[str] = mapped_column(String, unique=True, index=True)
+    external_id: Mapped[str | None] = mapped_column(
+        String, unique=True, index=True, nullable=True
+    )
     user_id: Mapped[UUID] = mapped_column(
         ForeignKey("user.id", ondelete="CASCADE"), unique=True, nullable=False
     )
@@ -4974,3 +4985,25 @@ class CodeInterpreterServer(Base):
 
     id: Mapped[int] = mapped_column(Integer, primary_key=True)
     server_enabled: Mapped[bool] = mapped_column(Boolean, nullable=False, default=True)
+
+
+class CacheStore(Base):
+    """Key-value cache table used by ``PostgresCacheBackend``.
+
+    Replaces Redis for simple KV caching, locks, and list operations
+    when ``CACHE_BACKEND=postgres`` (NO_VECTOR_DB deployments).
+
+    Intentionally separate from ``KVStore``:
+    - Stores raw bytes (LargeBinary) vs JSONB, matching Redis semantics.
+    - Has ``expires_at`` for TTL; rows are periodically garbage-collected.
+    - Holds ephemeral data (tokens, stop signals, lock state) not
+      persistent application config, so cleanup can be aggressive.
+    """
+
+    __tablename__ = "cache_store"
+
+    key: Mapped[str] = mapped_column(String, primary_key=True)
+    value: Mapped[bytes | None] = mapped_column(LargeBinary, nullable=True)
+    expires_at: Mapped[datetime.datetime | None] = mapped_column(
+        DateTime(timezone=True), nullable=True
+    )

backend/onyx/db/projects.py

@@ -52,7 +52,7 @@ def create_user_files(
 ) -> CategorizedFilesResult:
 
     # Categorize the files
-    categorized_files = categorize_uploaded_files(files)
+    categorized_files = categorize_uploaded_files(files, db_session)
     # NOTE: At the moment, zip metadata is not used for user files.
     # Should revisit to decide whether this should be a feature.
     upload_response = upload_files(categorized_files.acceptable, FileOrigin.USER_FILE)

backend/onyx/db/search_settings.py

@@ -129,7 +129,7 @@ def get_current_search_settings(db_session: Session) -> SearchSettings:
     latest_settings = result.scalars().first()
 
     if not latest_settings:
-        raise RuntimeError("No search settings specified, DB is not in a valid state")
+        raise RuntimeError("No search settings specified; DB is not in a valid state.")
     return latest_settings
 
 

backend/onyx/db/tools.py

@@ -13,12 +13,15 @@ from onyx.db.constants import UNSET
 from onyx.db.constants import UnsetType
 from onyx.db.enums import MCPServerStatus
 from onyx.db.models import MCPServer
+from onyx.db.models import OAuthConfig
 from onyx.db.models import Tool
 from onyx.db.models import ToolCall
 from onyx.server.features.tool.models import Header
 from onyx.tools.built_in_tools import BUILT_IN_TOOL_TYPES
 from onyx.utils.headers import HeaderItemDict
 from onyx.utils.logger import setup_logger
+from onyx.utils.postgres_sanitization import sanitize_json_like
+from onyx.utils.postgres_sanitization import sanitize_string
 
 if TYPE_CHECKING:
     pass
@@ -159,10 +162,26 @@ def update_tool(
         ]
     if passthrough_auth is not None:
         tool.passthrough_auth = passthrough_auth
+    old_oauth_config_id = tool.oauth_config_id
     if not isinstance(oauth_config_id, UnsetType):
         tool.oauth_config_id = oauth_config_id
-    db_session.commit()
+        db_session.flush()
 
+    # Clean up orphaned OAuthConfig if the oauth_config_id was changed
+    if (
+        old_oauth_config_id is not None
+        and not isinstance(oauth_config_id, UnsetType)
+        and old_oauth_config_id != oauth_config_id
+    ):
+        other_tools = db_session.scalars(
+            select(Tool).where(Tool.oauth_config_id == old_oauth_config_id)
+        ).all()
+        if not other_tools:
+            oauth_config = db_session.get(OAuthConfig, old_oauth_config_id)
+            if oauth_config:
+                db_session.delete(oauth_config)
+
+    db_session.commit()
     return tool
 
 
@@ -171,8 +190,21 @@ def delete_tool__no_commit(tool_id: int, db_session: Session) -> None:
     if tool is None:
         raise ValueError(f"Tool with ID {tool_id} does not exist")
 
+    oauth_config_id = tool.oauth_config_id
+
     db_session.delete(tool)
-    db_session.flush()  # Don't commit yet, let caller decide when to commit
+    db_session.flush()
+
+    # Clean up orphaned OAuthConfig if no other tools reference it
+    if oauth_config_id is not None:
+        other_tools = db_session.scalars(
+            select(Tool).where(Tool.oauth_config_id == oauth_config_id)
+        ).all()
+        if not other_tools:
+            oauth_config = db_session.get(OAuthConfig, oauth_config_id)
+            if oauth_config:
+                db_session.delete(oauth_config)
+                db_session.flush()
 
 
 def get_builtin_tool(
@@ -256,11 +288,13 @@ def create_tool_call_no_commit(
         tab_index=tab_index,
         tool_id=tool_id,
         tool_call_id=tool_call_id,
-        reasoning_tokens=reasoning_tokens,
-        tool_call_arguments=tool_call_arguments,
-        tool_call_response=tool_call_response,
+        reasoning_tokens=(
+            sanitize_string(reasoning_tokens) if reasoning_tokens else reasoning_tokens
+        ),
+        tool_call_arguments=sanitize_json_like(tool_call_arguments),
+        tool_call_response=sanitize_json_like(tool_call_response),
         tool_call_tokens=tool_call_tokens,
-        generated_images=generated_images,
+        generated_images=sanitize_json_like(generated_images),
     )
 
     db_session.add(tool_call)

backend/onyx/document_index/FILTER_SEMANTICS.md

@@ -0,0 +1,103 @@
+# Vector DB Filter Semantics
+
+How `IndexFilters` fields combine into the final query filter. Applies to both Vespa and OpenSearch.
+
+## Filter categories
+
+| Category | Fields | Join logic |
+|---|---|---|
+| **Visibility** | `hidden` | Always applied (unless `include_hidden`) |
+| **Tenant** | `tenant_id` | AND (multi-tenant only) |
+| **ACL** | `access_control_list` | OR within, AND with rest |
+| **Narrowing** | `source_type`, `tags`, `time_cutoff` | Each OR within, AND with rest |
+| **Knowledge scope** | `document_set`, `user_file_ids`, `attached_document_ids`, `hierarchy_node_ids` | OR within group, AND with rest |
+| **Additive scope** | `project_id`, `persona_id` | OR'd into knowledge scope **only when** a knowledge scope filter already exists |
+
+## How filters combine
+
+All categories are AND'd together. Within the knowledge scope category, individual filters are OR'd.
+
+```
+NOT hidden
+AND tenant = T                          -- if multi-tenant
+AND (acl contains A1 OR acl contains A2)
+AND (source_type = S1 OR ...)           -- if set
+AND (tag = T1 OR ...)                   -- if set
+AND <knowledge scope>                   -- see below
+AND time >= cutoff                      -- if set
+```
+
+## Knowledge scope rules
+
+The knowledge scope filter controls **what knowledge an assistant can access**.
+
+### No explicit knowledge attached
+
+When `document_set`, `user_file_ids`, `attached_document_ids`, and `hierarchy_node_ids` are all empty/None:
+
+- **No knowledge scope filter is applied.** The assistant can see everything (subject to ACL).
+- `project_id` and `persona_id` are ignored — they never restrict on their own.
+
+### One explicit knowledge type
+
+```
+-- Only document sets
+AND (document_sets contains "Engineering" OR document_sets contains "Legal")
+
+-- Only user files
+AND (document_id = "uuid-1" OR document_id = "uuid-2")
+```
+
+### Multiple explicit knowledge types (OR'd)
+
+```
+-- Document sets + user files
+AND (
+    document_sets contains "Engineering"
+    OR document_id = "uuid-1"
+)
+```
+
+### Explicit knowledge + overflowing user files
+
+When an explicit knowledge restriction is in effect **and** `project_id` or `persona_id` is set (user files overflowed the LLM context window), the additive scopes widen the filter:
+
+```
+-- Document sets + persona user files overflowed
+AND (
+    document_sets contains "Engineering"
+    OR personas contains 42
+)
+
+-- User files + project files overflowed
+AND (
+    document_id = "uuid-1"
+    OR user_project contains 7
+)
+```
+
+### Only project_id or persona_id (no explicit knowledge)
+
+No knowledge scope filter. The assistant searches everything.
+
+```
+-- Just ACL, no restriction
+NOT hidden
+AND (acl contains ...)
+```
+
+## Field reference
+
+| Filter field | Vespa field | Vespa type | Purpose |
+|---|---|---|---|
+| `document_set` | `document_sets` | `weightedset<string>` | Connector doc sets attached to assistant |
+| `user_file_ids` | `document_id` | `string` | User files uploaded to assistant |
+| `attached_document_ids` | `document_id` | `string` | Documents explicitly attached (OpenSearch only) |
+| `hierarchy_node_ids` | `ancestor_hierarchy_node_ids` | `array<int>` | Folder/space nodes (OpenSearch only) |
+| `project_id` | `user_project` | `array<int>` | Project tag for overflowing user files |
+| `persona_id` | `personas` | `array<int>` | Persona tag for overflowing user files |
+| `access_control_list` | `access_control_list` | `weightedset<string>` | ACL entries for the requesting user |
+| `source_type` | `source_type` | `string` | Connector source type (e.g. `web`, `jira`) |
+| `tags` | `metadata_list` | `array<string>` | Document metadata tags |
+| `time_cutoff` | `doc_updated_at` | `long` | Minimum document update timestamp |
+| `tenant_id` | `tenant_id` | `string` | Tenant isolation (multi-tenant) |

backend/onyx/document_index/document_index_utils.py

@@ -32,9 +32,6 @@ def get_multipass_config(search_settings: SearchSettings) -> MultipassConfig:
     Determines whether to enable multipass and large chunks by examining
     the current search settings and the embedder configuration.
     """
-    if not search_settings:
-        return MultipassConfig(multipass_indexing=False, enable_large_chunks=False)
-
     multipass = should_use_multipass(search_settings)
     enable_large_chunks = SearchSettings.can_use_large_chunks(
         multipass, search_settings.model_name, search_settings.provider_type

backend/onyx/document_index/factory.py

@@ -26,11 +26,10 @@ def get_default_document_index(
     To be used for retrieval only. Indexing should be done through both indices
     until Vespa is deprecated.
 
-    Pre-existing docstring for this function, although secondary indices are not
-    currently supported:
     Primary index is the index that is used for querying/updating etc. Secondary
     index is for when both the currently used index and the upcoming index both
-    need to be updated, updates are applied to both indices.
+    need to be updated. Updates are applied to both indices.
+    WARNING: In that case, get_all_document_indices should be used.
     """
     if DISABLE_VECTOR_DB:
         return DisabledDocumentIndex(
@@ -51,11 +50,26 @@ def get_default_document_index(
     opensearch_retrieval_enabled = get_opensearch_retrieval_state(db_session)
     if opensearch_retrieval_enabled:
         indexing_setting = IndexingSetting.from_db_model(search_settings)
+        secondary_indexing_setting = (
+            IndexingSetting.from_db_model(secondary_search_settings)
+            if secondary_search_settings
+            else None
+        )
         return OpenSearchOldDocumentIndex(
             index_name=search_settings.index_name,
             embedding_dim=indexing_setting.final_embedding_dim,
             embedding_precision=indexing_setting.embedding_precision,
             secondary_index_name=secondary_index_name,
+            secondary_embedding_dim=(
+                secondary_indexing_setting.final_embedding_dim
+                if secondary_indexing_setting
+                else None
+            ),
+            secondary_embedding_precision=(
+                secondary_indexing_setting.embedding_precision
+                if secondary_indexing_setting
+                else None
+            ),
             large_chunks_enabled=search_settings.large_chunks_enabled,
             secondary_large_chunks_enabled=secondary_large_chunks_enabled,
             multitenant=MULTI_TENANT,
@@ -86,8 +100,7 @@ def get_all_document_indices(
     Used for indexing only. Until Vespa is deprecated we will index into both
     document indices. Retrieval is done through only one index however.
 
-    Large chunks and secondary indices are not currently supported so we
-    hardcode appropriate values.
+    Large chunks are not currently supported so we hardcode appropriate values.
 
     NOTE: Make sure the Vespa index object is returned first. In the rare event
     that there is some conflict between indexing and the migration task, it is
@@ -123,13 +136,36 @@ def get_all_document_indices(
     opensearch_document_index: OpenSearchOldDocumentIndex | None = None
     if ENABLE_OPENSEARCH_INDEXING_FOR_ONYX:
         indexing_setting = IndexingSetting.from_db_model(search_settings)
+        secondary_indexing_setting = (
+            IndexingSetting.from_db_model(secondary_search_settings)
+            if secondary_search_settings
+            else None
+        )
         opensearch_document_index = OpenSearchOldDocumentIndex(
             index_name=search_settings.index_name,
             embedding_dim=indexing_setting.final_embedding_dim,
             embedding_precision=indexing_setting.embedding_precision,
-            secondary_index_name=None,
-            large_chunks_enabled=False,
-            secondary_large_chunks_enabled=None,
+            secondary_index_name=(
+                secondary_search_settings.index_name
+                if secondary_search_settings
+                else None
+            ),
+            secondary_embedding_dim=(
+                secondary_indexing_setting.final_embedding_dim
+                if secondary_indexing_setting
+                else None
+            ),
+            secondary_embedding_precision=(
+                secondary_indexing_setting.embedding_precision
+                if secondary_indexing_setting
+                else None
+            ),
+            large_chunks_enabled=search_settings.large_chunks_enabled,
+            secondary_large_chunks_enabled=(
+                secondary_search_settings.large_chunks_enabled
+                if secondary_search_settings
+                else None
+            ),
             multitenant=MULTI_TENANT,
             httpx_client=httpx_client,
         )

backend/onyx/document_index/opensearch/client.py

@@ -61,6 +61,25 @@ class SearchHit(BaseModel, Generic[SchemaDocumentModel]):
     explanation: dict[str, Any] | None = None
 
 
+class IndexInfo(BaseModel):
+    """
+    Represents information about an OpenSearch index.
+    """
+
+    model_config = {"frozen": True}
+
+    name: str
+    health: str
+    status: str
+    num_primary_shards: str
+    num_replica_shards: str
+    docs_count: str
+    docs_deleted: str
+    created_at: str
+    total_size: str
+    primary_shards_size: str
+
+
 def get_new_body_without_vectors(body: dict[str, Any]) -> dict[str, Any]:
     """Recursively replaces vectors in the body with their length.
 
@@ -159,8 +178,8 @@ class OpenSearchClient(AbstractContextManager):
         Raises:
             Exception: There was an error creating the search pipeline.
         """
-        result = self._client.search_pipeline.put(id=pipeline_id, body=pipeline_body)
-        if not result.get("acknowledged", False):
+        response = self._client.search_pipeline.put(id=pipeline_id, body=pipeline_body)
+        if not response.get("acknowledged", False):
             raise RuntimeError(f"Failed to create search pipeline {pipeline_id}.")
 
     @log_function_time(print_only=True, debug_only=True, include_args=True)
@@ -173,8 +192,8 @@ class OpenSearchClient(AbstractContextManager):
         Raises:
             Exception: There was an error deleting the search pipeline.
         """
-        result = self._client.search_pipeline.delete(id=pipeline_id)
-        if not result.get("acknowledged", False):
+        response = self._client.search_pipeline.delete(id=pipeline_id)
+        if not response.get("acknowledged", False):
             raise RuntimeError(f"Failed to delete search pipeline {pipeline_id}.")
 
     @log_function_time(print_only=True, debug_only=True, include_args=True)
@@ -198,6 +217,34 @@ class OpenSearchClient(AbstractContextManager):
             logger.error(f"Failed to put cluster settings: {response}.")
             return False
 
+    @log_function_time(print_only=True, debug_only=True)
+    def list_indices_with_info(self) -> list[IndexInfo]:
+        """
+        Lists the indices in the OpenSearch cluster with information about each
+        index.
+
+        Returns:
+            A list of IndexInfo objects for each index.
+        """
+        response = self._client.cat.indices(format="json")
+        indices: list[IndexInfo] = []
+        for raw_index_info in response:
+            indices.append(
+                IndexInfo(
+                    name=raw_index_info.get("index", ""),
+                    health=raw_index_info.get("health", ""),
+                    status=raw_index_info.get("status", ""),
+                    num_primary_shards=raw_index_info.get("pri", ""),
+                    num_replica_shards=raw_index_info.get("rep", ""),
+                    docs_count=raw_index_info.get("docs.count", ""),
+                    docs_deleted=raw_index_info.get("docs.deleted", ""),
+                    created_at=raw_index_info.get("creation.date.string", ""),
+                    total_size=raw_index_info.get("store.size", ""),
+                    primary_shards_size=raw_index_info.get("pri.store.size", ""),
+                )
+            )
+        return indices
+
     @log_function_time(print_only=True, debug_only=True)
     def ping(self) -> bool:
         """Pings the OpenSearch cluster.

backend/onyx/document_index/opensearch/opensearch_document_index.py

@@ -271,6 +271,9 @@ class OpenSearchOldDocumentIndex(OldDocumentIndex):
         embedding_dim: int,
         embedding_precision: EmbeddingPrecision,
         secondary_index_name: str | None,
+        secondary_embedding_dim: int | None,
+        secondary_embedding_precision: EmbeddingPrecision | None,
+        # NOTE: We do not support large chunks right now.
         large_chunks_enabled: bool,  # noqa: ARG002
         secondary_large_chunks_enabled: bool | None,  # noqa: ARG002
         multitenant: bool = False,
@@ -286,12 +289,25 @@ class OpenSearchOldDocumentIndex(OldDocumentIndex):
                 f"Expected {MULTI_TENANT}, got {multitenant}."
             )
         tenant_id = get_current_tenant_id()
+        tenant_state = TenantState(tenant_id=tenant_id, multitenant=multitenant)
         self._real_index = OpenSearchDocumentIndex(
-            tenant_state=TenantState(tenant_id=tenant_id, multitenant=multitenant),
+            tenant_state=tenant_state,
             index_name=index_name,
             embedding_dim=embedding_dim,
             embedding_precision=embedding_precision,
         )
+        self._secondary_real_index: OpenSearchDocumentIndex | None = None
+        if self.secondary_index_name:
+            if secondary_embedding_dim is None or secondary_embedding_precision is None:
+                raise ValueError(
+                    "Bug: Secondary index embedding dimension and precision are not set."
+                )
+            self._secondary_real_index = OpenSearchDocumentIndex(
+                tenant_state=tenant_state,
+                index_name=self.secondary_index_name,
+                embedding_dim=secondary_embedding_dim,
+                embedding_precision=secondary_embedding_precision,
+            )
 
     @staticmethod
     def register_multitenant_indices(
@@ -307,19 +323,38 @@ class OpenSearchOldDocumentIndex(OldDocumentIndex):
         self,
         primary_embedding_dim: int,
         primary_embedding_precision: EmbeddingPrecision,
-        secondary_index_embedding_dim: int | None,  # noqa: ARG002
-        secondary_index_embedding_precision: EmbeddingPrecision | None,  # noqa: ARG002
+        secondary_index_embedding_dim: int | None,
+        secondary_index_embedding_precision: EmbeddingPrecision | None,
     ) -> None:
-        # Only handle primary index for now, ignore secondary.
-        return self._real_index.verify_and_create_index_if_necessary(
+        self._real_index.verify_and_create_index_if_necessary(
             primary_embedding_dim, primary_embedding_precision
         )
+        if self.secondary_index_name:
+            if (
+                secondary_index_embedding_dim is None
+                or secondary_index_embedding_precision is None
+            ):
+                raise ValueError(
+                    "Bug: Secondary index embedding dimension and precision are not set."
+                )
+            assert (
+                self._secondary_real_index is not None
+            ), "Bug: Secondary index is not initialized."
+            self._secondary_real_index.verify_and_create_index_if_necessary(
+                secondary_index_embedding_dim, secondary_index_embedding_precision
+            )
 
     def index(
         self,
         chunks: list[DocMetadataAwareIndexChunk],
         index_batch_params: IndexBatchParams,
     ) -> set[OldDocumentInsertionRecord]:
+        """
+        NOTE: Do NOT consider the secondary index here. A separate indexing
+        pipeline will be responsible for indexing to the secondary index. This
+        design is not ideal and we should reconsider this when revamping index
+        swapping.
+        """
         # Convert IndexBatchParams to IndexingMetadata.
         chunk_counts: dict[str, IndexingMetadata.ChunkCounts] = {}
         for doc_id in index_batch_params.doc_id_to_new_chunk_cnt:
@@ -351,7 +386,20 @@ class OpenSearchOldDocumentIndex(OldDocumentIndex):
         tenant_id: str,  # noqa: ARG002
         chunk_count: int | None,
     ) -> int:
-        return self._real_index.delete(doc_id, chunk_count)
+        """
+        NOTE: Remember to handle the secondary index here. There is no separate
+        pipeline for deleting chunks in the secondary index. This design is not
+        ideal and we should reconsider this when revamping index swapping.
+        """
+        total_chunks_deleted = self._real_index.delete(doc_id, chunk_count)
+        if self.secondary_index_name:
+            assert (
+                self._secondary_real_index is not None
+            ), "Bug: Secondary index is not initialized."
+            total_chunks_deleted += self._secondary_real_index.delete(
+                doc_id, chunk_count
+            )
+        return total_chunks_deleted
 
     def update_single(
         self,
@@ -362,6 +410,11 @@ class OpenSearchOldDocumentIndex(OldDocumentIndex):
         fields: VespaDocumentFields | None,
         user_fields: VespaDocumentUserFields | None,
     ) -> None:
+        """
+        NOTE: Remember to handle the secondary index here. There is no separate
+        pipeline for updating chunks in the secondary index. This design is not
+        ideal and we should reconsider this when revamping index swapping.
+        """
         if fields is None and user_fields is None:
             logger.warning(
                 f"Tried to update document {doc_id} with no updated fields or user fields."
@@ -392,6 +445,11 @@ class OpenSearchOldDocumentIndex(OldDocumentIndex):
 
         try:
             self._real_index.update([update_request])
+            if self.secondary_index_name:
+                assert (
+                    self._secondary_real_index is not None
+                ), "Bug: Secondary index is not initialized."
+                self._secondary_real_index.update([update_request])
         except NotFoundError:
             logger.exception(
                 f"Tried to update document {doc_id} but at least one of its chunks was not found in OpenSearch. "
@@ -681,7 +739,8 @@ class OpenSearchDocumentIndex(DocumentIndex):
             The number of chunks successfully deleted.
         """
         logger.debug(
-            f"[OpenSearchDocumentIndex] Deleting document {document_id} from index {self._index_name}."
+            f"[OpenSearchDocumentIndex] Deleting document {document_id} from index "
+            f"{self._index_name}."
         )
         query_body = DocumentQuery.delete_from_document_id_query(
             document_id=document_id,
@@ -717,7 +776,8 @@ class OpenSearchDocumentIndex(DocumentIndex):
                 specified documents.
         """
         logger.debug(
-            f"[OpenSearchDocumentIndex] Updating {len(update_requests)} chunks for index {self._index_name}."
+            f"[OpenSearchDocumentIndex] Updating {len(update_requests)} chunks for index "
+            f"{self._index_name}."
         )
         for update_request in update_requests:
             properties_to_update: dict[str, Any] = dict()
@@ -773,9 +833,11 @@ class OpenSearchDocumentIndex(DocumentIndex):
                     # here.
                     # TODO(andrei): Fix the aforementioned race condition.
                     raise ChunkCountNotFoundError(
-                        f"Tried to update document {doc_id} but its chunk count is not known. Older versions of the "
-                        "application used to permit this but is not a supported state for a document when using OpenSearch. "
-                        "The document was likely just added to the indexing pipeline and the chunk count will be updated shortly."
+                        f"Tried to update document {doc_id} but its chunk count is not known. "
+                        "Older versions of the application used to permit this but is not a "
+                        "supported state for a document when using OpenSearch. The document was "
+                        "likely just added to the indexing pipeline and the chunk count will be "
+                        "updated shortly."
                     )
                 if doc_chunk_count == 0:
                     raise ValueError(
@@ -807,7 +869,8 @@ class OpenSearchDocumentIndex(DocumentIndex):
         chunk IDs vs querying for matching document chunks.
         """
         logger.debug(
-            f"[OpenSearchDocumentIndex] Retrieving {len(chunk_requests)} chunks for index {self._index_name}."
+            f"[OpenSearchDocumentIndex] Retrieving {len(chunk_requests)} chunks for index "
+            f"{self._index_name}."
         )
         results: list[InferenceChunk] = []
         for chunk_request in chunk_requests:
@@ -854,7 +917,8 @@ class OpenSearchDocumentIndex(DocumentIndex):
         num_to_retrieve: int,
     ) -> list[InferenceChunk]:
         logger.debug(
-            f"[OpenSearchDocumentIndex] Hybrid retrieving {num_to_retrieve} chunks for index {self._index_name}."
+            f"[OpenSearchDocumentIndex] Hybrid retrieving {num_to_retrieve} chunks for index "
+            f"{self._index_name}."
         )
         # TODO(andrei): This could be better, the caller should just make this
         # decision when passing in the query param. See the above comment in the
@@ -874,8 +938,10 @@ class OpenSearchDocumentIndex(DocumentIndex):
             index_filters=filters,
             include_hidden=False,
         )
-        # NOTE: Using z-score normalization here because it's better for hybrid search from a theoretical standpoint.
-        # Empirically on a small dataset of up to 10K docs, it's not very different. Likely more impactful at scale.
+        # NOTE: Using z-score normalization here because it's better for hybrid
+        # search from a theoretical standpoint. Empirically on a small dataset
+        # of up to 10K docs, it's not very different. Likely more impactful at
+        # scale.
         # https://opensearch.org/blog/introducing-the-z-score-normalization-technique-for-hybrid-search/
         search_hits: list[SearchHit[DocumentChunk]] = self._client.search(
             body=query_body,
@@ -902,7 +968,8 @@ class OpenSearchDocumentIndex(DocumentIndex):
         dirty: bool | None = None,  # noqa: ARG002
     ) -> list[InferenceChunk]:
         logger.debug(
-            f"[OpenSearchDocumentIndex] Randomly retrieving {num_to_retrieve} chunks for index {self._index_name}."
+            f"[OpenSearchDocumentIndex] Randomly retrieving {num_to_retrieve} chunks for index "
+            f"{self._index_name}."
         )
         query_body = DocumentQuery.get_random_search_query(
             tenant_state=self._tenant_state,
@@ -932,7 +999,8 @@ class OpenSearchDocumentIndex(DocumentIndex):
         complete.
         """
         logger.debug(
-            f"[OpenSearchDocumentIndex] Indexing {len(chunks)} raw chunks for index {self._index_name}."
+            f"[OpenSearchDocumentIndex] Indexing {len(chunks)} raw chunks for index "
+            f"{self._index_name}."
         )
         # Do not raise if the document already exists, just update. This is
         # because the document may already have been indexed during the

backend/onyx/document_index/opensearch/schema.py

@@ -243,7 +243,8 @@ class DocumentChunk(BaseModel):
             return value
         if not isinstance(value, int):
             raise ValueError(
-                f"Bug: Expected an int for the last_updated property from OpenSearch, got {type(value)} instead."
+                f"Bug: Expected an int for the last_updated property from OpenSearch, got "
+                f"{type(value)} instead."
             )
         return datetime.fromtimestamp(value, tz=timezone.utc)
 
@@ -284,19 +285,22 @@ class DocumentChunk(BaseModel):
         elif isinstance(value, TenantState):
             if MULTI_TENANT != value.multitenant:
                 raise ValueError(
-                    f"Bug: An existing TenantState object was supplied to the DocumentChunk model but its multi-tenant mode "
-                    f"({value.multitenant}) does not match the program's current global tenancy state."
+                    f"Bug: An existing TenantState object was supplied to the DocumentChunk model "
+                    f"but its multi-tenant mode ({value.multitenant}) does not match the program's "
+                    "current global tenancy state."
                 )
             return value
         elif not isinstance(value, str):
             raise ValueError(
-                f"Bug: Expected a str for the tenant_id property from OpenSearch, got {type(value)} instead."
+                f"Bug: Expected a str for the tenant_id property from OpenSearch, got "
+                f"{type(value)} instead."
             )
         else:
             if not MULTI_TENANT:
                 raise ValueError(
-                    "Bug: Got a non-null str for the tenant_id property from OpenSearch but multi-tenant mode is not enabled. "
-                    "This is unexpected because in single-tenant mode we don't expect to see a tenant_id."
+                    "Bug: Got a non-null str for the tenant_id property from OpenSearch but "
+                    "multi-tenant mode is not enabled. This is unexpected because in single-tenant "
+                    "mode we don't expect to see a tenant_id."
                 )
             return TenantState(tenant_id=value, multitenant=MULTI_TENANT)
 
@@ -352,8 +356,10 @@ class DocumentSchema:
             "properties": {
                 TITLE_FIELD_NAME: {
                     "type": "text",
-                    # Language analyzer (e.g. english) stems at index and search time for variant matching.
-                    # Configure via OPENSEARCH_TEXT_ANALYZER. Existing indices need reindexing after a change.
+                    # Language analyzer (e.g. english) stems at index and search
+                    # time for variant matching. Configure via
+                    # OPENSEARCH_TEXT_ANALYZER. Existing indices need reindexing
+                    # after a change.
                     "analyzer": OPENSEARCH_TEXT_ANALYZER,
                     "fields": {
                         # Subfield accessed as title.keyword. Not indexed for

backend/onyx/document_index/opensearch/search.py

@@ -698,41 +698,6 @@ class DocumentQuery:
             """
             return {"terms": {ANCESTOR_HIERARCHY_NODE_IDS_FIELD_NAME: node_ids}}
 
-        def _get_assistant_knowledge_filter(
-            attached_doc_ids: list[str] | None,
-            node_ids: list[int] | None,
-            file_ids: list[UUID] | None,
-            document_sets: list[str] | None,
-        ) -> dict[str, Any]:
-            """Combined filter for assistant knowledge.
-
-            When an assistant has attached knowledge, search should be scoped to:
-            - Documents explicitly attached (by document ID), OR
-            - Documents under attached hierarchy nodes (by ancestor node IDs), OR
-            - User-uploaded files attached to the assistant, OR
-            - Documents in the assistant's document sets (if any)
-            """
-            knowledge_filter: dict[str, Any] = {
-                "bool": {"should": [], "minimum_should_match": 1}
-            }
-            if attached_doc_ids:
-                knowledge_filter["bool"]["should"].append(
-                    _get_attached_document_id_filter(attached_doc_ids)
-                )
-            if node_ids:
-                knowledge_filter["bool"]["should"].append(
-                    _get_hierarchy_node_filter(node_ids)
-                )
-            if file_ids:
-                knowledge_filter["bool"]["should"].append(
-                    _get_user_file_id_filter(file_ids)
-                )
-            if document_sets:
-                knowledge_filter["bool"]["should"].append(
-                    _get_document_set_filter(document_sets)
-                )
-            return knowledge_filter
-
         filter_clauses: list[dict[str, Any]] = []
 
         if not include_hidden:
@@ -758,41 +723,53 @@ class DocumentQuery:
             # document's metadata list.
             filter_clauses.append(_get_tag_filter(tags))
 
-        # Check if this is an assistant knowledge search (has any assistant-scoped knowledge)
-        has_assistant_knowledge = (
+        # Knowledge scope: explicit knowledge attachments restrict what
+        # an assistant can see.  When none are set the assistant
+        # searches everything.
+        #
+        # project_id / persona_id are additive: they make overflowing
+        # user files findable but must NOT trigger the restriction on
+        # their own (an agent with no explicit knowledge should search
+        # everything).
+        has_knowledge_scope = (
             attached_document_ids
             or hierarchy_node_ids
             or user_file_ids
             or document_sets
         )
 
-        if has_assistant_knowledge:
-            # If assistant has attached knowledge, scope search to that knowledge.
-            # Document sets are included in the OR filter so directly attached
-            # docs are always findable even if not in the document sets.
-            filter_clauses.append(
-                _get_assistant_knowledge_filter(
-                    attached_document_ids,
-                    hierarchy_node_ids,
-                    user_file_ids,
-                    document_sets,
+        if has_knowledge_scope:
+            knowledge_filter: dict[str, Any] = {
+                "bool": {"should": [], "minimum_should_match": 1}
+            }
+            if attached_document_ids:
+                knowledge_filter["bool"]["should"].append(
+                    _get_attached_document_id_filter(attached_document_ids)
                 )
-            )
-        elif user_file_ids:
-            # Fallback for non-assistant user file searches (e.g., project searches)
-            # If at least one user file ID is provided, the caller will only
-            # retrieve documents where the document ID is in this input list of
-            # file IDs.
-            filter_clauses.append(_get_user_file_id_filter(user_file_ids))
-
-        if project_id is not None:
-            # If a project ID is provided, the caller will only retrieve
-            # documents where the project ID provided here is present in the
-            # document's user projects list.
-            filter_clauses.append(_get_user_project_filter(project_id))
-
-        if persona_id is not None:
-            filter_clauses.append(_get_persona_filter(persona_id))
+            if hierarchy_node_ids:
+                knowledge_filter["bool"]["should"].append(
+                    _get_hierarchy_node_filter(hierarchy_node_ids)
+                )
+            if user_file_ids:
+                knowledge_filter["bool"]["should"].append(
+                    _get_user_file_id_filter(user_file_ids)
+                )
+            if document_sets:
+                knowledge_filter["bool"]["should"].append(
+                    _get_document_set_filter(document_sets)
+                )
+            # Additive: widen scope to also cover overflowing user
+            # files, but only when an explicit restriction is already
+            # in effect.
+            if project_id is not None:
+                knowledge_filter["bool"]["should"].append(
+                    _get_user_project_filter(project_id)
+                )
+            if persona_id is not None:
+                knowledge_filter["bool"]["should"].append(
+                    _get_persona_filter(persona_id)
+                )
+            filter_clauses.append(knowledge_filter)
 
         if time_cutoff is not None:
             # If a time cutoff is provided, the caller will only retrieve

backend/onyx/document_index/vespa/chunk_retrieval.py

@@ -1,5 +1,6 @@
 import json
 import string
+import time
 from collections.abc import Callable
 from collections.abc import Mapping
 from datetime import datetime
@@ -18,6 +19,7 @@ from onyx.background.celery.tasks.opensearch_migration.transformer import (
 )
 from onyx.configs.app_configs import LOG_VESPA_TIMING_INFORMATION
 from onyx.configs.app_configs import VESPA_LANGUAGE_OVERRIDE
+from onyx.configs.app_configs import VESPA_MIGRATION_REQUEST_TIMEOUT_S
 from onyx.context.search.models import IndexFilters
 from onyx.context.search.models import InferenceChunkUncleaned
 from onyx.document_index.interfaces import VespaChunkRequest
@@ -338,12 +340,18 @@ def get_all_chunks_paginated(
             params["continuation"] = continuation_token
 
         response: httpx.Response | None = None
+        start_time = time.monotonic()
         try:
-            with get_vespa_http_client() as http_client:
+            with get_vespa_http_client(
+                timeout=VESPA_MIGRATION_REQUEST_TIMEOUT_S
+            ) as http_client:
                 response = http_client.get(url, params=params)
                 response.raise_for_status()
         except httpx.HTTPError as e:
-            error_base = f"Failed to get chunks from Vespa slice {slice_id} with continuation token {continuation_token}."
+            error_base = (
+                f"Failed to get chunks from Vespa slice {slice_id} with continuation token "
+                f"{continuation_token} in {time.monotonic() - start_time:.3f} seconds."
+            )
             logger.exception(
                 f"Request URL: {e.request.url}\n"
                 f"Request Headers: {e.request.headers}\n"

backend/onyx/document_index/vespa/index.py

@@ -465,6 +465,12 @@ class VespaIndex(DocumentIndex):
         chunks: list[DocMetadataAwareIndexChunk],
         index_batch_params: IndexBatchParams,
     ) -> set[OldDocumentInsertionRecord]:
+        """
+        NOTE: Do NOT consider the secondary index here. A separate indexing
+        pipeline will be responsible for indexing to the secondary index. This
+        design is not ideal and we should reconsider this when revamping index
+        swapping.
+        """
         if len(index_batch_params.doc_id_to_previous_chunk_cnt) != len(
             index_batch_params.doc_id_to_new_chunk_cnt
         ):
@@ -659,6 +665,10 @@ class VespaIndex(DocumentIndex):
         """Note: if the document id does not exist, the update will be a no-op and the
         function will complete with no errors or exceptions.
         Handle other exceptions if you wish to implement retry behavior
+
+        NOTE: Remember to handle the secondary index here. There is no separate
+        pipeline for updating chunks in the secondary index. This design is not
+        ideal and we should reconsider this when revamping index swapping.
         """
         if fields is None and user_fields is None:
             logger.warning(
@@ -679,13 +689,6 @@ class VespaIndex(DocumentIndex):
                 f"Bug: Tenant ID mismatch. Expected {tenant_state.tenant_id}, got {tenant_id}."
             )
 
-        vespa_document_index = VespaDocumentIndex(
-            index_name=self.index_name,
-            tenant_state=tenant_state,
-            large_chunks_enabled=self.large_chunks_enabled,
-            httpx_client=self.httpx_client,
-        )
-
         project_ids: set[int] | None = None
         if user_fields is not None and user_fields.user_projects is not None:
             project_ids = set(user_fields.user_projects)
@@ -705,7 +708,20 @@ class VespaIndex(DocumentIndex):
             persona_ids=persona_ids,
         )
 
-        vespa_document_index.update([update_request])
+        indices = [self.index_name]
+        if self.secondary_index_name:
+            indices.append(self.secondary_index_name)
+
+        for index_name in indices:
+            vespa_document_index = VespaDocumentIndex(
+                index_name=index_name,
+                tenant_state=tenant_state,
+                large_chunks_enabled=self.index_to_large_chunks_enabled.get(
+                    index_name, False
+                ),
+                httpx_client=self.httpx_client,
+            )
+            vespa_document_index.update([update_request])
 
     def delete_single(
         self,
@@ -714,6 +730,11 @@ class VespaIndex(DocumentIndex):
         tenant_id: str,
         chunk_count: int | None,
     ) -> int:
+        """
+        NOTE: Remember to handle the secondary index here. There is no separate
+        pipeline for deleting chunks in the secondary index. This design is not
+        ideal and we should reconsider this when revamping index swapping.
+        """
         tenant_state = TenantState(
             tenant_id=get_current_tenant_id(),
             multitenant=MULTI_TENANT,
@@ -726,13 +747,25 @@ class VespaIndex(DocumentIndex):
             raise ValueError(
                 f"Bug: Tenant ID mismatch. Expected {tenant_state.tenant_id}, got {tenant_id}."
             )
-        vespa_document_index = VespaDocumentIndex(
-            index_name=self.index_name,
-            tenant_state=tenant_state,
-            large_chunks_enabled=self.large_chunks_enabled,
-            httpx_client=self.httpx_client,
-        )
-        return vespa_document_index.delete(document_id=doc_id, chunk_count=chunk_count)
+        indices = [self.index_name]
+        if self.secondary_index_name:
+            indices.append(self.secondary_index_name)
+
+        total_chunks_deleted = 0
+        for index_name in indices:
+            vespa_document_index = VespaDocumentIndex(
+                index_name=index_name,
+                tenant_state=tenant_state,
+                large_chunks_enabled=self.index_to_large_chunks_enabled.get(
+                    index_name, False
+                ),
+                httpx_client=self.httpx_client,
+            )
+            total_chunks_deleted += vespa_document_index.delete(
+                document_id=doc_id, chunk_count=chunk_count
+            )
+
+        return total_chunks_deleted
 
     def id_based_retrieval(
         self,

backend/onyx/document_index/vespa/shared_utils/utils.py

@@ -52,7 +52,9 @@ def replace_invalid_doc_id_characters(text: str) -> str:
     return text.replace("'", "_")
 
 
-def get_vespa_http_client(no_timeout: bool = False, http2: bool = True) -> httpx.Client:
+def get_vespa_http_client(
+    no_timeout: bool = False, http2: bool = True, timeout: int | None = None
+) -> httpx.Client:
     """
     Configures and returns an HTTP client for communicating with Vespa,
     including authentication if needed.
@@ -64,7 +66,7 @@ def get_vespa_http_client(no_timeout: bool = False, http2: bool = True) -> httpx
             else None
         ),
         verify=False if not MANAGED_VESPA else True,
-        timeout=None if no_timeout else VESPA_REQUEST_TIMEOUT,
+        timeout=None if no_timeout else (timeout or VESPA_REQUEST_TIMEOUT),
         http2=http2,
     )
 

backend/onyx/document_index/vespa/shared_utils/vespa_request_builders.py

@@ -23,11 +23,8 @@ from shared_configs.configs import MULTI_TENANT
 logger = setup_logger()
 
 
-def build_tenant_id_filter(tenant_id: str, include_trailing_and: bool = False) -> str:
-    filter_str = f'({TENANT_ID} contains "{tenant_id}")'
-    if include_trailing_and:
-        filter_str += " and "
-    return filter_str
+def build_tenant_id_filter(tenant_id: str) -> str:
+    return f'({TENANT_ID} contains "{tenant_id}")'
 
 
 def build_vespa_filters(
@@ -37,30 +34,22 @@ def build_vespa_filters(
     remove_trailing_and: bool = False,  # Set to True when using as a complete Vespa query
 ) -> str:
     def _build_or_filters(key: str, vals: list[str] | None) -> str:
-        """For string-based 'contains' filters, e.g. WSET fields or array<string> fields."""
+        """For string-based 'contains' filters, e.g. WSET fields or array<string> fields.
+        Returns a bare clause like '(key contains "v1" or key contains "v2")' or ""."""
         if not key or not vals:
             return ""
         eq_elems = [f'{key} contains "{val}"' for val in vals if val]
         if not eq_elems:
             return ""
-        or_clause = " or ".join(eq_elems)
-        return f"({or_clause}) and "
+        return f"({' or '.join(eq_elems)})"
 
     def _build_int_or_filters(key: str, vals: list[int] | None) -> str:
-        """
-        For an integer field filter.
-        If vals is not None, we want *only* docs whose key matches one of vals.
-        """
-        # If `vals` is None => skip the filter entirely
+        """For an integer field filter.
+        Returns a bare clause or ""."""
         if vals is None or not vals:
             return ""
-
-        # Otherwise build the OR filter
         eq_elems = [f"{key} = {val}" for val in vals]
-        or_clause = " or ".join(eq_elems)
-        result = f"({or_clause}) and "
-
-        return result
+        return f"({' or '.join(eq_elems)})"
 
     def _build_kg_filter(
         kg_entities: list[str] | None,
@@ -73,16 +62,12 @@ def build_vespa_filters(
         combined_filter_parts = []
 
         def _build_kge(entity: str) -> str:
-            # TYPE-SUBTYPE::ID -> "TYPE-SUBTYPE::ID"
-            # TYPE-SUBTYPE::*  -> ({prefix: true}"TYPE-SUBTYPE")
-            # TYPE::*          -> ({prefix: true}"TYPE")
             GENERAL = "::*"
             if entity.endswith(GENERAL):
                 return f'({{prefix: true}}"{entity.split(GENERAL, 1)[0]}")'
             else:
                 return f'"{entity}"'
 
-        # OR the entities (give new design)
         if kg_entities:
             filter_parts = []
             for kg_entity in kg_entities:
@@ -104,8 +89,7 @@ def build_vespa_filters(
 
         # TODO: remove kg terms entirely from prompts and codebase
 
-        # AND the combined filter parts
-        return f"({' and '.join(combined_filter_parts)}) and "
+        return f"({' and '.join(combined_filter_parts)})"
 
     def _build_kg_source_filters(
         kg_sources: list[str] | None,
@@ -114,16 +98,14 @@ def build_vespa_filters(
             return ""
 
         source_phrases = [f'{DOCUMENT_ID} contains "{source}"' for source in kg_sources]
-
-        return f"({' or '.join(source_phrases)}) and "
+        return f"({' or '.join(source_phrases)})"
 
     def _build_kg_chunk_id_zero_only_filter(
         kg_chunk_id_zero_only: bool,
     ) -> str:
         if not kg_chunk_id_zero_only:
             return ""
-
-        return "(chunk_id = 0 ) and "
+        return "(chunk_id = 0)"
 
     def _build_time_filter(
         cutoff: datetime | None,
@@ -135,8 +117,8 @@ def build_vespa_filters(
         cutoff_secs = int(cutoff.timestamp())
 
         if include_untimed:
-            return f"!({DOC_UPDATED_AT} < {cutoff_secs}) and "
-        return f"({DOC_UPDATED_AT} >= {cutoff_secs}) and "
+            return f"!({DOC_UPDATED_AT} < {cutoff_secs})"
+        return f"({DOC_UPDATED_AT} >= {cutoff_secs})"
 
     def _build_user_project_filter(
         project_id: int | None,
@@ -147,8 +129,7 @@ def build_vespa_filters(
             pid = int(project_id)
         except Exception:
             return ""
-        # Vespa YQL 'contains' expects a string literal; quote the integer
-        return f'({USER_PROJECT} contains "{pid}") and '
+        return f'({USER_PROJECT} contains "{pid}")'
 
     def _build_persona_filter(
         persona_id: int | None,
@@ -160,73 +141,94 @@ def build_vespa_filters(
         except Exception:
             logger.warning(f"Invalid persona ID: {persona_id}")
             return ""
-        return f'({PERSONAS} contains "{pid}") and '
+        return f'({PERSONAS} contains "{pid}")'
 
-    # Start building the filter string
-    filter_str = f"!({HIDDEN}=true) and " if not include_hidden else ""
+    def _append(parts: list[str], clause: str) -> None:
+        if clause:
+            parts.append(clause)
+
+    # Collect all top-level filter clauses, then join with " and " at the end.
+    filter_parts: list[str] = []
+
+    if not include_hidden:
+        filter_parts.append(f"!({HIDDEN}=true)")
 
     # TODO: add error condition if MULTI_TENANT and no tenant_id filter is set
-    # If running in multi-tenant mode
     if filters.tenant_id and MULTI_TENANT:
-        filter_str += build_tenant_id_filter(
-            filters.tenant_id, include_trailing_and=True
-        )
+        filter_parts.append(build_tenant_id_filter(filters.tenant_id))
 
     # ACL filters
     if filters.access_control_list is not None:
-        filter_str += _build_or_filters(
-            ACCESS_CONTROL_LIST, filters.access_control_list
+        _append(
+            filter_parts,
+            _build_or_filters(ACCESS_CONTROL_LIST, filters.access_control_list),
         )
 
     # Source type filters
     source_strs = (
         [s.value for s in filters.source_type] if filters.source_type else None
     )
-    filter_str += _build_or_filters(SOURCE_TYPE, source_strs)
+    _append(filter_parts, _build_or_filters(SOURCE_TYPE, source_strs))
 
     # Tag filters
     tag_attributes = None
     if filters.tags:
-        # build e.g. "tag_key|tag_value"
         tag_attributes = [
             f"{tag.tag_key}{INDEX_SEPARATOR}{tag.tag_value}" for tag in filters.tags
         ]
-    filter_str += _build_or_filters(METADATA_LIST, tag_attributes)
+    _append(filter_parts, _build_or_filters(METADATA_LIST, tag_attributes))
 
-    # Document sets
-    filter_str += _build_or_filters(DOCUMENT_SETS, filters.document_set)
+    # Knowledge scope: explicit knowledge attachments (document_sets,
+    # user_file_ids) restrict what an assistant can see.  When none are
+    # set, the assistant can see everything.
+    #
+    # project_id / persona_id are additive: they make overflowing user
+    # files findable in Vespa but must NOT trigger the restriction on
+    # their own (an agent with no explicit knowledge should search
+    # everything).
+    knowledge_scope_parts: list[str] = []
+
+    _append(
+        knowledge_scope_parts, _build_or_filters(DOCUMENT_SETS, filters.document_set)
+    )
 
-    # Convert UUIDs to strings for user_file_ids
     user_file_ids_str = (
         [str(uuid) for uuid in filters.user_file_ids] if filters.user_file_ids else None
     )
-    filter_str += _build_or_filters(DOCUMENT_ID, user_file_ids_str)
+    _append(knowledge_scope_parts, _build_or_filters(DOCUMENT_ID, user_file_ids_str))
 
-    # User project filter (array<int> attribute membership)
-    filter_str += _build_user_project_filter(filters.project_id)
+    # Only include project/persona scopes when an explicit knowledge
+    # restriction is already in effect — they widen the scope to also
+    # cover overflowing user files but never restrict on their own.
+    if knowledge_scope_parts:
+        _append(knowledge_scope_parts, _build_user_project_filter(filters.project_id))
+        _append(knowledge_scope_parts, _build_persona_filter(filters.persona_id))
 
-    # Persona filter (array<int> attribute membership)
-    filter_str += _build_persona_filter(filters.persona_id)
+    if len(knowledge_scope_parts) > 1:
+        filter_parts.append("(" + " or ".join(knowledge_scope_parts) + ")")
+    elif len(knowledge_scope_parts) == 1:
+        filter_parts.append(knowledge_scope_parts[0])
 
     # Time filter
-    filter_str += _build_time_filter(filters.time_cutoff)
+    _append(filter_parts, _build_time_filter(filters.time_cutoff))
 
     # # Knowledge Graph Filters
-    # filter_str += _build_kg_filter(
+    # _append(filter_parts, _build_kg_filter(
     #     kg_entities=filters.kg_entities,
     #     kg_relationships=filters.kg_relationships,
     #     kg_terms=filters.kg_terms,
-    # )
+    # ))
 
-    # filter_str += _build_kg_source_filters(filters.kg_sources)
+    # _append(filter_parts, _build_kg_source_filters(filters.kg_sources))
 
-    # filter_str += _build_kg_chunk_id_zero_only_filter(
+    # _append(filter_parts, _build_kg_chunk_id_zero_only_filter(
     #     filters.kg_chunk_id_zero_only or False
-    # )
+    # ))
 
-    # Trim trailing " and "
-    if remove_trailing_and and filter_str.endswith(" and "):
-        filter_str = filter_str[:-5]
+    filter_str = " and ".join(filter_parts)
+
+    if filter_str and not remove_trailing_and:
+        filter_str += " and "
 
     return filter_str
 

backend/onyx/error_handling/error_codes.py

@@ -0,0 +1,101 @@
+"""
+Standardized error codes for the Onyx backend.
+
+Usage:
+    from onyx.error_handling.error_codes import OnyxErrorCode
+    from onyx.error_handling.exceptions import OnyxError
+
+    raise OnyxError(OnyxErrorCode.UNAUTHENTICATED, "Token expired")
+"""
+
+from enum import Enum
+
+
+class OnyxErrorCode(Enum):
+    """
+    Each member is a tuple of (error_code_string, http_status_code).
+
+    The error_code_string is a stable, machine-readable identifier that
+    API consumers can match on. The http_status_code is the default HTTP
+    status to return.
+    """
+
+    # ------------------------------------------------------------------
+    # Authentication (401)
+    # ------------------------------------------------------------------
+    UNAUTHENTICATED = ("UNAUTHENTICATED", 401)
+    INVALID_TOKEN = ("INVALID_TOKEN", 401)
+    TOKEN_EXPIRED = ("TOKEN_EXPIRED", 401)
+    CSRF_FAILURE = ("CSRF_FAILURE", 403)
+
+    # ------------------------------------------------------------------
+    # Authorization (403)
+    # ------------------------------------------------------------------
+    UNAUTHORIZED = ("UNAUTHORIZED", 403)
+    INSUFFICIENT_PERMISSIONS = ("INSUFFICIENT_PERMISSIONS", 403)
+    ADMIN_ONLY = ("ADMIN_ONLY", 403)
+    EE_REQUIRED = ("EE_REQUIRED", 403)
+
+    # ------------------------------------------------------------------
+    # Validation / Bad Request (400)
+    # ------------------------------------------------------------------
+    VALIDATION_ERROR = ("VALIDATION_ERROR", 400)
+    INVALID_INPUT = ("INVALID_INPUT", 400)
+    MISSING_REQUIRED_FIELD = ("MISSING_REQUIRED_FIELD", 400)
+
+    # ------------------------------------------------------------------
+    # Not Found (404)
+    # ------------------------------------------------------------------
+    NOT_FOUND = ("NOT_FOUND", 404)
+    CONNECTOR_NOT_FOUND = ("CONNECTOR_NOT_FOUND", 404)
+    CREDENTIAL_NOT_FOUND = ("CREDENTIAL_NOT_FOUND", 404)
+    PERSONA_NOT_FOUND = ("PERSONA_NOT_FOUND", 404)
+    DOCUMENT_NOT_FOUND = ("DOCUMENT_NOT_FOUND", 404)
+    SESSION_NOT_FOUND = ("SESSION_NOT_FOUND", 404)
+    USER_NOT_FOUND = ("USER_NOT_FOUND", 404)
+
+    # ------------------------------------------------------------------
+    # Conflict (409)
+    # ------------------------------------------------------------------
+    CONFLICT = ("CONFLICT", 409)
+    DUPLICATE_RESOURCE = ("DUPLICATE_RESOURCE", 409)
+
+    # ------------------------------------------------------------------
+    # Rate Limiting / Quotas (429 / 402)
+    # ------------------------------------------------------------------
+    RATE_LIMITED = ("RATE_LIMITED", 429)
+    SEAT_LIMIT_EXCEEDED = ("SEAT_LIMIT_EXCEEDED", 402)
+
+    # ------------------------------------------------------------------
+    # Connector / Credential Errors (400-range)
+    # ------------------------------------------------------------------
+    CONNECTOR_VALIDATION_FAILED = ("CONNECTOR_VALIDATION_FAILED", 400)
+    CREDENTIAL_INVALID = ("CREDENTIAL_INVALID", 400)
+    CREDENTIAL_EXPIRED = ("CREDENTIAL_EXPIRED", 401)
+
+    # ------------------------------------------------------------------
+    # Server Errors (5xx)
+    # ------------------------------------------------------------------
+    INTERNAL_ERROR = ("INTERNAL_ERROR", 500)
+    NOT_IMPLEMENTED = ("NOT_IMPLEMENTED", 501)
+    SERVICE_UNAVAILABLE = ("SERVICE_UNAVAILABLE", 503)
+    BAD_GATEWAY = ("BAD_GATEWAY", 502)
+    LLM_PROVIDER_ERROR = ("LLM_PROVIDER_ERROR", 502)
+    GATEWAY_TIMEOUT = ("GATEWAY_TIMEOUT", 504)
+
+    def __init__(self, code: str, status_code: int) -> None:
+        self.code = code
+        self.status_code = status_code
+
+    def detail(self, message: str | None = None) -> dict[str, str]:
+        """Build a structured error detail dict.
+
+        Returns a dict like:
+            {"error_code": "UNAUTHENTICATED", "message": "Token expired"}
+
+        If no message is supplied, the error code itself is used as the message.
+        """
+        return {
+            "error_code": self.code,
+            "message": message or self.code,
+        }

backend/onyx/error_handling/exceptions.py

@@ -0,0 +1,83 @@
+"""OnyxError — the single exception type for all Onyx business errors.
+
+Raise ``OnyxError`` instead of ``HTTPException`` in business code.  A global
+FastAPI exception handler (registered via ``register_onyx_exception_handlers``)
+converts it into a JSON response with the standard
+``{"error_code": "...", "message": "..."}`` shape.
+
+Usage::
+
+    from onyx.error_handling.error_codes import OnyxErrorCode
+    from onyx.error_handling.exceptions import OnyxError
+
+    raise OnyxError(OnyxErrorCode.NOT_FOUND, "Session not found")
+
+For upstream errors with a dynamic HTTP status (e.g. billing service),
+use ``status_code_override``::
+
+    raise OnyxError(
+        OnyxErrorCode.BAD_GATEWAY,
+        detail,
+        status_code_override=upstream_status,
+    )
+"""
+
+from fastapi import FastAPI
+from fastapi import Request
+from fastapi.responses import JSONResponse
+
+from onyx.error_handling.error_codes import OnyxErrorCode
+from onyx.utils.logger import setup_logger
+
+logger = setup_logger()
+
+
+class OnyxError(Exception):
+    """Structured error that maps to a specific ``OnyxErrorCode``.
+
+    Attributes:
+        error_code: The ``OnyxErrorCode`` enum member.
+        message: Human-readable message (defaults to the error code string).
+        status_code: HTTP status — either overridden or from the error code.
+    """
+
+    def __init__(
+        self,
+        error_code: OnyxErrorCode,
+        message: str | None = None,
+        *,
+        status_code_override: int | None = None,
+    ) -> None:
+        resolved_message = message or error_code.code
+        super().__init__(resolved_message)
+        self.error_code = error_code
+        self.message = resolved_message
+        self._status_code_override = status_code_override
+
+    @property
+    def status_code(self) -> int:
+        return self._status_code_override or self.error_code.status_code
+
+
+def register_onyx_exception_handlers(app: FastAPI) -> None:
+    """Register a global handler that converts ``OnyxError`` to JSON responses.
+
+    Must be called *after* the app is created but *before* it starts serving.
+    The handler logs at WARNING for 4xx and ERROR for 5xx.
+    """
+
+    @app.exception_handler(OnyxError)
+    async def _handle_onyx_error(
+        request: Request,  # noqa: ARG001
+        exc: OnyxError,
+    ) -> JSONResponse:
+        status_code = exc.status_code
+        if status_code >= 500:
+            logger.error(f"OnyxError {exc.error_code.code}: {exc.message}")
+        elif status_code >= 400:
+            logger.warning(f"OnyxError {exc.error_code.code}: {exc.message}")
+
+        return JSONResponse(
+            status_code=status_code,
+            content=exc.error_code.detail(exc.message),
+        )

backend/onyx/federated_connectors/oauth_utils.py

@@ -4,39 +4,33 @@ import base64
 import json
 import uuid
 from typing import Any
-from typing import cast
-from typing import Dict
-from typing import Optional
 
+from onyx.cache.factory import get_cache_backend
 from onyx.configs.app_configs import WEB_DOMAIN
-from onyx.redis.redis_pool import get_redis_client
 from onyx.utils.logger import setup_logger
 
 logger = setup_logger()
 
-# Redis key prefix for OAuth state
 OAUTH_STATE_PREFIX = "federated_oauth"
-# Default TTL for OAuth state (5 minutes)
-OAUTH_STATE_TTL = 300
+OAUTH_STATE_TTL = 300  # 5 minutes
 
 
 class OAuthSession:
-    """Represents an OAuth session stored in Redis."""
+    """Represents an OAuth session stored in the cache backend."""
 
     def __init__(
         self,
         federated_connector_id: int,
         user_id: str,
-        redirect_uri: Optional[str] = None,
-        additional_data: Optional[Dict[str, Any]] = None,
+        redirect_uri: str | None = None,
+        additional_data: dict[str, Any] | None = None,
     ):
         self.federated_connector_id = federated_connector_id
         self.user_id = user_id
         self.redirect_uri = redirect_uri
         self.additional_data = additional_data or {}
 
-    def to_dict(self) -> Dict[str, Any]:
-        """Convert to dictionary for Redis storage."""
+    def to_dict(self) -> dict[str, Any]:
         return {
             "federated_connector_id": self.federated_connector_id,
             "user_id": self.user_id,
@@ -45,8 +39,7 @@ class OAuthSession:
         }
 
     @classmethod
-    def from_dict(cls, data: Dict[str, Any]) -> "OAuthSession":
-        """Create from dictionary retrieved from Redis."""
+    def from_dict(cls, data: dict[str, Any]) -> "OAuthSession":
         return cls(
             federated_connector_id=data["federated_connector_id"],
             user_id=data["user_id"],
@@ -58,31 +51,27 @@ class OAuthSession:
 def generate_oauth_state(
     federated_connector_id: int,
     user_id: str,
-    redirect_uri: Optional[str] = None,
-    additional_data: Optional[Dict[str, Any]] = None,
+    redirect_uri: str | None = None,
+    additional_data: dict[str, Any] | None = None,
     ttl: int = OAUTH_STATE_TTL,
 ) -> str:
     """
-    Generate a secure state parameter and store session data in Redis.
+    Generate a secure state parameter and store session data in the cache backend.
 
     Args:
         federated_connector_id: ID of the federated connector
         user_id: ID of the user initiating OAuth
         redirect_uri: Optional redirect URI after OAuth completion
         additional_data: Any additional data to store with the session
-        ttl: Time-to-live in seconds for the Redis key
+        ttl: Time-to-live in seconds for the cache key
 
     Returns:
         Base64-encoded state parameter
     """
     # Generate a random UUID for the state
     state_uuid = uuid.uuid4()
+    state_b64 = base64.urlsafe_b64encode(state_uuid.bytes).decode("utf-8").rstrip("=")
 
-    # Convert UUID to base64 for URL-safe state parameter
-    state_bytes = state_uuid.bytes
-    state_b64 = base64.urlsafe_b64encode(state_bytes).decode("utf-8").rstrip("=")
-
-    # Create session object
     session = OAuthSession(
         federated_connector_id=federated_connector_id,
         user_id=user_id,
@@ -90,15 +79,9 @@ def generate_oauth_state(
         additional_data=additional_data,
     )
 
-    # Store in Redis with TTL
-    redis_client = get_redis_client()
-    redis_key = f"{OAUTH_STATE_PREFIX}:{state_uuid}"
-
-    redis_client.set(
-        redis_key,
-        json.dumps(session.to_dict()),
-        ex=ttl,
-    )
+    cache = get_cache_backend()
+    cache_key = f"{OAUTH_STATE_PREFIX}:{state_uuid}"
+    cache.set(cache_key, json.dumps(session.to_dict()), ex=ttl)
 
     logger.info(
         f"Generated OAuth state for federated_connector_id={federated_connector_id}, "
@@ -125,18 +108,15 @@ def verify_oauth_state(state: str) -> OAuthSession:
     state_bytes = base64.urlsafe_b64decode(padded_state)
     state_uuid = uuid.UUID(bytes=state_bytes)
 
-    # Look up in Redis
-    redis_client = get_redis_client()
-    redis_key = f"{OAUTH_STATE_PREFIX}:{state_uuid}"
+    cache = get_cache_backend()
+    cache_key = f"{OAUTH_STATE_PREFIX}:{state_uuid}"
 
-    session_data = cast(bytes, redis_client.get(redis_key))
+    session_data = cache.get(cache_key)
     if not session_data:
-        raise ValueError(f"OAuth state not found in Redis: {state}")
+        raise ValueError(f"OAuth state not found: {state}")
 
-    # Delete the key after retrieval (one-time use)
-    redis_client.delete(redis_key)
+    cache.delete(cache_key)
 
-    # Parse and return session
     session_dict = json.loads(session_data)
     return OAuthSession.from_dict(session_dict)
 

backend/onyx/indexing/indexing_pipeline.py

@@ -49,7 +49,6 @@ from onyx.indexing.embedder import IndexingEmbedder
 from onyx.indexing.models import DocAwareChunk
 from onyx.indexing.models import IndexingBatchAdapter
 from onyx.indexing.models import UpdatableChunkData
-from onyx.indexing.postgres_sanitization import sanitize_documents_for_postgres
 from onyx.indexing.vector_db_insertion import write_chunks_to_vector_db_with_backoff
 from onyx.llm.factory import get_default_llm_with_vision
 from onyx.llm.factory import get_llm_for_contextual_rag
@@ -65,6 +64,7 @@ from onyx.prompts.contextual_retrieval import CONTEXTUAL_RAG_PROMPT1
 from onyx.prompts.contextual_retrieval import CONTEXTUAL_RAG_PROMPT2
 from onyx.prompts.contextual_retrieval import DOCUMENT_SUMMARY_PROMPT
 from onyx.utils.logger import setup_logger
+from onyx.utils.postgres_sanitization import sanitize_documents_for_postgres
 from onyx.utils.threadpool_concurrency import run_functions_tuples_in_parallel
 from onyx.utils.timing import log_function_time
 

backend/onyx/key_value_store/store.py

@@ -1,13 +1,11 @@
 import json
 from typing import cast
 
-from redis.client import Redis
-
+from onyx.cache.interface import CacheBackend
 from onyx.db.engine.sql_engine import get_session_with_current_tenant
 from onyx.db.models import KVStore
 from onyx.key_value_store.interface import KeyValueStore
 from onyx.key_value_store.interface import KvKeyNotFoundError
-from onyx.redis.redis_pool import get_redis_client
 from onyx.utils.logger import setup_logger
 from onyx.utils.special_types import JSON_ro
 
@@ -20,22 +18,27 @@ KV_REDIS_KEY_EXPIRATION = 60 * 60 * 24  # 1 Day
 
 
 class PgRedisKVStore(KeyValueStore):
-    def __init__(self, redis_client: Redis | None = None) -> None:
-        # If no redis_client is provided, fall back to the context var
-        if redis_client is not None:
-            self.redis_client = redis_client
-        else:
-            self.redis_client = get_redis_client()
+    def __init__(self, cache: CacheBackend | None = None) -> None:
+        self._cache = cache
+
+    def _get_cache(self) -> CacheBackend:
+        if self._cache is None:
+            from onyx.cache.factory import get_cache_backend
+
+            self._cache = get_cache_backend()
+        return self._cache
 
     def store(self, key: str, val: JSON_ro, encrypt: bool = False) -> None:
-        # Not encrypted in Redis, but encrypted in Postgres
+        # Not encrypted in Cache backend (typically Redis), but encrypted in Postgres
         try:
-            self.redis_client.set(
+            self._get_cache().set(
                 REDIS_KEY_PREFIX + key, json.dumps(val), ex=KV_REDIS_KEY_EXPIRATION
             )
         except Exception as e:
-            # Fallback gracefully to Postgres if Redis fails
-            logger.error(f"Failed to set value in Redis for key '{key}': {str(e)}")
+            # Fallback gracefully to Postgres if Cache backend fails
+            logger.error(
+                f"Failed to set value in Cache backend for key '{key}': {str(e)}"
+            )
 
         encrypted_val = val if encrypt else None
         plain_val = val if not encrypt else None
@@ -53,16 +56,12 @@ class PgRedisKVStore(KeyValueStore):
     def load(self, key: str, refresh_cache: bool = False) -> JSON_ro:
         if not refresh_cache:
             try:
-                redis_value = self.redis_client.get(REDIS_KEY_PREFIX + key)
-                if redis_value:
-                    if not isinstance(redis_value, bytes):
-                        raise ValueError(
-                            f"Redis value for key '{key}' is not a bytes object"
-                        )
-                    return json.loads(redis_value.decode("utf-8"))
+                cached = self._get_cache().get(REDIS_KEY_PREFIX + key)
+                if cached is not None:
+                    return json.loads(cached.decode("utf-8"))
             except Exception as e:
                 logger.error(
-                    f"Failed to get value from Redis for key '{key}': {str(e)}"
+                    f"Failed to get value from cache for key '{key}': {str(e)}"
                 )
 
         with get_session_with_current_tenant() as db_session:
@@ -79,21 +78,21 @@ class PgRedisKVStore(KeyValueStore):
                 value = None
 
             try:
-                self.redis_client.set(
+                self._get_cache().set(
                     REDIS_KEY_PREFIX + key,
                     json.dumps(value),
                     ex=KV_REDIS_KEY_EXPIRATION,
                 )
             except Exception as e:
-                logger.error(f"Failed to set value in Redis for key '{key}': {str(e)}")
+                logger.error(f"Failed to set value in cache for key '{key}': {str(e)}")
 
             return cast(JSON_ro, value)
 
     def delete(self, key: str) -> None:
         try:
-            self.redis_client.delete(REDIS_KEY_PREFIX + key)
+            self._get_cache().delete(REDIS_KEY_PREFIX + key)
         except Exception as e:
-            logger.error(f"Failed to delete value from Redis for key '{key}': {str(e)}")
+            logger.error(f"Failed to delete value from cache for key '{key}': {str(e)}")
 
         with get_session_with_current_tenant() as db_session:
             result = db_session.query(KVStore).filter_by(key=key).delete()

backend/onyx/llm/constants.py

@@ -22,6 +22,7 @@ class LlmProviderNames(str, Enum):
     OPENROUTER = "openrouter"
     AZURE = "azure"
     OLLAMA_CHAT = "ollama_chat"
+    LM_STUDIO = "lm_studio"
     MISTRAL = "mistral"
     LITELLM_PROXY = "litellm_proxy"
 
@@ -41,6 +42,7 @@ WELL_KNOWN_PROVIDER_NAMES = [
     LlmProviderNames.OPENROUTER,
     LlmProviderNames.AZURE,
     LlmProviderNames.OLLAMA_CHAT,
+    LlmProviderNames.LM_STUDIO,
 ]
 
 
@@ -56,6 +58,7 @@ PROVIDER_DISPLAY_NAMES: dict[str, str] = {
     LlmProviderNames.AZURE: "Azure",
     "ollama": "Ollama",
     LlmProviderNames.OLLAMA_CHAT: "Ollama",
+    LlmProviderNames.LM_STUDIO: "LM Studio",
     "groq": "Groq",
     "anyscale": "Anyscale",
     "deepseek": "DeepSeek",
@@ -103,6 +106,7 @@ AGGREGATOR_PROVIDERS: set[str] = {
     LlmProviderNames.BEDROCK_CONVERSE,
     LlmProviderNames.OPENROUTER,
     LlmProviderNames.OLLAMA_CHAT,
+    LlmProviderNames.LM_STUDIO,
     LlmProviderNames.VERTEX_AI,
     LlmProviderNames.AZURE,
 }

backend/onyx/llm/factory.py

@@ -20,7 +20,9 @@ from onyx.llm.multi_llm import LitellmLLM
 from onyx.llm.override_models import LLMOverride
 from onyx.llm.utils import get_max_input_tokens_from_llm_provider
 from onyx.llm.utils import model_supports_image_input
-from onyx.llm.well_known_providers.constants import OLLAMA_API_KEY_CONFIG_KEY
+from onyx.llm.well_known_providers.constants import (
+    PROVIDERS_WITH_SPECIAL_API_KEY_HANDLING,
+)
 from onyx.natural_language_processing.utils import get_tokenizer
 from onyx.server.manage.llm.models import LLMProviderView
 from onyx.utils.headers import build_llm_extra_headers
@@ -32,14 +34,18 @@ logger = setup_logger()
 def _build_provider_extra_headers(
     provider: str, custom_config: dict[str, str] | None
 ) -> dict[str, str]:
-    if provider == LlmProviderNames.OLLAMA_CHAT and custom_config:
-        raw_api_key = custom_config.get(OLLAMA_API_KEY_CONFIG_KEY)
-        api_key = raw_api_key.strip() if raw_api_key else None
+    if provider in PROVIDERS_WITH_SPECIAL_API_KEY_HANDLING and custom_config:
+        raw = custom_config.get(PROVIDERS_WITH_SPECIAL_API_KEY_HANDLING[provider])
+        api_key = raw.strip() if raw else None
         if not api_key:
             return {}
-        if not api_key.lower().startswith("bearer "):
-            api_key = f"Bearer {api_key}"
-        return {"Authorization": api_key}
+        return {
+            "Authorization": (
+                api_key
+                if api_key.lower().startswith("bearer ")
+                else f"Bearer {api_key}"
+            )
+        }
 
     # Passing these will put Onyx on the OpenRouter leaderboard
     elif provider == LlmProviderNames.OPENROUTER:

backend/onyx/llm/litellm_singleton/monkey_patches.py

@@ -67,6 +67,18 @@ Status checked against LiteLLM v1.81.6-nightly (2026-02-02):
    STATUS: STILL NEEDED - litellm_core_utils/litellm_logging.py lines 3185-3199 set
            usage as a dict with chat completion format instead of keeping it as
            ResponseAPIUsage. Our patch creates a deep copy before modification.
+
+7. Responses API metadata=None TypeError (_patch_responses_metadata_none):
+   - LiteLLM's @client decorator wrapper in utils.py uses kwargs.get("metadata", {})
+     to check for router calls, but when metadata is explicitly None (key exists with
+     value None), the default {} is not used
+   - This causes "argument of type 'NoneType' is not iterable" TypeError which swallows
+     the real exception (e.g. AuthenticationError for wrong API key)
+   - Surfaces as: APIConnectionError: OpenAIException - argument of type 'NoneType' is
+     not iterable
+   STATUS: STILL NEEDED - litellm/utils.py wrapper function (line 1721) does not guard
+           against metadata being explicitly None. Triggered when Responses API bridge
+           passes **litellm_params containing metadata=None.
 """
 
 import time
@@ -725,6 +737,44 @@ def _patch_logging_assembled_streaming_response() -> None:
     LiteLLMLoggingObj._get_assembled_streaming_response = _patched_get_assembled_streaming_response  # type: ignore[method-assign]
 
 
+def _patch_responses_metadata_none() -> None:
+    """
+    Patches litellm.responses to normalize metadata=None to metadata={} in kwargs.
+
+    LiteLLM's @client decorator wrapper in utils.py (line 1721) does:
+        _is_litellm_router_call = "model_group" in kwargs.get("metadata", {})
+    When metadata is explicitly None in kwargs, kwargs.get("metadata", {}) returns
+    None (the key exists, so the default is not used), causing:
+        TypeError: argument of type 'NoneType' is not iterable
+
+    This swallows the real exception (e.g. AuthenticationError) and surfaces as:
+        APIConnectionError: OpenAIException - argument of type 'NoneType' is not iterable
+
+    This happens when the Responses API bridge calls litellm.responses() with
+    **litellm_params which may contain metadata=None.
+
+    STATUS: STILL NEEDED - litellm/utils.py wrapper function uses kwargs.get("metadata", {})
+            which does not guard against metadata being explicitly None. Same pattern exists
+            on line 1407 for async path.
+    """
+    import litellm as _litellm
+    from functools import wraps
+
+    original_responses = _litellm.responses
+
+    if getattr(original_responses, "_metadata_patched", False):
+        return
+
+    @wraps(original_responses)
+    def _patched_responses(*args: Any, **kwargs: Any) -> Any:
+        if kwargs.get("metadata") is None:
+            kwargs["metadata"] = {}
+        return original_responses(*args, **kwargs)
+
+    _patched_responses._metadata_patched = True  # type: ignore[attr-defined]
+    _litellm.responses = _patched_responses
+
+
 def apply_monkey_patches() -> None:
     """
     Apply all necessary monkey patches to LiteLLM for compatibility.
@@ -736,6 +786,7 @@ def apply_monkey_patches() -> None:
     - Patching AzureOpenAIResponsesAPIConfig.should_fake_stream to enable native streaming
     - Patching ResponsesAPIResponse.model_construct to fix usage format in all code paths
     - Patching LiteLLMLoggingObj._get_assembled_streaming_response to avoid mutating original response
+    - Patching litellm.responses to fix metadata=None causing TypeError in error handling
     """
     _patch_ollama_chunk_parser()
     _patch_openai_responses_parallel_tool_calls()
@@ -743,3 +794,4 @@ def apply_monkey_patches() -> None:
     _patch_azure_responses_should_fake_stream()
     _patch_responses_api_usage_format()
     _patch_logging_assembled_streaming_response()
+    _patch_responses_metadata_none()

backend/onyx/llm/model_metadata_enrichments.json

@@ -1512,6 +1512,10 @@
     "display_name": "Claude Opus 4.5",
     "model_vendor": "anthropic"
   },
+  "claude-opus-4-6": {
+    "display_name": "Claude Opus 4.6",
+    "model_vendor": "anthropic"
+  },
   "claude-opus-4-5-20251101": {
     "display_name": "Claude Opus 4.5",
     "model_vendor": "anthropic",
@@ -1526,6 +1530,10 @@
     "display_name": "Claude Sonnet 4.5",
     "model_vendor": "anthropic"
   },
+  "claude-sonnet-4-6": {
+    "display_name": "Claude Sonnet 4.6",
+    "model_vendor": "anthropic"
+  },
   "claude-sonnet-4-5-20250929": {
     "display_name": "Claude Sonnet 4.5",
     "model_vendor": "anthropic",
@@ -2516,6 +2524,10 @@
     "model_vendor": "openai",
     "model_version": "2025-10-06"
   },
+  "gpt-5.4": {
+    "display_name": "GPT-5.4",
+    "model_vendor": "openai"
+  },
   "gpt-5.2-pro-2025-12-11": {
     "display_name": "GPT-5.2 Pro",
     "model_vendor": "openai",

backend/onyx/llm/multi_llm.py

@@ -42,6 +42,7 @@ from onyx.llm.well_known_providers.constants import AWS_SECRET_ACCESS_KEY_KWARG
 from onyx.llm.well_known_providers.constants import (
     AWS_SECRET_ACCESS_KEY_KWARG_ENV_VAR_FORMAT,
 )
+from onyx.llm.well_known_providers.constants import LM_STUDIO_API_KEY_CONFIG_KEY
 from onyx.llm.well_known_providers.constants import OLLAMA_API_KEY_CONFIG_KEY
 from onyx.llm.well_known_providers.constants import VERTEX_CREDENTIALS_FILE_KWARG
 from onyx.llm.well_known_providers.constants import (
@@ -92,6 +93,98 @@ def _prompt_to_dicts(prompt: LanguageModelInput) -> list[dict[str, Any]]:
     return [prompt.model_dump(exclude_none=True)]
 
 
+def _normalize_content(raw: Any) -> str:
+    """Normalize a message content field to a plain string.
+
+    Content can be a string, None, or a list of content-block dicts
+    (e.g. [{"type": "text", "text": "..."}]).
+    """
+    if raw is None:
+        return ""
+    if isinstance(raw, str):
+        return raw
+    if isinstance(raw, list):
+        return "\n".join(
+            block.get("text", "") if isinstance(block, dict) else str(block)
+            for block in raw
+        )
+    return str(raw)
+
+
+def _strip_tool_content_from_messages(
+    messages: list[dict[str, Any]],
+) -> list[dict[str, Any]]:
+    """Convert tool-related messages to plain text.
+
+    Bedrock's Converse API requires toolConfig when messages contain
+    toolUse/toolResult content blocks. When no tools are provided for the
+    current request, we must convert any tool-related history into plain text
+    to avoid the "toolConfig field must be defined" error.
+
+    This is the same approach used by _OllamaHistoryMessageFormatter.
+    """
+    result: list[dict[str, Any]] = []
+    for msg in messages:
+        role = msg.get("role")
+        tool_calls = msg.get("tool_calls")
+
+        if role == "assistant" and tool_calls:
+            # Convert structured tool calls to text representation
+            tool_call_lines = []
+            for tc in tool_calls:
+                func = tc.get("function", {})
+                name = func.get("name", "unknown")
+                args = func.get("arguments", "{}")
+                tc_id = tc.get("id", "")
+                tool_call_lines.append(
+                    f"[Tool Call] name={name} id={tc_id} args={args}"
+                )
+
+            existing_content = _normalize_content(msg.get("content"))
+            parts = (
+                [existing_content] + tool_call_lines
+                if existing_content
+                else tool_call_lines
+            )
+            new_msg = {
+                "role": "assistant",
+                "content": "\n".join(parts),
+            }
+            result.append(new_msg)
+
+        elif role == "tool":
+            # Convert tool response to user message with text content
+            tool_call_id = msg.get("tool_call_id", "")
+            content = _normalize_content(msg.get("content"))
+            tool_result_text = f"[Tool Result] id={tool_call_id}\n{content}"
+            # Merge into previous user message if it is also a converted
+            # tool result to avoid consecutive user messages (Bedrock requires
+            # strict user/assistant alternation).
+            if (
+                result
+                and result[-1]["role"] == "user"
+                and "[Tool Result]" in result[-1].get("content", "")
+            ):
+                result[-1]["content"] += "\n\n" + tool_result_text
+            else:
+                result.append({"role": "user", "content": tool_result_text})
+
+        else:
+            result.append(msg)
+
+    return result
+
+
+def _messages_contain_tool_content(messages: list[dict[str, Any]]) -> bool:
+    """Check if any messages contain tool-related content blocks."""
+    for msg in messages:
+        if msg.get("role") == "tool":
+            return True
+        if msg.get("role") == "assistant" and msg.get("tool_calls"):
+            return True
+    return False
+
+
 def _is_vertex_model_rejecting_output_config(model_name: str) -> bool:
     normalized_model_name = model_name.lower()
     return any(
@@ -157,6 +250,9 @@ class LitellmLLM(LLM):
                 elif model_provider == LlmProviderNames.OLLAMA_CHAT:
                     if k == OLLAMA_API_KEY_CONFIG_KEY:
                         model_kwargs["api_key"] = v
+                elif model_provider == LlmProviderNames.LM_STUDIO:
+                    if k == LM_STUDIO_API_KEY_CONFIG_KEY:
+                        model_kwargs["api_key"] = v
                 elif model_provider == LlmProviderNames.BEDROCK:
                     if k == AWS_REGION_NAME_KWARG:
                         model_kwargs[k] = v
@@ -173,6 +269,19 @@ class LitellmLLM(LLM):
                     elif k == AWS_SECRET_ACCESS_KEY_KWARG_ENV_VAR_FORMAT:
                         model_kwargs[AWS_SECRET_ACCESS_KEY_KWARG] = v
 
+        # LM Studio: LiteLLM defaults to "fake-api-key" when no key is provided,
+        # which LM Studio rejects. Ensure we always pass an explicit key (or empty
+        # string) to prevent LiteLLM from injecting its fake default.
+        if model_provider == LlmProviderNames.LM_STUDIO:
+            model_kwargs.setdefault("api_key", "")
+
+            # Users provide the server root (e.g. http://localhost:1234) but LiteLLM
+            # needs /v1 for OpenAI-compatible calls.
+            if self._api_base is not None:
+                base = self._api_base.rstrip("/")
+                self._api_base = base if base.endswith("/v1") else f"{base}/v1"
+                model_kwargs["api_base"] = self._api_base
+
         # Default vertex_location to "global" if not provided for Vertex AI
         # Latest gemini models are only available through the global region
         if (
@@ -404,13 +513,30 @@ class LitellmLLM(LLM):
                 else nullcontext()
             )
             with env_ctx:
+                messages = _prompt_to_dicts(prompt)
+
+                # Bedrock's Converse API requires toolConfig when messages
+                # contain toolUse/toolResult content blocks. When no tools are
+                # provided for this request but the history contains tool
+                # content from previous turns, strip it to plain text.
+                is_bedrock = self._model_provider in {
+                    LlmProviderNames.BEDROCK,
+                    LlmProviderNames.BEDROCK_CONVERSE,
+                }
+                if (
+                    is_bedrock
+                    and not tools
+                    and _messages_contain_tool_content(messages)
+                ):
+                    messages = _strip_tool_content_from_messages(messages)
+
                 response = litellm.completion(
                     mock_response=get_llm_mock_response() or MOCK_LLM_RESPONSE,
                     model=model,
                     base_url=self._api_base or None,
                     api_version=self._api_version or None,
                     custom_llm_provider=self._custom_llm_provider or None,
-                    messages=_prompt_to_dicts(prompt),
+                    messages=messages,
                     tools=tools,
                     tool_choice=tool_choice,
                     stream=stream,

backend/onyx/llm/utils.py

@@ -322,7 +322,7 @@ def test_llm(llm: LLM) -> str | None:
     error_msg = None
     for _ in range(2):
         try:
-            llm.invoke(UserMessage(content="Do not respond"))
+            llm.invoke(UserMessage(content="Do not respond"), max_tokens=50)
             return None
         except Exception as e:
             error_msg = str(e)

backend/onyx/llm/well_known_providers/auto_update_service.py

@@ -13,44 +13,38 @@ from datetime import datetime
 import httpx
 from sqlalchemy.orm import Session
 
+from onyx.cache.factory import get_cache_backend
 from onyx.configs.app_configs import AUTO_LLM_CONFIG_URL
 from onyx.db.llm import fetch_auto_mode_providers
 from onyx.db.llm import sync_auto_mode_models
 from onyx.llm.well_known_providers.auto_update_models import LLMRecommendations
-from onyx.redis.redis_pool import get_redis_client
 from onyx.utils.logger import setup_logger
 
 logger = setup_logger()
 
-# Redis key for caching the last updated timestamp (per-tenant)
-_REDIS_KEY_LAST_UPDATED_AT = "auto_llm_update:last_updated_at"
+_CACHE_KEY_LAST_UPDATED_AT = "auto_llm_update:last_updated_at"
+_CACHE_TTL_SECONDS = 60 * 60 * 24  # 24 hours
 
 
 def _get_cached_last_updated_at() -> datetime | None:
-    """Get the cached last_updated_at timestamp from Redis."""
     try:
-        redis_client = get_redis_client()
-        value = redis_client.get(_REDIS_KEY_LAST_UPDATED_AT)
-        if value and isinstance(value, bytes):
-            # Value is bytes, decode to string then parse as ISO format
+        value = get_cache_backend().get(_CACHE_KEY_LAST_UPDATED_AT)
+        if value is not None:
             return datetime.fromisoformat(value.decode("utf-8"))
     except Exception as e:
-        logger.warning(f"Failed to get cached last_updated_at from Redis: {e}")
+        logger.warning(f"Failed to get cached last_updated_at: {e}")
     return None
 
 
 def _set_cached_last_updated_at(updated_at: datetime) -> None:
-    """Set the cached last_updated_at timestamp in Redis."""
     try:
-        redis_client = get_redis_client()
-        # Store as ISO format string, with 24 hour expiration
-        redis_client.set(
-            _REDIS_KEY_LAST_UPDATED_AT,
+        get_cache_backend().set(
+            _CACHE_KEY_LAST_UPDATED_AT,
             updated_at.isoformat(),
-            ex=60 * 60 * 24,  # 24 hours
+            ex=_CACHE_TTL_SECONDS,
         )
     except Exception as e:
-        logger.warning(f"Failed to set cached last_updated_at in Redis: {e}")
+        logger.warning(f"Failed to set cached last_updated_at: {e}")
 
 
 def fetch_llm_recommendations_from_github(
@@ -148,9 +142,8 @@ def sync_llm_models_from_github(
 
 
 def reset_cache() -> None:
-    """Reset the cache timestamp in Redis. Useful for testing."""
+    """Reset the cache timestamp. Useful for testing."""
     try:
-        redis_client = get_redis_client()
-        redis_client.delete(_REDIS_KEY_LAST_UPDATED_AT)
+        get_cache_backend().delete(_CACHE_KEY_LAST_UPDATED_AT)
     except Exception as e:
-        logger.warning(f"Failed to reset cache in Redis: {e}")
+        logger.warning(f"Failed to reset cache: {e}")

backend/onyx/llm/well_known_providers/constants.py

@@ -1,52 +1,27 @@
+from onyx.llm.constants import LlmProviderNames
+
 OPENAI_PROVIDER_NAME = "openai"
-# Curated list of OpenAI models to show by default in the UI
-OPENAI_VISIBLE_MODEL_NAMES = {
-    "gpt-5",
-    "gpt-5-mini",
-    "o1",
-    "o3-mini",
-    "gpt-4o",
-    "gpt-4o-mini",
-}
 
 BEDROCK_PROVIDER_NAME = "bedrock"
-BEDROCK_DEFAULT_MODEL = "anthropic.claude-3-5-sonnet-20241022-v2:0"
-
-
-def _fallback_bedrock_regions() -> list[str]:
-    # Fall back to a conservative set of well-known Bedrock regions if boto3 data isn't available.
-    return [
-        "us-east-1",
-        "us-east-2",
-        "us-gov-east-1",
-        "us-gov-west-1",
-        "us-west-2",
-        "ap-northeast-1",
-        "ap-south-1",
-        "ap-southeast-1",
-        "ap-southeast-2",
-        "ap-east-1",
-        "ca-central-1",
-        "eu-central-1",
-        "eu-west-2",
-    ]
 
 
 OLLAMA_PROVIDER_NAME = "ollama_chat"
 OLLAMA_API_KEY_CONFIG_KEY = "OLLAMA_API_KEY"
 
+LM_STUDIO_PROVIDER_NAME = "lm_studio"
+LM_STUDIO_API_KEY_CONFIG_KEY = "LM_STUDIO_API_KEY"
+
+# Providers that use optional Bearer auth from custom_config
+PROVIDERS_WITH_SPECIAL_API_KEY_HANDLING: dict[str, str] = {
+    LlmProviderNames.OLLAMA_CHAT: OLLAMA_API_KEY_CONFIG_KEY,
+    LlmProviderNames.LM_STUDIO: LM_STUDIO_API_KEY_CONFIG_KEY,
+}
+
 # OpenRouter
 OPENROUTER_PROVIDER_NAME = "openrouter"
 
 ANTHROPIC_PROVIDER_NAME = "anthropic"
 
-# Curated list of Anthropic models to show by default in the UI
-ANTHROPIC_VISIBLE_MODEL_NAMES = {
-    "claude-opus-4-5",
-    "claude-sonnet-4-5",
-    "claude-haiku-4-5",
-}
-
 AZURE_PROVIDER_NAME = "azure"
 
 
@@ -54,13 +29,6 @@ VERTEXAI_PROVIDER_NAME = "vertex_ai"
 VERTEX_CREDENTIALS_FILE_KWARG = "vertex_credentials"
 VERTEX_CREDENTIALS_FILE_KWARG_ENV_VAR_FORMAT = "CREDENTIALS_FILE"
 VERTEX_LOCATION_KWARG = "vertex_location"
-VERTEXAI_DEFAULT_MODEL = "gemini-2.5-flash"
-# Curated list of Vertex AI models to show by default in the UI
-VERTEXAI_VISIBLE_MODEL_NAMES = {
-    "gemini-2.5-flash",
-    "gemini-2.5-flash-lite",
-    "gemini-2.5-pro",
-}
 
 AWS_REGION_NAME_KWARG = "aws_region_name"
 AWS_REGION_NAME_KWARG_ENV_VAR_FORMAT = "AWS_REGION_NAME"

backend/onyx/llm/well_known_providers/llm_provider_options.py

@@ -15,6 +15,7 @@ from onyx.llm.well_known_providers.auto_update_service import (
 from onyx.llm.well_known_providers.constants import ANTHROPIC_PROVIDER_NAME
 from onyx.llm.well_known_providers.constants import AZURE_PROVIDER_NAME
 from onyx.llm.well_known_providers.constants import BEDROCK_PROVIDER_NAME
+from onyx.llm.well_known_providers.constants import LM_STUDIO_PROVIDER_NAME
 from onyx.llm.well_known_providers.constants import OLLAMA_PROVIDER_NAME
 from onyx.llm.well_known_providers.constants import OPENAI_PROVIDER_NAME
 from onyx.llm.well_known_providers.constants import OPENROUTER_PROVIDER_NAME
@@ -44,6 +45,7 @@ def _get_provider_to_models_map() -> dict[str, list[str]]:
         ANTHROPIC_PROVIDER_NAME: get_anthropic_model_names(),
         VERTEXAI_PROVIDER_NAME: get_vertexai_model_names(),
         OLLAMA_PROVIDER_NAME: [],  # Dynamic - fetched from Ollama API
+        LM_STUDIO_PROVIDER_NAME: [],  # Dynamic - fetched from LM Studio API
         OPENROUTER_PROVIDER_NAME: [],  # Dynamic - fetched from OpenRouter API
     }
 
@@ -323,6 +325,7 @@ def get_provider_display_name(provider_name: str) -> str:
     _ONYX_PROVIDER_DISPLAY_NAMES: dict[str, str] = {
         OPENAI_PROVIDER_NAME: "ChatGPT (OpenAI)",
         OLLAMA_PROVIDER_NAME: "Ollama",
+        LM_STUDIO_PROVIDER_NAME: "LM Studio",
         ANTHROPIC_PROVIDER_NAME: "Claude (Anthropic)",
         AZURE_PROVIDER_NAME: "Azure OpenAI",
         BEDROCK_PROVIDER_NAME: "Amazon Bedrock",

backend/onyx/llm/well_known_providers/recommended-models.json

@@ -1,12 +1,12 @@
 {
   "version": "1.1",
-  "updated_at": "2026-02-05T00:00:00Z",
+  "updated_at": "2026-03-05T00:00:00Z",
   "providers": {
     "openai": {
-      "default_model": { "name": "gpt-5.2" },
+      "default_model": { "name": "gpt-5.4" },
       "additional_visible_models": [
-        { "name": "gpt-5-mini" },
-        { "name": "gpt-4.1" }
+        { "name": "gpt-5.4" },
+        { "name": "gpt-5.2" }
       ]
     },
     "anthropic": {
@@ -16,6 +16,10 @@
           "name": "claude-opus-4-6",
           "display_name": "Claude Opus 4.6"
         },
+        {
+          "name": "claude-sonnet-4-6",
+          "display_name": "Claude Sonnet 4.6"
+        },
         {
           "name": "claude-opus-4-5",
           "display_name": "Claude Opus 4.5"