nit

backend/alembic/env.py

@@ -1,4 +1,4 @@
-from typing import Any
+from typing import Any, Literal
 from onyx.db.engine.iam_auth import get_iam_auth_token
 from onyx.configs.app_configs import USE_IAM_AUTH
 from onyx.configs.app_configs import POSTGRES_HOST
@@ -19,6 +19,7 @@ from logging.config import fileConfig
 
 from alembic import context
 from sqlalchemy.ext.asyncio import create_async_engine
+from sqlalchemy.sql.schema import SchemaItem
 from onyx.configs.constants import SSL_CERT_FILE
 from shared_configs.configs import (
     MULTI_TENANT,
@@ -44,6 +45,8 @@ if config.config_file_name is not None and config.attributes.get(
 
 target_metadata = [Base.metadata, ResultModelBase.metadata]
 
+EXCLUDE_TABLES = {"kombu_queue", "kombu_message"}
+
 logger = logging.getLogger(__name__)
 
 ssl_context: ssl.SSLContext | None = None
@@ -53,6 +56,25 @@ if USE_IAM_AUTH:
     ssl_context = ssl.create_default_context(cafile=SSL_CERT_FILE)
 
 
+def include_object(
+    object: SchemaItem,  # noqa: ARG001
+    name: str | None,
+    type_: Literal[
+        "schema",
+        "table",
+        "column",
+        "index",
+        "unique_constraint",
+        "foreign_key_constraint",
+    ],
+    reflected: bool,  # noqa: ARG001
+    compare_to: SchemaItem | None,  # noqa: ARG001
+) -> bool:
+    if type_ == "table" and name in EXCLUDE_TABLES:
+        return False
+    return True
+
+
 def filter_tenants_by_range(
     tenant_ids: list[str], start_range: int | None = None, end_range: int | None = None
 ) -> list[str]:
@@ -209,6 +231,7 @@ def do_run_migrations(
     context.configure(
         connection=connection,
         target_metadata=target_metadata,  # type: ignore
+        include_object=include_object,
         version_table_schema=schema_name,
         include_schemas=True,
         compare_type=True,
@@ -382,6 +405,7 @@ def run_migrations_offline() -> None:
                 url=url,
                 target_metadata=target_metadata,  # type: ignore
                 literal_binds=True,
+                include_object=include_object,
                 version_table_schema=schema,
                 include_schemas=True,
                 script_location=config.get_main_option("script_location"),
@@ -423,6 +447,7 @@ def run_migrations_offline() -> None:
                 url=url,
                 target_metadata=target_metadata,  # type: ignore
                 literal_binds=True,
+                include_object=include_object,
                 version_table_schema=schema,
                 include_schemas=True,
                 script_location=config.get_main_option("script_location"),
@@ -465,6 +490,7 @@ def run_migrations_online() -> None:
             context.configure(
                 connection=connection,
                 target_metadata=target_metadata,  # type: ignore
+                include_object=include_object,
                 version_table_schema=schema_name,
                 include_schemas=True,
                 compare_type=True,

backend/alembic_tenants/env.py

@@ -1,9 +1,11 @@
 import asyncio
 from logging.config import fileConfig
+from typing import Literal
 
 from sqlalchemy import pool
 from sqlalchemy.engine import Connection
 from sqlalchemy.ext.asyncio import create_async_engine
+from sqlalchemy.schema import SchemaItem
 
 from alembic import context
 from onyx.db.engine.sql_engine import build_connection_string
@@ -33,6 +35,27 @@ target_metadata = [PublicBase.metadata]
 # my_important_option = config.get_main_option("my_important_option")
 # ... etc.
 
+EXCLUDE_TABLES = {"kombu_queue", "kombu_message"}
+
+
+def include_object(
+    object: SchemaItem,  # noqa: ARG001
+    name: str | None,
+    type_: Literal[
+        "schema",
+        "table",
+        "column",
+        "index",
+        "unique_constraint",
+        "foreign_key_constraint",
+    ],
+    reflected: bool,  # noqa: ARG001
+    compare_to: SchemaItem | None,  # noqa: ARG001
+) -> bool:
+    if type_ == "table" and name in EXCLUDE_TABLES:
+        return False
+    return True
+
 
 def run_migrations_offline() -> None:
     """Run migrations in 'offline' mode.
@@ -62,6 +85,7 @@ def do_run_migrations(connection: Connection) -> None:
     context.configure(
         connection=connection,
         target_metadata=target_metadata,  # type: ignore[arg-type]
+        include_object=include_object,
     )
 
     with context.begin_transaction():

backend/onyx/background/README.md

@@ -56,6 +56,7 @@ Then it cycles through its tasks as scheduled by Celery Beat:
 | `check_for_user_file_processing` | 20s | Checks for user uploads → dispatches to `USER_FILE_PROCESSING` queue |
 | `check_for_checkpoint_cleanup` | 1h | Cleans up old indexing checkpoints |
 | `check_for_index_attempt_cleanup` | 30m | Cleans up old index attempts |
+| `kombu_message_cleanup_task` | periodic | Cleans orphaned Kombu messages from DB (Kombu being the messaging framework used by Celery) |
 | `celery_beat_heartbeat` | 1m | Heartbeat for Beat watchdog |
 
 Watchdog is a separate Python process managed by supervisord which runs alongside celery workers. It checks the ONYX_CELERY_BEAT_HEARTBEAT_KEY in

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

@@ -317,6 +317,7 @@ celery_app.autodiscover_tasks(
             "onyx.background.celery.tasks.docprocessing",
             "onyx.background.celery.tasks.evals",
             "onyx.background.celery.tasks.hierarchyfetching",
+            "onyx.background.celery.tasks.periodic",
             "onyx.background.celery.tasks.pruning",
             "onyx.background.celery.tasks.shared",
             "onyx.background.celery.tasks.vespa",

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

@@ -0,0 +1,138 @@
+#####
+# Periodic Tasks
+#####
+import json
+from typing import Any
+
+from celery import shared_task
+from celery.contrib.abortable import AbortableTask  # type: ignore
+from celery.exceptions import TaskRevokedError
+from sqlalchemy import inspect
+from sqlalchemy import text
+from sqlalchemy.orm import Session
+
+from onyx.background.celery.apps.app_base import task_logger
+from onyx.configs.app_configs import JOB_TIMEOUT
+from onyx.configs.constants import OnyxCeleryTask
+from onyx.configs.constants import PostgresAdvisoryLocks
+from onyx.db.engine.sql_engine import get_session_with_current_tenant
+
+
+@shared_task(
+    name=OnyxCeleryTask.KOMBU_MESSAGE_CLEANUP_TASK,
+    soft_time_limit=JOB_TIMEOUT,
+    bind=True,
+    base=AbortableTask,
+)
+def kombu_message_cleanup_task(self: Any, tenant_id: str) -> int:  # noqa: ARG001
+    """Runs periodically to clean up the kombu_message table"""
+
+    # we will select messages older than this amount to clean up
+    KOMBU_MESSAGE_CLEANUP_AGE = 7  # days
+    KOMBU_MESSAGE_CLEANUP_PAGE_LIMIT = 1000
+
+    ctx = {}
+    ctx["last_processed_id"] = 0
+    ctx["deleted"] = 0
+    ctx["cleanup_age"] = KOMBU_MESSAGE_CLEANUP_AGE
+    ctx["page_limit"] = KOMBU_MESSAGE_CLEANUP_PAGE_LIMIT
+    with get_session_with_current_tenant() as db_session:
+        # Exit the task if we can't take the advisory lock
+        result = db_session.execute(
+            text("SELECT pg_try_advisory_lock(:id)"),
+            {"id": PostgresAdvisoryLocks.KOMBU_MESSAGE_CLEANUP_LOCK_ID.value},
+        ).scalar()
+        if not result:
+            return 0
+
+        while True:
+            if self.is_aborted():
+                raise TaskRevokedError("kombu_message_cleanup_task was aborted.")
+
+            b = kombu_message_cleanup_task_helper(ctx, db_session)
+            if not b:
+                break
+
+            db_session.commit()
+
+    if ctx["deleted"] > 0:
+        task_logger.info(
+            f"Deleted {ctx['deleted']} orphaned messages from kombu_message."
+        )
+
+    return ctx["deleted"]
+
+
+def kombu_message_cleanup_task_helper(ctx: dict, db_session: Session) -> bool:
+    """
+    Helper function to clean up old messages from the `kombu_message` table that are no longer relevant.
+
+    This function retrieves messages from the `kombu_message` table that are no longer visible and
+    older than a specified interval. It checks if the corresponding task_id exists in the
+    `celery_taskmeta` table. If the task_id does not exist, the message is deleted.
+
+    Args:
+        ctx (dict): A context dictionary containing configuration parameters such as:
+            - 'cleanup_age' (int): The age in days after which messages are considered old.
+            - 'page_limit' (int): The maximum number of messages to process in one batch.
+            - 'last_processed_id' (int): The ID of the last processed message to handle pagination.
+            - 'deleted' (int): A counter to track the number of deleted messages.
+        db_session (Session): The SQLAlchemy database session for executing queries.
+
+    Returns:
+        bool: Returns True if there are more rows to process, False if not.
+    """
+
+    inspector = inspect(db_session.bind)
+    if not inspector:
+        return False
+
+    # With the move to redis as celery's broker and backend, kombu tables may not even exist.
+    # We can fail silently.
+    if not inspector.has_table("kombu_message"):
+        return False
+
+    query = text(
+        """
+    SELECT id, timestamp, payload
+    FROM kombu_message WHERE visible = 'false'
+    AND timestamp < CURRENT_TIMESTAMP - INTERVAL :interval_days
+    AND id > :last_processed_id
+    ORDER BY id
+    LIMIT :page_limit
+"""
+    )
+    kombu_messages = db_session.execute(
+        query,
+        {
+            "interval_days": f"{ctx['cleanup_age']} days",
+            "page_limit": ctx["page_limit"],
+            "last_processed_id": ctx["last_processed_id"],
+        },
+    ).fetchall()
+
+    if len(kombu_messages) == 0:
+        return False
+
+    for msg in kombu_messages:
+        payload = json.loads(msg[2])
+        task_id = payload["headers"]["id"]
+
+        # Check if task_id exists in celery_taskmeta
+        task_exists = db_session.execute(
+            text("SELECT 1 FROM celery_taskmeta WHERE task_id = :task_id"),
+            {"task_id": task_id},
+        ).fetchone()
+
+        # If task_id does not exist, delete the message
+        if not task_exists:
+            result = db_session.execute(
+                text("DELETE FROM kombu_message WHERE id = :message_id"),
+                {"message_id": msg[0]},
+            )
+            if result.rowcount > 0:  # type: ignore
+                ctx["deleted"] += 1
+
+        ctx["last_processed_id"] = msg[0]
+
+    return True

backend/onyx/configs/constants.py

@@ -12,11 +12,6 @@ SLACK_USER_TOKEN_PREFIX = "xoxp-"
 SLACK_BOT_TOKEN_PREFIX = "xoxb-"
 ONYX_EMAILABLE_LOGO_MAX_DIM = 512
 
-# The mask_string() function in encryption.py uses "•" (U+2022 BULLET) to mask secrets.
-MASK_CREDENTIAL_CHAR = "\u2022"
-# Pattern produced by mask_string for strings >= 14 chars: "abcd...wxyz" (exactly 11 chars)
-MASK_CREDENTIAL_LONG_RE = re.compile(r"^.{4}\.{3}.{4}$")
-
 SOURCE_TYPE = "source_type"
 # stored in the `metadata` of a chunk. Used to signify that this chunk should
 # not be used for QA. For example, Google Drive file types which can't be parsed
@@ -396,6 +391,10 @@ class MilestoneRecordType(str, Enum):
     REQUESTED_CONNECTOR = "requested_connector"
 
 
+class PostgresAdvisoryLocks(Enum):
+    KOMBU_MESSAGE_CLEANUP_LOCK_ID = auto()
+
+
 class OnyxCeleryQueues:
     # "celery" is the default queue defined by celery and also the queue
     # we are running in the primary worker to run system tasks
@@ -578,6 +577,7 @@ class OnyxCeleryTask:
     MONITOR_PROCESS_MEMORY = "monitor_process_memory"
     CELERY_BEAT_HEARTBEAT = "celery_beat_heartbeat"
 
+    KOMBU_MESSAGE_CLEANUP_TASK = "kombu_message_cleanup_task"
     CONNECTOR_PERMISSION_SYNC_GENERATOR_TASK = (
         "connector_permission_sync_generator_task"
     )

backend/onyx/db/federated.py

@@ -8,8 +8,6 @@ from sqlalchemy.orm import selectinload
 from sqlalchemy.orm import Session
 
 from onyx.configs.constants import FederatedConnectorSource
-from onyx.configs.constants import MASK_CREDENTIAL_CHAR
-from onyx.configs.constants import MASK_CREDENTIAL_LONG_RE
 from onyx.db.engine.sql_engine import get_session_with_current_tenant
 from onyx.db.models import DocumentSet
 from onyx.db.models import FederatedConnector
@@ -47,23 +45,6 @@ def fetch_all_federated_connectors_parallel() -> list[FederatedConnector]:
         return fetch_all_federated_connectors(db_session)
 
 
-def _reject_masked_credentials(credentials: dict[str, Any]) -> None:
-    """Raise if any credential string value contains mask placeholder characters.
-
-    mask_string() has two output formats:
-    - Short strings (< 14 chars): "••••••••••••" (U+2022 BULLET)
-    - Long strings (>= 14 chars): "abcd...wxyz" (first4 + "..." + last4)
-    Both must be rejected.
-    """
-    for key, val in credentials.items():
-        if isinstance(val, str) and (
-            MASK_CREDENTIAL_CHAR in val or MASK_CREDENTIAL_LONG_RE.match(val)
-        ):
-            raise ValueError(
-                f"Credential field '{key}' contains masked placeholder characters. Please provide the actual credential value."
-            )
-
-
 def validate_federated_connector_credentials(
     source: FederatedConnectorSource,
     credentials: dict[str, Any],
@@ -85,8 +66,6 @@ def create_federated_connector(
     config: dict[str, Any] | None = None,
 ) -> FederatedConnector:
     """Create a new federated connector with credential and config validation."""
-    _reject_masked_credentials(credentials)
-
     # Validate credentials before creating
     if not validate_federated_connector_credentials(source, credentials):
         raise ValueError(
@@ -298,8 +277,6 @@ def update_federated_connector(
     )
 
     if credentials is not None:
-        _reject_masked_credentials(credentials)
-
         # Validate credentials before updating
         if not validate_federated_connector_credentials(
             federated_connector.source, credentials

backend/tests/unit/federated_connector/test_reject_masked_credentials.py

@@ -1,58 +0,0 @@
-import pytest
-
-from onyx.configs.constants import MASK_CREDENTIAL_CHAR
-from onyx.db.federated import _reject_masked_credentials
-
-
-class TestRejectMaskedCredentials:
-    """Verify that masked credential values are never accepted for DB writes.
-
-    mask_string() has two output formats:
-    - Short strings (< 14 chars): "••••••••••••" (U+2022 BULLET)
-    - Long strings (>= 14 chars): "abcd...wxyz" (first4 + "..." + last4)
-    _reject_masked_credentials must catch both.
-    """
-
-    def test_rejects_fully_masked_value(self) -> None:
-        masked = MASK_CREDENTIAL_CHAR * 12  # "••••••••••••"
-        with pytest.raises(ValueError, match="masked placeholder"):
-            _reject_masked_credentials({"client_id": masked})
-
-    def test_rejects_long_string_masked_value(self) -> None:
-        """mask_string returns 'first4...last4' for long strings — the real
-        format used for OAuth credentials like client_id and client_secret."""
-        with pytest.raises(ValueError, match="masked placeholder"):
-            _reject_masked_credentials({"client_id": "1234...7890"})
-
-    def test_rejects_when_any_field_is_masked(self) -> None:
-        """Even if client_id is real, a masked client_secret must be caught."""
-        with pytest.raises(ValueError, match="client_secret"):
-            _reject_masked_credentials(
-                {
-                    "client_id": "1234567890.1234567890",
-                    "client_secret": MASK_CREDENTIAL_CHAR * 12,
-                }
-            )
-
-    def test_accepts_real_credentials(self) -> None:
-        # Should not raise
-        _reject_masked_credentials(
-            {
-                "client_id": "1234567890.1234567890",
-                "client_secret": "test_client_secret_value",
-            }
-        )
-
-    def test_accepts_empty_dict(self) -> None:
-        # Should not raise — empty credentials are handled elsewhere
-        _reject_masked_credentials({})
-
-    def test_ignores_non_string_values(self) -> None:
-        # Non-string values (None, bool, int) should pass through
-        _reject_masked_credentials(
-            {
-                "client_id": "real_value",
-                "redirect_uri": None,
-                "some_flag": True,
-            }
-        )

cli/cmd/agents.go

@@ -6,7 +6,6 @@ import (
 	"text/tabwriter"
 
 	"github.com/onyx-dot-app/onyx/cli/internal/api"
-	"github.com/onyx-dot-app/onyx/cli/internal/config"
 	"github.com/onyx-dot-app/onyx/cli/internal/exitcodes"
 	"github.com/spf13/cobra"
 )
@@ -25,7 +24,7 @@ Use --json for machine-readable output.`,
   onyx-cli agents --json
   onyx-cli agents --json | jq '.[].name'`,
 		RunE: func(cmd *cobra.Command, args []string) error {
-			cfg := config.Load()
+			cfg := loadConfig(cmd)
 			if !cfg.IsConfigured() {
 				return exitcodes.New(exitcodes.NotConfigured, "onyx CLI is not configured\n  Run: onyx-cli configure")
 			}

cli/cmd/ask.go

@@ -11,7 +11,6 @@ import (
 	"syscall"
 
 	"github.com/onyx-dot-app/onyx/cli/internal/api"
-	"github.com/onyx-dot-app/onyx/cli/internal/config"
 	"github.com/onyx-dot-app/onyx/cli/internal/exitcodes"
 	"github.com/onyx-dot-app/onyx/cli/internal/models"
 	"github.com/onyx-dot-app/onyx/cli/internal/overflow"
@@ -49,7 +48,7 @@ to a temp file. Set --max-output 0 to disable truncation.`,
   cat error.log | onyx-cli ask --prompt "Find the root cause"
   echo "what is onyx?" | onyx-cli ask`,
 		RunE: func(cmd *cobra.Command, args []string) error {
-			cfg := config.Load()
+			cfg := loadConfig(cmd)
 			if !cfg.IsConfigured() {
 				return exitcodes.New(exitcodes.NotConfigured, "onyx CLI is not configured\n  Run: onyx-cli configure")
 			}

cli/cmd/chat.go

@@ -21,7 +21,7 @@ an interactive setup wizard will guide you through configuration.`,
 		Example: `  onyx-cli chat
   onyx-cli`,
 		RunE: func(cmd *cobra.Command, args []string) error {
-			cfg := config.Load()
+			cfg := loadConfig(cmd)
 
 			// First-run: onboarding
 			if !config.ConfigExists() || !cfg.IsConfigured() {

cli/cmd/configure.go

@@ -69,7 +69,7 @@ Use --dry-run to test the connection without saving the configuration.`,
 				return exitcodes.New(exitcodes.BadRequest, "both --server-url and --api-key are required for non-interactive setup\n  Run 'onyx-cli configure' without flags for interactive setup")
 			}
 
-			cfg := config.Load()
+			cfg := loadConfig(cmd)
 			onboarding.Run(&cfg)
 			return nil
 		},

cli/cmd/experiments.go

@@ -12,7 +12,7 @@ func newExperimentsCmd() *cobra.Command {
 		Use:   "experiments",
 		Short: "List experimental features and their status",
 		RunE: func(cmd *cobra.Command, args []string) error {
-			cfg := config.Load()
+			cfg := loadConfig(cmd)
 			_, _ = fmt.Fprintln(cmd.OutOrStdout(), config.ExperimentsText(cfg.Features))
 			return nil
 		},

cli/cmd/root.go

@@ -13,6 +13,20 @@ import (
 	"github.com/spf13/cobra"
 )
 
+// loadConfig loads the CLI config, using the --config-file persistent flag if set.
+func loadConfig(cmd *cobra.Command) config.OnyxCliConfig {
+	cf, _ := cmd.Flags().GetString("config-file")
+	return config.Load(cf)
+}
+
+// effectiveConfigPath returns the config file path, respecting --config-file.
+func effectiveConfigPath(cmd *cobra.Command) string {
+	if cf, _ := cmd.Flags().GetString("config-file"); cf != "" {
+		return cf
+	}
+	return config.ConfigFilePath()
+}
+
 // Version and Commit are set via ldflags at build time.
 var (
 	Version string
@@ -29,7 +43,7 @@ func fullVersion() string {
 func printVersion(cmd *cobra.Command) {
 	_, _ = fmt.Fprintf(cmd.OutOrStdout(), "Client version: %s\n", fullVersion())
 
-	cfg := config.Load()
+	cfg := loadConfig(cmd)
 	if !cfg.IsConfigured() {
 		_, _ = fmt.Fprintf(cmd.OutOrStdout(), "Server version: unknown (not configured)\n")
 		return
@@ -84,6 +98,8 @@ func Execute() error {
 	}
 
 	rootCmd.PersistentFlags().BoolVar(&opts.Debug, "debug", false, "run in debug mode")
+	rootCmd.PersistentFlags().String("config-file", "",
+		"Path to config file (default: "+config.ConfigFilePath()+")")
 
 	// Custom --version flag instead of Cobra's built-in (which only shows one version string)
 	var showVersion bool

cli/cmd/serve.go

@@ -50,16 +50,14 @@ func sessionEnv(s ssh.Session, key string) string {
 	return ""
 }
 
-func validateAPIKey(serverURL string, apiKey string) error {
+func validateAPIKey(serverCfg config.OnyxCliConfig, apiKey string) error {
 	trimmedKey := strings.TrimSpace(apiKey)
 	if len(trimmedKey) > maxAPIKeyLength {
 		return fmt.Errorf("API key is too long (max %d characters)", maxAPIKeyLength)
 	}
 
-	cfg := config.OnyxCliConfig{
-		ServerURL: serverURL,
-		APIKey:    trimmedKey,
-	}
+	cfg := serverCfg
+	cfg.APIKey = trimmedKey
 	client := api.NewClient(cfg)
 	ctx, cancel := context.WithTimeout(context.Background(), apiKeyValidationTimeout)
 	defer cancel()
@@ -83,7 +81,7 @@ type authValidatedMsg struct {
 
 type authModel struct {
 	input     textinput.Model
-	serverURL string
+	serverCfg config.OnyxCliConfig
 	state     authState
 	apiKey    string // set on successful validation
 	errMsg    string
@@ -91,7 +89,7 @@ type authModel struct {
 	aborted   bool
 }
 
-func newAuthModel(serverURL, initialErr string) authModel {
+func newAuthModel(serverCfg config.OnyxCliConfig, initialErr string) authModel {
 	ti := textinput.New()
 	ti.Prompt = "  API Key: "
 	ti.EchoMode = textinput.EchoPassword
@@ -102,7 +100,7 @@ func newAuthModel(serverURL, initialErr string) authModel {
 
 	return authModel{
 		input:     ti,
-		serverURL: serverURL,
+		serverCfg: serverCfg,
 		errMsg:    initialErr,
 	}
 }
@@ -138,9 +136,9 @@ func (m authModel) Update(msg tea.Msg) (authModel, tea.Cmd) {
 			}
 			m.state = authValidating
 			m.errMsg = ""
-			serverURL := m.serverURL
+			serverCfg := m.serverCfg
 			return m, func() tea.Msg {
-				return authValidatedMsg{key: key, err: validateAPIKey(serverURL, key)}
+				return authValidatedMsg{key: key, err: validateAPIKey(serverCfg, key)}
 			}
 		}
 
@@ -171,12 +169,13 @@ func (m authModel) Update(msg tea.Msg) (authModel, tea.Cmd) {
 }
 
 func (m authModel) View() string {
-	settingsURL := strings.TrimRight(m.serverURL, "/") + "/app/settings/accounts-access"
+	serverURL := m.serverCfg.ServerURL
+	settingsURL := strings.TrimRight(serverURL, "/") + "/app/settings/accounts-access"
 
 	var b strings.Builder
 	b.WriteString("\n")
 	b.WriteString("  \x1b[1;35mOnyx CLI\x1b[0m\n")
-	b.WriteString("  \x1b[90m" + m.serverURL + "\x1b[0m\n")
+	b.WriteString("  \x1b[90m" + serverURL + "\x1b[0m\n")
 	b.WriteString("\n")
 	b.WriteString("  Generate an API key at:\n")
 	b.WriteString("  \x1b[4;34m" + settingsURL + "\x1b[0m\n")
@@ -215,7 +214,7 @@ type serveModel struct {
 
 func newServeModel(serverCfg config.OnyxCliConfig, initialErr string) serveModel {
 	return serveModel{
-		auth:      newAuthModel(serverCfg.ServerURL, initialErr),
+		auth:      newAuthModel(serverCfg, initialErr),
 		serverCfg: serverCfg,
 	}
 }
@@ -238,11 +237,8 @@ func (m serveModel) Update(msg tea.Msg) (tea.Model, tea.Cmd) {
 			return m, tea.Quit
 		}
 		if m.auth.apiKey != "" {
-			cfg := config.OnyxCliConfig{
-				ServerURL:      m.serverCfg.ServerURL,
-				APIKey:         m.auth.apiKey,
-				DefaultAgentID: m.serverCfg.DefaultAgentID,
-			}
+			cfg := m.serverCfg
+			cfg.APIKey = m.auth.apiKey
 			m.tui = tui.NewModel(cfg)
 			m.authed = true
 			w, h := m.width, m.height
@@ -280,6 +276,8 @@ func newServeCmd() *cobra.Command {
 		rateLimitPerMin   int
 		rateLimitBurst    int
 		rateLimitCache    int
+		serverURL         string
+		apiServerURL      string
 	)
 
 	cmd := &cobra.Command{
@@ -300,9 +298,18 @@ environment variable (the --host-key flag takes precedence).`,
 		Example: `  onyx-cli serve --port 2222
   ssh localhost -p 2222
   onyx-cli serve --host 0.0.0.0 --port 2222
-  onyx-cli serve --idle-timeout 30m --max-session-timeout 2h`,
+  onyx-cli serve --idle-timeout 30m --max-session-timeout 2h
+  onyx-cli serve --server-url https://my-onyx.example.com
+  onyx-cli serve --api-server-url http://api_server:8080  # bypass nginx
+  onyx-cli serve --config-file /etc/onyx-cli/config.json  # global flag`,
 		RunE: func(cmd *cobra.Command, args []string) error {
-			serverCfg := config.Load()
+			serverCfg := loadConfig(cmd)
+			if cmd.Flags().Changed("server-url") {
+				serverCfg.ServerURL = serverURL
+			}
+			if cmd.Flags().Changed("api-server-url") {
+				serverCfg.InternalURL = apiServerURL
+			}
 			if serverCfg.ServerURL == "" {
 				return exitcodes.New(exitcodes.NotConfigured, "server URL is not configured\n  Run: onyx-cli configure")
 			}
@@ -333,7 +340,7 @@ environment variable (the --host-key flag takes precedence).`,
 				var envErr string
 
 				if apiKey != "" {
-					if err := validateAPIKey(serverCfg.ServerURL, apiKey); err != nil {
+					if err := validateAPIKey(serverCfg, apiKey); err != nil {
 						envErr = fmt.Sprintf("ONYX_API_KEY from SSH environment is invalid: %s", err.Error())
 						apiKey = ""
 					}
@@ -341,11 +348,8 @@ environment variable (the --host-key flag takes precedence).`,
 
 				if apiKey != "" {
 					// Env key is valid — go straight to the TUI.
-					cfg := config.OnyxCliConfig{
-						ServerURL:      serverCfg.ServerURL,
-						APIKey:         apiKey,
-						DefaultAgentID: serverCfg.DefaultAgentID,
-					}
+					cfg := serverCfg
+					cfg.APIKey = apiKey
 					return tui.NewModel(cfg), []tea.ProgramOption{
 						tea.WithAltScreen(),
 						tea.WithMouseCellMotion(),
@@ -446,6 +450,10 @@ environment variable (the --host-key flag takes precedence).`,
 		defaultServeRateLimitCacheSize,
 		"Maximum number of IP limiter entries tracked in memory",
 	)
+	cmd.Flags().StringVar(&serverURL, "server-url", "",
+		"Onyx server URL (overrides config file and $"+config.EnvServerURL+")")
+	cmd.Flags().StringVar(&apiServerURL, "api-server-url", "",
+		"API server URL for direct access, bypassing nginx (overrides $"+config.EnvAPIServerURL+")")
 
 	return cmd
 }

cli/cmd/validate.go

@@ -4,10 +4,10 @@ import (
 	"context"
 	"errors"
 	"fmt"
+	"os"
 	"time"
 
 	"github.com/onyx-dot-app/onyx/cli/internal/api"
-	"github.com/onyx-dot-app/onyx/cli/internal/config"
 	"github.com/onyx-dot-app/onyx/cli/internal/exitcodes"
 	"github.com/onyx-dot-app/onyx/cli/internal/version"
 	log "github.com/sirupsen/logrus"
@@ -23,19 +23,21 @@ is valid. Also reports the server version and warns if it is below the
 minimum required.`,
 		Example: `  onyx-cli validate-config`,
 		RunE: func(cmd *cobra.Command, args []string) error {
+			cfgPath := effectiveConfigPath(cmd)
+
 			// Check config file
-			if !config.ConfigExists() {
-				return exitcodes.Newf(exitcodes.NotConfigured, "config file not found at %s\n  Run: onyx-cli configure", config.ConfigFilePath())
+			if _, err := os.Stat(cfgPath); err != nil {
+				return exitcodes.Newf(exitcodes.NotConfigured, "config file not found at %s\n  Run: onyx-cli configure", cfgPath)
 			}
 
-			cfg := config.Load()
+			cfg := loadConfig(cmd)
 
 			// Check API key
 			if !cfg.IsConfigured() {
 				return exitcodes.New(exitcodes.NotConfigured, "API key is missing\n  Run: onyx-cli configure")
 			}
 
-			_, _ = fmt.Fprintf(cmd.OutOrStdout(), "Config:  %s\n", config.ConfigFilePath())
+			_, _ = fmt.Fprintf(cmd.OutOrStdout(), "Config:  %s\n", cfgPath)
 			_, _ = fmt.Fprintf(cmd.OutOrStdout(), "Server:  %s\n", cfg.ServerURL)
 
 			// Test connection

cli/internal/api/client.go

@@ -16,18 +16,30 @@ import (
 
 	"github.com/onyx-dot-app/onyx/cli/internal/config"
 	"github.com/onyx-dot-app/onyx/cli/internal/models"
+	log "github.com/sirupsen/logrus"
 )
 
 // Client is the Onyx API client.
 type Client struct {
-	baseURL        string
+	serverURL      string       // root server URL (for reachability checks)
+	baseURL        string       // API base URL (includes /api when going through nginx)
 	apiKey         string
 	httpClient     *http.Client // default 30s timeout for quick requests
 	longHTTPClient *http.Client // 5min timeout for streaming/uploads
 }
 
 // NewClient creates a new API client from config.
+//
+// When InternalURL is set, requests go directly to the API server (no /api
+// prefix needed — mirrors INTERNAL_URL in the web server). Otherwise,
+// requests go through the nginx proxy at ServerURL which strips /api.
 func NewClient(cfg config.OnyxCliConfig) *Client {
+	baseURL := apiBaseURL(cfg)
+	log.WithFields(log.Fields{
+		"server_url":   cfg.ServerURL,
+		"internal_url": cfg.InternalURL,
+		"base_url":     baseURL,
+	}).Debug("creating API client")
 	var transport *http.Transport
 	if t, ok := http.DefaultTransport.(*http.Transport); ok {
 		transport = t.Clone()
@@ -35,8 +47,9 @@ func NewClient(cfg config.OnyxCliConfig) *Client {
 		transport = &http.Transport{}
 	}
 	return &Client{
-		baseURL: strings.TrimRight(cfg.ServerURL, "/"),
-		apiKey:  cfg.APIKey,
+		serverURL: strings.TrimRight(cfg.ServerURL, "/"),
+		baseURL:   baseURL,
+		apiKey:    cfg.APIKey,
 		httpClient: &http.Client{
 			Timeout:   30 * time.Second,
 			Transport: transport,
@@ -48,14 +61,27 @@ func NewClient(cfg config.OnyxCliConfig) *Client {
 	}
 }
 
+// apiBaseURL returns the base URL for API requests. When InternalURL is set,
+// it points directly at the API server. Otherwise it goes through the nginx
+// proxy at ServerURL/api.
+func apiBaseURL(cfg config.OnyxCliConfig) string {
+	if cfg.InternalURL != "" {
+		return strings.TrimRight(cfg.InternalURL, "/")
+	}
+	return strings.TrimRight(cfg.ServerURL, "/") + "/api"
+}
+
 // UpdateConfig replaces the client's config.
 func (c *Client) UpdateConfig(cfg config.OnyxCliConfig) {
-	c.baseURL = strings.TrimRight(cfg.ServerURL, "/")
+	c.serverURL = strings.TrimRight(cfg.ServerURL, "/")
+	c.baseURL = apiBaseURL(cfg)
 	c.apiKey = cfg.APIKey
 }
 
 func (c *Client) newRequest(ctx context.Context, method, path string, body io.Reader) (*http.Request, error) {
-	req, err := http.NewRequestWithContext(ctx, method, c.baseURL+path, body)
+	url := c.baseURL + path
+	log.WithFields(log.Fields{"method": method, "url": url}).Debug("API request")
+	req, err := http.NewRequestWithContext(ctx, method, url, body)
 	if err != nil {
 		return nil, err
 	}
@@ -87,12 +113,16 @@ func (c *Client) doJSON(ctx context.Context, method, path string, reqBody any, r
 
 	resp, err := c.httpClient.Do(req)
 	if err != nil {
+		log.WithError(err).WithField("url", req.URL.String()).Debug("API request failed")
 		return err
 	}
 	defer func() { _ = resp.Body.Close() }()
 
+	log.WithFields(log.Fields{"url": req.URL.String(), "status": resp.StatusCode}).Debug("API response")
+
 	if resp.StatusCode < 200 || resp.StatusCode >= 300 {
 		respBody, _ := io.ReadAll(resp.Body)
+		log.WithFields(log.Fields{"status": resp.StatusCode, "body": string(respBody)}).Debug("API error response")
 		return &OnyxAPIError{StatusCode: resp.StatusCode, Detail: string(respBody)}
 	}
 
@@ -105,16 +135,26 @@ func (c *Client) doJSON(ctx context.Context, method, path string, reqBody any, r
 // TestConnection checks if the server is reachable and credentials are valid.
 // Returns nil on success, or an error with a descriptive message on failure.
 func (c *Client) TestConnection(ctx context.Context) error {
-	// Step 1: Basic reachability
-	req, err := c.newRequest(ctx, "GET", "/", nil)
+	// Step 1: Basic reachability (hit the server root, not the API prefix)
+	reachURL := c.serverURL
+	if reachURL == "" {
+		reachURL = c.baseURL
+	}
+	log.WithFields(log.Fields{
+		"reach_url": reachURL,
+		"base_url":  c.baseURL,
+	}).Debug("testing connection — step 1: reachability")
+	req, err := http.NewRequestWithContext(ctx, "GET", reachURL+"/", nil)
 	if err != nil {
-		return fmt.Errorf("cannot connect to %s: %w", c.baseURL, err)
+		return fmt.Errorf("cannot connect to %s: %w", reachURL, err)
 	}
 
 	resp, err := c.httpClient.Do(req)
 	if err != nil {
-		return fmt.Errorf("cannot connect to %s — is the server running?", c.baseURL)
+		log.WithError(err).Debug("reachability check failed")
+		return fmt.Errorf("cannot connect to %s — is the server running?", reachURL)
 	}
+	log.WithField("status", resp.StatusCode).Debug("reachability check response")
 	_ = resp.Body.Close()
 
 	serverHeader := strings.ToLower(resp.Header.Get("Server"))
@@ -127,7 +167,8 @@ func (c *Client) TestConnection(ctx context.Context) error {
 	}
 
 	// Step 2: Authenticated check
-	req2, err := c.newRequest(ctx, "GET", "/api/me", nil)
+	log.WithField("url", c.baseURL+"/me").Debug("testing connection — step 2: auth check")
+	req2, err := c.newRequest(ctx, "GET", "/me", nil)
 	if err != nil {
 		return fmt.Errorf("server reachable but API error: %w", err)
 	}
@@ -167,7 +208,7 @@ func (c *Client) TestConnection(ctx context.Context) error {
 // ListAgents returns visible agents.
 func (c *Client) ListAgents(ctx context.Context) ([]models.AgentSummary, error) {
 	var raw []models.AgentSummary
-	if err := c.doJSON(ctx, "GET", "/api/persona", nil, &raw); err != nil {
+	if err := c.doJSON(ctx, "GET", "/persona", nil, &raw); err != nil {
 		return nil, err
 	}
 	var result []models.AgentSummary
@@ -184,7 +225,7 @@ func (c *Client) ListChatSessions(ctx context.Context) ([]models.ChatSessionDeta
 	var resp struct {
 		Sessions []models.ChatSessionDetails `json:"sessions"`
 	}
-	if err := c.doJSON(ctx, "GET", "/api/chat/get-user-chat-sessions", nil, &resp); err != nil {
+	if err := c.doJSON(ctx, "GET", "/chat/get-user-chat-sessions", nil, &resp); err != nil {
 		return nil, err
 	}
 	return resp.Sessions, nil
@@ -193,7 +234,7 @@ func (c *Client) ListChatSessions(ctx context.Context) ([]models.ChatSessionDeta
 // GetChatSession returns full details for a session.
 func (c *Client) GetChatSession(ctx context.Context, sessionID string) (*models.ChatSessionDetailResponse, error) {
 	var resp models.ChatSessionDetailResponse
-	if err := c.doJSON(ctx, "GET", "/api/chat/get-chat-session/"+sessionID, nil, &resp); err != nil {
+	if err := c.doJSON(ctx, "GET", "/chat/get-chat-session/"+sessionID, nil, &resp); err != nil {
 		return nil, err
 	}
 	return &resp, nil
@@ -210,7 +251,7 @@ func (c *Client) RenameChatSession(ctx context.Context, sessionID string, name *
 	var resp struct {
 		NewName string `json:"new_name"`
 	}
-	if err := c.doJSON(ctx, "PUT", "/api/chat/rename-chat-session", payload, &resp); err != nil {
+	if err := c.doJSON(ctx, "PUT", "/chat/rename-chat-session", payload, &resp); err != nil {
 		return "", err
 	}
 	return resp.NewName, nil
@@ -236,7 +277,7 @@ func (c *Client) UploadFile(ctx context.Context, filePath string) (*models.FileD
 	}
 	_ = writer.Close()
 
-	req, err := c.newRequest(ctx, "POST", "/api/user/projects/file/upload", &buf)
+	req, err := c.newRequest(ctx, "POST", "/user/projects/file/upload", &buf)
 	if err != nil {
 		return nil, err
 	}
@@ -275,7 +316,7 @@ func (c *Client) GetBackendVersion(ctx context.Context) (string, error) {
 	var resp struct {
 		BackendVersion string `json:"backend_version"`
 	}
-	if err := c.doJSON(ctx, "GET", "/api/version", nil, &resp); err != nil {
+	if err := c.doJSON(ctx, "GET", "/version", nil, &resp); err != nil {
 		return "", err
 	}
 	return resp.BackendVersion, nil
@@ -283,7 +324,7 @@ func (c *Client) GetBackendVersion(ctx context.Context) (string, error) {
 
 // StopChatSession sends a stop signal for a streaming session (best-effort).
 func (c *Client) StopChatSession(ctx context.Context, sessionID string) {
-	req, err := c.newRequest(ctx, "POST", "/api/chat/stop-chat-session/"+sessionID, nil)
+	req, err := c.newRequest(ctx, "POST", "/chat/stop-chat-session/"+sessionID, nil)
 	if err != nil {
 		return
 	}

cli/internal/api/stream.go

@@ -64,7 +64,7 @@ func (c *Client) SendMessageStream(
 			return
 		}
 
-		req, err := http.NewRequestWithContext(ctx, "POST", c.baseURL+"/api/chat/send-chat-message", nil)
+		req, err := http.NewRequestWithContext(ctx, "POST", c.baseURL+"/chat/send-chat-message", nil)
 		if err != nil {
 			ch <- models.ErrorEvent{Error: fmt.Sprintf("request error: %v", err), IsRetryable: false}
 			return

cli/internal/config/config.go

@@ -10,6 +10,7 @@ import (
 
 const (
 	EnvServerURL      = "ONYX_SERVER_URL"
+	EnvAPIServerURL   = "ONYX_API_SERVER_URL"
 	EnvAPIKey         = "ONYX_API_KEY"
 	EnvAgentID        = "ONYX_PERSONA_ID"
 	EnvSSHHostKey     = "ONYX_SSH_HOST_KEY"
@@ -27,6 +28,7 @@ type Features struct {
 // OnyxCliConfig holds the CLI configuration.
 type OnyxCliConfig struct {
 	ServerURL      string   `json:"server_url"`
+	InternalURL    string   `json:"internal_url,omitempty"`
 	APIKey         string   `json:"api_key"`
 	DefaultAgentID int      `json:"default_persona_id"`
 	Features       Features `json:"features,omitempty"`
@@ -78,30 +80,47 @@ func ConfigExists() bool {
 	return err == nil
 }
 
-// LoadFromDisk reads config from the file only, without applying environment
-// variable overrides. Use this when you need the persisted config values
-// (e.g., to preserve them during a save operation).
-func LoadFromDisk() OnyxCliConfig {
+// LoadFromDisk reads config from the given file path without applying
+// environment variable overrides. Use this when you need the persisted
+// config values (e.g., to preserve them during a save operation).
+// If no path is provided, the default config file path is used.
+func LoadFromDisk(path ...string) OnyxCliConfig {
+	p := ConfigFilePath()
+	if len(path) > 0 && path[0] != "" {
+		p = path[0]
+	}
+
 	cfg := DefaultConfig()
 
-	data, err := os.ReadFile(ConfigFilePath())
+	data, err := os.ReadFile(p)
 	if err == nil {
 		if jsonErr := json.Unmarshal(data, &cfg); jsonErr != nil {
-			fmt.Fprintf(os.Stderr, "warning: config file %s is malformed: %v (using defaults)\n", ConfigFilePath(), jsonErr)
+			fmt.Fprintf(os.Stderr, "warning: config file %s is malformed: %v (using defaults)\n", p, jsonErr)
 		}
 	}
 
 	return cfg
 }
 
-// Load reads config from file and applies environment variable overrides.
-func Load() OnyxCliConfig {
-	cfg := LoadFromDisk()
+// Load reads config from the given file path and applies environment variable
+// overrides. If no path is provided, the default config file path is used.
+func Load(path ...string) OnyxCliConfig {
+	cfg := LoadFromDisk(path...)
+	applyEnvOverrides(&cfg)
+	return cfg
+}
 
-	// Environment overrides
+func applyEnvOverrides(cfg *OnyxCliConfig) {
 	if v := os.Getenv(EnvServerURL); v != "" {
 		cfg.ServerURL = v
 	}
+	// ONYX_API_SERVER_URL takes precedence; fall back to INTERNAL_URL
+	// (the env var used by the web server) for compatibility.
+	if v := os.Getenv(EnvAPIServerURL); v != "" {
+		cfg.InternalURL = v
+	} else if v := os.Getenv("INTERNAL_URL"); v != "" {
+		cfg.InternalURL = v
+	}
 	if v := os.Getenv(EnvAPIKey); v != "" {
 		cfg.APIKey = v
 	}
@@ -117,8 +136,6 @@ func Load() OnyxCliConfig {
 			fmt.Fprintf(os.Stderr, "warning: invalid value %q for %s (expected true/false), ignoring\n", v, EnvStreamMarkdown)
 		}
 	}
-
-	return cfg
 }
 
 // Save writes the config to disk, creating parent directories if needed.

web/src/app/app/message/MultiModelPanel.tsx

@@ -1,126 +0,0 @@
-"use client";
-
-import { useCallback } from "react";
-import { Button } from "@opal/components";
-import { Text } from "@opal/components";
-import { ContentAction } from "@opal/layouts";
-import { SvgEyeOff, SvgX } from "@opal/icons";
-import { getProviderIcon } from "@/app/admin/configuration/llm/utils";
-import AgentMessage, {
-  AgentMessageProps,
-} from "@/app/app/message/messageComponents/AgentMessage";
-import { cn } from "@/lib/utils";
-import { markdown } from "@opal/utils";
-
-export interface MultiModelPanelProps {
-  /** Provider name for icon lookup */
-  provider: string;
-  /** Model name for icon lookup and display */
-  modelName: string;
-  /** Display-friendly model name */
-  displayName: string;
-  /** Whether this panel is the preferred/selected response */
-  isPreferred: boolean;
-  /** Whether this panel is currently hidden */
-  isHidden: boolean;
-  /** Whether this is a non-preferred panel in selection mode (pushed off-screen) */
-  isNonPreferredInSelection: boolean;
-  /** Callback when user clicks this panel to select as preferred */
-  onSelect: () => void;
-  /** Callback to hide/show this panel */
-  onToggleVisibility: () => void;
-  /** Props to pass through to AgentMessage */
-  agentMessageProps: AgentMessageProps;
-}
-
-/**
- * A single model's response panel within the multi-model view.
- *
- * Renders in two states:
- * - **Hidden** — compact header strip only (provider icon + strikethrough name + show button).
- * - **Visible** — full header plus `AgentMessage` body. Clicking anywhere on a
- *   visible non-preferred panel marks it as preferred.
- *
- * The `isNonPreferredInSelection` flag disables pointer events on the body and
- * hides the footer so the panel acts as a passive comparison surface.
- */
-export default function MultiModelPanel({
-  provider,
-  modelName,
-  displayName,
-  isPreferred,
-  isHidden,
-  isNonPreferredInSelection,
-  onSelect,
-  onToggleVisibility,
-  agentMessageProps,
-}: MultiModelPanelProps) {
-  const ProviderIcon = getProviderIcon(provider, modelName);
-
-  const handlePanelClick = useCallback(() => {
-    if (!isHidden && !isPreferred) onSelect();
-  }, [isHidden, isPreferred, onSelect]);
-
-  const header = (
-    <div
-      className={cn(
-        "rounded-12",
-        isPreferred ? "bg-background-tint-02" : "bg-background-tint-00"
-      )}
-    >
-      <ContentAction
-        sizePreset="main-ui"
-        variant="body"
-        paddingVariant="lg"
-        icon={ProviderIcon}
-        title={isHidden ? markdown(`~~${displayName}~~`) : displayName}
-        rightChildren={
-          <div className="flex items-center gap-1 px-2">
-            {isPreferred && (
-              <span className="text-action-link-05 shrink-0">
-                <Text font="secondary-body" color="inherit" nowrap>
-                  Preferred Response
-                </Text>
-              </span>
-            )}
-            {!isPreferred && (
-              <Button
-                prominence="tertiary"
-                icon={isHidden ? SvgEyeOff : SvgX}
-                size="md"
-                onClick={(e) => {
-                  e.stopPropagation();
-                  onToggleVisibility();
-                }}
-                tooltip={isHidden ? "Show response" : "Hide response"}
-              />
-            )}
-          </div>
-        }
-      />
-    </div>
-  );
-
-  // Hidden/collapsed panel — just the header row
-  if (isHidden) {
-    return header;
-  }
-
-  return (
-    <div
-      className={cn(
-        "flex flex-col gap-3 min-w-0 rounded-16 transition-colors",
-        !isPreferred && "cursor-pointer hover:bg-background-tint-02"
-      )}
-      onClick={handlePanelClick}
-    >
-      {header}
-      <div className={cn(isNonPreferredInSelection && "pointer-events-none")}>
-        <AgentMessage
-          {...agentMessageProps}
-          hideFooter={isNonPreferredInSelection}
-        />
-      </div>
-    </div>
-  );
-}

web/src/app/app/message/MultiModelResponseView.tsx

@@ -1,372 +0,0 @@
-"use client";
-
-import { useState, useCallback, useMemo, useEffect, useRef } from "react";
-import { FullChatState } from "@/app/app/message/messageComponents/interfaces";
-import { Message } from "@/app/app/interfaces";
-import { LlmManager } from "@/lib/hooks";
-import { RegenerationFactory } from "@/app/app/message/messageComponents/AgentMessage";
-import MultiModelPanel from "@/app/app/message/MultiModelPanel";
-import { MultiModelResponse } from "@/app/app/message/interfaces";
-import { cn } from "@/lib/utils";
-
-export interface MultiModelResponseViewProps {
-  responses: MultiModelResponse[];
-  chatState: FullChatState;
-  llmManager: LlmManager | null;
-  onRegenerate?: RegenerationFactory;
-  parentMessage?: Message | null;
-  otherMessagesCanSwitchTo?: number[];
-  onMessageSelection?: (nodeId: number) => void;
-  /** Called whenever the set of hidden panel indices changes */
-  onHiddenPanelsChange?: (hidden: Set<number>) => void;
-}
-
-// How many pixels of a non-preferred panel are visible at the viewport edge
-const PEEK_W = 64;
-// Uniform panel width used in the selection-mode carousel
-const SELECTION_PANEL_W = 400;
-// Compact width for hidden panels in the carousel track
-const HIDDEN_PANEL_W = 220;
-// Generation-mode panel widths (from Figma)
-const GEN_PANEL_W_2 = 640; // 2 panels side-by-side
-const GEN_PANEL_W_3 = 436; // 3 panels side-by-side
-// Gap between panels — matches CSS gap-6 (24px)
-const PANEL_GAP = 24;
-// Minimum panel width before horizontal scroll kicks in
-const MIN_PANEL_W = 300;
-
-/**
- * Renders N model responses side-by-side with two layout modes:
- *
- * **Generation mode** — equal-width panels in a horizontally-scrollable row.
- * Panel width is determined by the number of visible (non-hidden) panels.
- *
- * **Selection mode** — activated when the user clicks a panel to mark it as
- * preferred. All panels (including hidden ones) sit in a fixed-width carousel
- * track. A CSS `translateX` transform slides the track so the preferred panel
- * is centered in the viewport; the other panels peek in from the edges through
- * a mask gradient. Non-preferred visible panels are height-capped to the
- * preferred panel's measured height, dimmed at 50% opacity, and receive a
- * bottom fade-out overlay.
- *
- * Hidden panels render as a compact header-only strip at `HIDDEN_PANEL_W` in
- * both modes and are excluded from layout width calculations.
- */
-export default function MultiModelResponseView({
-  responses,
-  chatState,
-  llmManager,
-  onRegenerate,
-  parentMessage,
-  otherMessagesCanSwitchTo,
-  onMessageSelection,
-  onHiddenPanelsChange,
-}: MultiModelResponseViewProps) {
-  const [preferredIndex, setPreferredIndex] = useState<number | null>(null);
-  const [hiddenPanels, setHiddenPanels] = useState<Set<number>>(new Set());
-  // Controls animation: false = panels at start position, true = panels at peek position
-  const [selectionEntered, setSelectionEntered] = useState(false);
-  // Measures the overflow-hidden carousel container for responsive preferred-panel sizing.
-  const [trackContainerW, setTrackContainerW] = useState(0);
-  const roRef = useRef<ResizeObserver | null>(null);
-  const trackContainerRef = useCallback((el: HTMLDivElement | null) => {
-    if (roRef.current) {
-      roRef.current.disconnect();
-      roRef.current = null;
-    }
-    if (!el) return;
-    const ro = new ResizeObserver(([entry]) => {
-      setTrackContainerW(entry?.contentRect.width ?? 0);
-    });
-    ro.observe(el);
-    setTrackContainerW(el.offsetWidth);
-    roRef.current = ro;
-  }, []);
-
-  // Measures the preferred panel's height to cap non-preferred panels in selection mode.
-  const [preferredPanelHeight, setPreferredPanelHeight] = useState<
-    number | null
-  >(null);
-  const preferredRoRef = useRef<ResizeObserver | null>(null);
-  // Tracks which non-preferred panels overflow the preferred height cap
-  const [overflowingPanels, setOverflowingPanels] = useState<Set<number>>(
-    new Set()
-  );
-
-  const preferredPanelRef = useCallback((el: HTMLDivElement | null) => {
-    if (preferredRoRef.current) {
-      preferredRoRef.current.disconnect();
-      preferredRoRef.current = null;
-    }
-    if (!el) {
-      setPreferredPanelHeight(null);
-      return;
-    }
-    const ro = new ResizeObserver(([entry]) => {
-      setPreferredPanelHeight(entry?.contentRect.height ?? 0);
-    });
-    ro.observe(el);
-    setPreferredPanelHeight(el.offsetHeight);
-    preferredRoRef.current = ro;
-  }, []);
-
-  const isGenerating = useMemo(
-    () => responses.some((r) => r.isGenerating),
-    [responses]
-  );
-
-  // Non-hidden responses — used for layout width decisions and selection-mode gating
-  const visibleResponses = useMemo(
-    () => responses.filter((r) => !hiddenPanels.has(r.modelIndex)),
-    [responses, hiddenPanels]
-  );
-
-  const toggleVisibility = useCallback(
-    (modelIndex: number) => {
-      setHiddenPanels((prev) => {
-        const next = new Set(prev);
-        if (next.has(modelIndex)) {
-          next.delete(modelIndex);
-        } else {
-          // Don't hide the last visible panel
-          const visibleCount = responses.length - next.size;
-          if (visibleCount <= 1) return prev;
-          next.add(modelIndex);
-        }
-        onHiddenPanelsChange?.(next);
-        return next;
-      });
-    },
-    [responses.length, onHiddenPanelsChange]
-  );
-
-  const handleSelectPreferred = useCallback(
-    (modelIndex: number) => {
-      if (isGenerating) return;
-      setPreferredIndex(modelIndex);
-      const response = responses.find((r) => r.modelIndex === modelIndex);
-      if (!response) return;
-      if (onMessageSelection) {
-        onMessageSelection(response.nodeId);
-      }
-    },
-    [isGenerating, responses, onMessageSelection]
-  );
-
-  // Clear preferred selection when generation starts
-  useEffect(() => {
-    if (isGenerating) {
-      setPreferredIndex(null);
-    }
-  }, [isGenerating]);
-
-  // Find preferred panel position — used for both the selection guard and carousel layout
-  const preferredIdx = responses.findIndex(
-    (r) => r.modelIndex === preferredIndex
-  );
-
-  // Selection mode when preferred is set, found in responses, not generating, and at least 2 visible panels
-  const showSelectionMode =
-    preferredIndex !== null &&
-    preferredIdx !== -1 &&
-    !isGenerating &&
-    visibleResponses.length > 1;
-
-  // Trigger the slide-out animation one frame after entering selection mode
-  useEffect(() => {
-    if (!showSelectionMode) {
-      setSelectionEntered(false);
-      return;
-    }
-    const raf = requestAnimationFrame(() => setSelectionEntered(true));
-    return () => cancelAnimationFrame(raf);
-  }, [showSelectionMode]);
-
-  // Build panel props — isHidden reflects actual hidden state
-  const buildPanelProps = useCallback(
-    (response: MultiModelResponse, isNonPreferred: boolean) => ({
-      provider: response.provider,
-      modelName: response.modelName,
-      displayName: response.displayName,
-      isPreferred: preferredIndex === response.modelIndex,
-      isHidden: hiddenPanels.has(response.modelIndex),
-      isNonPreferredInSelection: isNonPreferred,
-      onSelect: () => handleSelectPreferred(response.modelIndex),
-      onToggleVisibility: () => toggleVisibility(response.modelIndex),
-      agentMessageProps: {
-        rawPackets: response.packets,
-        packetCount: response.packetCount,
-        chatState,
-        nodeId: response.nodeId,
-        messageId: response.messageId,
-        currentFeedback: response.currentFeedback,
-        llmManager,
-        otherMessagesCanSwitchTo,
-        onMessageSelection,
-        onRegenerate,
-        parentMessage,
-      },
-    }),
-    [
-      preferredIndex,
-      hiddenPanels,
-      handleSelectPreferred,
-      toggleVisibility,
-      chatState,
-      llmManager,
-      otherMessagesCanSwitchTo,
-      onMessageSelection,
-      onRegenerate,
-      parentMessage,
-    ]
-  );
-
-  if (showSelectionMode) {
-    // ── Selection Layout (transform-based carousel) ──
-    //
-    // All panels (including hidden) sit in the track at their original A/B/C positions.
-    // Hidden panels use HIDDEN_PANEL_W; non-preferred use SELECTION_PANEL_W;
-    // preferred uses dynamicPrefW (up to GEN_PANEL_W_2).
-    const n = responses.length;
-
-    const dynamicPrefW =
-      trackContainerW > 0
-        ? Math.min(trackContainerW - 2 * (PEEK_W + PANEL_GAP), GEN_PANEL_W_2)
-        : GEN_PANEL_W_2;
-
-    const selectionWidths = responses.map((r, i) => {
-      if (hiddenPanels.has(r.modelIndex)) return HIDDEN_PANEL_W;
-      if (i === preferredIdx) return dynamicPrefW;
-      return SELECTION_PANEL_W;
-    });
-
-    const panelLeftEdges = selectionWidths.reduce<number[]>((acc, w, i) => {
-      acc.push(i === 0 ? 0 : acc[i - 1]! + selectionWidths[i - 1]! + PANEL_GAP);
-      return acc;
-    }, []);
-
-    const preferredCenterInTrack =
-      panelLeftEdges[preferredIdx]! + selectionWidths[preferredIdx]! / 2;
-
-    // Start position: hidden panels at HIDDEN_PANEL_W, visible at SELECTION_PANEL_W
-    const uniformTrackW =
-      responses.reduce(
-        (sum, r) =>
-          sum +
-          (hiddenPanels.has(r.modelIndex) ? HIDDEN_PANEL_W : SELECTION_PANEL_W),
-        0
-      ) +
-      (n - 1) * PANEL_GAP;
-
-    const trackTransform = selectionEntered
-      ? `translateX(${trackContainerW / 2 - preferredCenterInTrack}px)`
-      : `translateX(${(trackContainerW - uniformTrackW) / 2}px)`;
-
-    return (
-      <div
-        ref={trackContainerRef}
-        className="w-full overflow-hidden"
-        style={{
-          maskImage: `linear-gradient(to right, transparent 0px, black ${PEEK_W}px, black calc(100% - ${PEEK_W}px), transparent 100%)`,
-          WebkitMaskImage: `linear-gradient(to right, transparent 0px, black ${PEEK_W}px, black calc(100% - ${PEEK_W}px), transparent 100%)`,
-        }}
-      >
-        <div
-          className="flex items-start"
-          style={{
-            gap: `${PANEL_GAP}px`,
-            transition: selectionEntered
-              ? "transform 0.45s cubic-bezier(0.2, 0, 0, 1)"
-              : "none",
-            transform: trackTransform,
-          }}
-        >
-          {responses.map((r, i) => {
-            const isHidden = hiddenPanels.has(r.modelIndex);
-            const isPref = r.modelIndex === preferredIndex;
-            const isNonPref = !isHidden && !isPref;
-            const finalW = selectionWidths[i]!;
-            const startW = isHidden ? HIDDEN_PANEL_W : SELECTION_PANEL_W;
-            const capped = isNonPref && preferredPanelHeight != null;
-            const overflows = capped && overflowingPanels.has(r.modelIndex);
-            return (
-              <div
-                key={r.modelIndex}
-                ref={(el) => {
-                  if (isPref) preferredPanelRef(el);
-                  if (capped && el) {
-                    const doesOverflow = el.scrollHeight > el.clientHeight;
-                    setOverflowingPanels((prev) => {
-                      const had = prev.has(r.modelIndex);
-                      if (doesOverflow === had) return prev;
-                      const next = new Set(prev);
-                      if (doesOverflow) next.add(r.modelIndex);
-                      else next.delete(r.modelIndex);
-                      return next;
-                    });
-                  }
-                }}
-                style={{
-                  width: `${selectionEntered ? finalW : startW}px`,
-                  flexShrink: 0,
-                  transition: selectionEntered
-                    ? "width 0.45s cubic-bezier(0.2, 0, 0, 1)"
-                    : "none",
-                  maxHeight: capped ? preferredPanelHeight : undefined,
-                  overflow: capped ? "hidden" : undefined,
-                  position: capped ? "relative" : undefined,
-                }}
-              >
-                <div className={cn(isNonPref && "opacity-50")}>
-                  <MultiModelPanel {...buildPanelProps(r, isNonPref)} />
-                </div>
-                {overflows && (
-                  <div
-                    className="absolute inset-x-0 bottom-0 h-24 pointer-events-none"
-                    style={{
-                      background:
-                        "linear-gradient(to top, var(--background-tint-01) 0%, transparent 100%)",
-                    }}
-                  />
-                )}
-              </div>
-            );
-          })}
-        </div>
-      </div>
-    );
-  }
-
-  // ── Generation Layout (equal panels side-by-side) ──
-  // Panel width based on number of visible (non-hidden) panels.
-  const panelWidth =
-    visibleResponses.length <= 2 ? GEN_PANEL_W_2 : GEN_PANEL_W_3;
-
-  return (
-    <div className="overflow-x-auto">
-      <div className="flex gap-6 items-start w-fit mx-auto">
-        {responses.map((r) => {
-          const isHidden = hiddenPanels.has(r.modelIndex);
-          return (
-            <div
-              key={r.modelIndex}
-              style={
-                isHidden
-                  ? {
-                      width: HIDDEN_PANEL_W,
-                      minWidth: HIDDEN_PANEL_W,
-                      maxWidth: HIDDEN_PANEL_W,
-                      flexShrink: 0,
-                      overflow: "hidden" as const,
-                    }
-                  : { width: panelWidth, minWidth: MIN_PANEL_W }
-              }
-            >
-              <MultiModelPanel {...buildPanelProps(r, false)} />
-            </div>
-          );
-        })}
-      </div>
-    </div>
-  );
-}

web/src/app/app/message/interfaces.ts

@@ -1,16 +0,0 @@
-import { Packet } from "@/app/app/services/streamingModels";
-import { FeedbackType } from "@/app/app/interfaces";
-
-export interface MultiModelResponse {
-  modelIndex: number;
-  provider: string;
-  modelName: string;
-  displayName: string;
-  packets: Packet[];
-  packetCount: number;
-  nodeId: number;
-  messageId?: number;
-
-  currentFeedback?: FeedbackType | null;
-  isGenerating?: boolean;
-}

web/src/app/app/message/messageComponents/AgentMessage.tsx

@@ -49,8 +49,6 @@ export interface AgentMessageProps {
   parentMessage?: Message | null;
   // Duration in seconds for processing this message (agent messages only)
   processingDurationSeconds?: number;
-  /** Hide the feedback/toolbar footer (used in multi-model non-preferred panels) */
-  hideFooter?: boolean;
 }
 
 // TODO: Consider more robust comparisons:
@@ -78,8 +76,7 @@ function arePropsEqual(
     prev.parentMessage?.messageId === next.parentMessage?.messageId &&
     prev.llmManager?.isLoadingProviders ===
       next.llmManager?.isLoadingProviders &&
-    prev.processingDurationSeconds === next.processingDurationSeconds &&
-    prev.hideFooter === next.hideFooter
+    prev.processingDurationSeconds === next.processingDurationSeconds
     // Skip: chatState.regenerate, chatState.setPresentingDocument,
     //       most of llmManager, onMessageSelection (function/object props)
   );
@@ -98,7 +95,6 @@ const AgentMessage = React.memo(function AgentMessage({
   onRegenerate,
   parentMessage,
   processingDurationSeconds,
-  hideFooter,
 }: AgentMessageProps) {
   const markdownRef = useRef<HTMLDivElement>(null);
   const finalAnswerRef = useRef<HTMLDivElement>(null);
@@ -330,7 +326,7 @@ const AgentMessage = React.memo(function AgentMessage({
       </div>
 
       {/* Feedback buttons - only show when streaming and rendering complete */}
-      {isComplete && !hideFooter && (
+      {isComplete && (
         <MessageToolbar
           nodeId={nodeId}
           messageId={messageId}

web/src/components/admin/federated/FederatedConnectorForm.tsx

@@ -133,7 +133,7 @@ async function createFederatedConnector(
 
 async function updateFederatedConnector(
   id: number,
-  credentials: CredentialForm | null,
+  credentials: CredentialForm,
   config?: ConfigForm
 ): Promise<{ success: boolean; message: string }> {
   try {
@@ -143,7 +143,7 @@ async function updateFederatedConnector(
         "Content-Type": "application/json",
       },
       body: JSON.stringify({
-        credentials: credentials ?? undefined,
+        credentials,
         config: config || {},
       }),
     });
@@ -201,9 +201,7 @@ export function FederatedConnectorForm({
   const isEditMode = connectorId !== undefined;
 
   const [formState, setFormState] = useState<FormState>({
-    // In edit mode, don't populate credentials with masked values from the API.
-    // Masked values (e.g. "••••••••••••") would be saved back and corrupt the real credentials.
-    credentials: isEditMode ? {} : preloadedConnectorData?.credentials || {},
+    credentials: preloadedConnectorData?.credentials || {},
     config: preloadedConnectorData?.config || {},
     schema: preloadedCredentialSchema?.credentials || null,
     configurationSchema: null,
@@ -211,7 +209,6 @@ export function FederatedConnectorForm({
     configurationSchemaError: null,
     connectorError: null,
   });
-  const [credentialsModified, setCredentialsModified] = useState(false);
   const [isSubmitting, setIsSubmitting] = useState(false);
   const [submitMessage, setSubmitMessage] = useState<string | null>(null);
   const [submitSuccess, setSubmitSuccess] = useState<boolean | null>(null);
@@ -336,7 +333,6 @@ export function FederatedConnectorForm({
   }
 
   const handleCredentialChange = (key: string, value: string) => {
-    setCredentialsModified(true);
     setFormState((prev) => ({
       ...prev,
       credentials: {
@@ -358,11 +354,6 @@ export function FederatedConnectorForm({
 
   const handleValidateCredentials = async () => {
     if (!formState.schema) return;
-    if (isEditMode && !credentialsModified) {
-      setSubmitMessage("Enter new credential values before validating.");
-      setSubmitSuccess(false);
-      return;
-    }
 
     setIsValidating(true);
     setSubmitMessage(null);
@@ -420,10 +411,8 @@ export function FederatedConnectorForm({
     setSubmitSuccess(null);
 
     try {
-      const shouldValidateCredentials = !isEditMode || credentialsModified;
-
-      // Validate required fields (skip for credentials in edit mode when unchanged)
-      if (formState.schema && shouldValidateCredentials) {
+      // Validate required fields
+      if (formState.schema) {
         const missingRequired = Object.entries(formState.schema)
           .filter(
             ([key, field]) => field.required && !formState.credentials[key]
@@ -453,20 +442,16 @@ export function FederatedConnectorForm({
       }
       setConfigValidationErrors({});
 
-      // Validate credentials before creating/updating (skip in edit mode when unchanged)
-      if (shouldValidateCredentials) {
-        const validation = await validateCredentials(
-          connector,
-          formState.credentials
-        );
-        if (!validation.success) {
-          setSubmitMessage(
-            `Credential validation failed: ${validation.message}`
-          );
-          setSubmitSuccess(false);
-          setIsSubmitting(false);
-          return;
-        }
+      // Validate credentials before creating/updating
+      const validation = await validateCredentials(
+        connector,
+        formState.credentials
+      );
+      if (!validation.success) {
+        setSubmitMessage(`Credential validation failed: ${validation.message}`);
+        setSubmitSuccess(false);
+        setIsSubmitting(false);
+        return;
       }
 
       // Create or update the connector
@@ -474,7 +459,7 @@ export function FederatedConnectorForm({
         isEditMode && connectorId
           ? await updateFederatedConnector(
               connectorId,
-              credentialsModified ? formState.credentials : null,
+              formState.credentials,
               formState.config
             )
           : await createFederatedConnector(
@@ -553,16 +538,14 @@ export function FederatedConnectorForm({
               id={fieldKey}
               type={fieldSpec.secret ? "password" : "text"}
               placeholder={
-                isEditMode && !credentialsModified
-                  ? "••••••••  (leave blank to keep current value)"
-                  : fieldSpec.example
-                    ? String(fieldSpec.example)
-                    : fieldSpec.description
+                fieldSpec.example
+                  ? String(fieldSpec.example)
+                  : fieldSpec.description
               }
               value={formState.credentials[fieldKey] || ""}
               onChange={(e) => handleCredentialChange(fieldKey, e.target.value)}
               className="w-96"
-              required={!isEditMode && fieldSpec.required}
+              required={fieldSpec.required}
             />
           </div>
         ))}

web/src/components/chat/ProviderContext.tsx

@@ -1,10 +1,25 @@
 "use client";
-import { LLMProviderDescriptor } from "@/interfaces/llm";
-import React, { createContext, useContext, useCallback } from "react";
+import {
+  WellKnownLLMProviderDescriptor,
+  LLMProviderDescriptor,
+} from "@/interfaces/llm";
+import React, {
+  createContext,
+  useContext,
+  useState,
+  useEffect,
+  useCallback,
+} from "react";
+import { useUser } from "@/providers/UserProvider";
 import { useLLMProviders } from "@/hooks/useLLMProviders";
+import { useLLMProviderOptions } from "@/lib/hooks/useLLMProviderOptions";
+import { testDefaultProvider as testDefaultProviderSvc } from "@/lib/llmConfig/svc";
 
 interface ProviderContextType {
+  shouldShowConfigurationNeeded: boolean;
+  providerOptions: WellKnownLLMProviderDescriptor[];
   refreshProviderInfo: () => Promise<void>;
+  // Expose configured provider instances for components that need it (e.g., onboarding)
   llmProviders: LLMProviderDescriptor[] | undefined;
   isLoadingProviders: boolean;
   hasProviders: boolean;
@@ -14,26 +29,79 @@ const ProviderContext = createContext<ProviderContextType | undefined>(
   undefined
 );
 
+const DEFAULT_LLM_PROVIDER_TEST_COMPLETE_KEY = "defaultLlmProviderTestComplete";
+
+function checkDefaultLLMProviderTestComplete() {
+  if (typeof window === "undefined") return true;
+  return (
+    localStorage.getItem(DEFAULT_LLM_PROVIDER_TEST_COMPLETE_KEY) === "true"
+  );
+}
+
+function setDefaultLLMProviderTestComplete() {
+  if (typeof window === "undefined") return;
+  localStorage.setItem(DEFAULT_LLM_PROVIDER_TEST_COMPLETE_KEY, "true");
+}
+
 export function ProviderContextProvider({
   children,
 }: {
   children: React.ReactNode;
 }) {
+  const { user } = useUser();
+
+  // Use SWR hooks instead of raw fetch
   const {
     llmProviders,
     isLoading: isLoadingProviders,
     refetch: refetchProviders,
   } = useLLMProviders();
+  const { llmProviderOptions: providerOptions, refetch: refetchOptions } =
+    useLLMProviderOptions();
+
+  const [defaultCheckSuccessful, setDefaultCheckSuccessful] =
+    useState<boolean>(true);
+
+  // Test the default provider - only runs if test hasn't passed yet
+  const testDefaultProvider = useCallback(async () => {
+    const shouldCheck =
+      !checkDefaultLLMProviderTestComplete() &&
+      (!user || user.role === "admin");
+
+    if (shouldCheck) {
+      const success = await testDefaultProviderSvc();
+      setDefaultCheckSuccessful(success);
+      if (success) {
+        setDefaultLLMProviderTestComplete();
+      }
+    }
+  }, [user]);
+
+  // Test default provider on mount
+  useEffect(() => {
+    testDefaultProvider();
+  }, [testDefaultProvider]);
 
   const hasProviders = (llmProviders?.length ?? 0) > 0;
+  const validProviderExists = hasProviders && defaultCheckSuccessful;
+
+  const shouldShowConfigurationNeeded =
+    !validProviderExists && (providerOptions?.length ?? 0) > 0;
 
   const refreshProviderInfo = useCallback(async () => {
-    await refetchProviders();
-  }, [refetchProviders]);
+    // Refetch provider lists and re-test default provider if needed
+    await Promise.all([
+      refetchProviders(),
+      refetchOptions(),
+      testDefaultProvider(),
+    ]);
+  }, [refetchProviders, refetchOptions, testDefaultProvider]);
 
   return (
     <ProviderContext.Provider
       value={{
+        shouldShowConfigurationNeeded,
+        providerOptions: providerOptions ?? [],
         refreshProviderInfo,
         llmProviders,
         isLoadingProviders,

web/src/hooks/__tests__/useShowOnboarding.test.tsx

@@ -17,6 +17,7 @@ const mockProviderStatus = {
   llmProviders: [] as unknown[],
   isLoadingProviders: false,
   hasProviders: false,
+  providerOptions: [],
   refreshProviderInfo: jest.fn(),
 };
 
@@ -70,6 +71,7 @@ describe("useShowOnboarding", () => {
     mockProviderStatus.llmProviders = [];
     mockProviderStatus.isLoadingProviders = false;
     mockProviderStatus.hasProviders = false;
+    mockProviderStatus.providerOptions = [];
   });
 
   it("returns showOnboarding=false while providers are loading", () => {
@@ -196,6 +198,7 @@ describe("useShowOnboarding", () => {
       OnboardingStep.Welcome
     );
     expect(result.current.onboardingActions).toBeDefined();
+    expect(result.current.llmDescriptors).toEqual([]);
   });
 
   describe("localStorage persistence", () => {

web/src/hooks/useMultiModelChat.ts

@@ -1,192 +0,0 @@
-"use client";
-
-import { useState, useCallback, useEffect, useMemo } from "react";
-import {
-  MAX_MODELS,
-  SelectedModel,
-} from "@/refresh-components/popovers/ModelSelector";
-import { LLMOverride } from "@/app/app/services/lib";
-import { LlmManager } from "@/lib/hooks";
-import { buildLlmOptions } from "@/refresh-components/popovers/LLMPopover";
-
-export interface UseMultiModelChatReturn {
-  /** Currently selected models for multi-model comparison. */
-  selectedModels: SelectedModel[];
-  /** Whether multi-model mode is active (>1 model selected). */
-  isMultiModelActive: boolean;
-  /** Add a model to the selection. */
-  addModel: (model: SelectedModel) => void;
-  /** Remove a model by index. */
-  removeModel: (index: number) => void;
-  /** Replace a model at a specific index with a new one. */
-  replaceModel: (index: number, model: SelectedModel) => void;
-  /** Clear all selected models. */
-  clearModels: () => void;
-  /** Build the LLMOverride[] array from selectedModels. */
-  buildLlmOverrides: () => LLMOverride[];
-  /**
-   * Restore multi-model selection from model version strings (e.g. from chat history).
-   * Matches against available llmOptions to reconstruct full SelectedModel objects.
-   */
-  restoreFromModelNames: (modelNames: string[]) => void;
-  /**
-   * Switch to a single model by name (after user picks a preferred response).
-   * Matches against llmOptions to find the full SelectedModel.
-   */
-  selectSingleModel: (modelName: string) => void;
-}
-
-export default function useMultiModelChat(
-  llmManager: LlmManager
-): UseMultiModelChatReturn {
-  const [selectedModels, setSelectedModels] = useState<SelectedModel[]>([]);
-  const [defaultInitialized, setDefaultInitialized] = useState(false);
-
-  // Initialize with the default model from llmManager once providers load
-  const llmOptions = useMemo(
-    () =>
-      llmManager.llmProviders ? buildLlmOptions(llmManager.llmProviders) : [],
-    [llmManager.llmProviders]
-  );
-
-  useEffect(() => {
-    if (defaultInitialized) return;
-    if (llmOptions.length === 0) return;
-    const { currentLlm } = llmManager;
-    // Don't initialize if currentLlm hasn't loaded yet
-    if (!currentLlm.modelName) return;
-    const match = llmOptions.find(
-      (opt) =>
-        opt.provider === currentLlm.provider &&
-        opt.modelName === currentLlm.modelName
-    );
-    if (match) {
-      setSelectedModels([
-        {
-          name: match.name,
-          provider: match.provider,
-          modelName: match.modelName,
-          displayName: match.displayName,
-        },
-      ]);
-      setDefaultInitialized(true);
-    }
-  }, [llmOptions, llmManager.currentLlm, defaultInitialized]);
-
-  const isMultiModelActive = selectedModels.length > 1;
-
-  const addModel = useCallback((model: SelectedModel) => {
-    setSelectedModels((prev) => {
-      if (prev.length >= MAX_MODELS) return prev;
-      if (
-        prev.some(
-          (m) =>
-            m.provider === model.provider && m.modelName === model.modelName
-        )
-      ) {
-        return prev;
-      }
-      return [...prev, model];
-    });
-  }, []);
-
-  const removeModel = useCallback((index: number) => {
-    setSelectedModels((prev) => prev.filter((_, i) => i !== index));
-  }, []);
-
-  const replaceModel = useCallback((index: number, model: SelectedModel) => {
-    setSelectedModels((prev) => {
-      // Don't replace with a model that's already selected elsewhere
-      if (
-        prev.some(
-          (m, i) =>
-            i !== index &&
-            m.provider === model.provider &&
-            m.modelName === model.modelName
-        )
-      ) {
-        return prev;
-      }
-      const next = [...prev];
-      next[index] = model;
-      return next;
-    });
-  }, []);
-
-  const clearModels = useCallback(() => {
-    setSelectedModels([]);
-  }, []);
-
-  const restoreFromModelNames = useCallback(
-    (modelNames: string[]) => {
-      if (modelNames.length < 2 || llmOptions.length === 0) return;
-      const restored: SelectedModel[] = [];
-      for (const name of modelNames) {
-        // Try matching by modelName (raw version string like "claude-opus-4-6")
-        // or by displayName (friendly name like "Claude Opus 4.6")
-        const match = llmOptions.find(
-          (opt) =>
-            opt.modelName === name ||
-            opt.displayName === name ||
-            opt.name === name
-        );
-        if (match) {
-          restored.push({
-            name: match.name,
-            provider: match.provider,
-            modelName: match.modelName,
-            displayName: match.displayName,
-          });
-        }
-      }
-      if (restored.length >= 2) {
-        setSelectedModels(restored.slice(0, MAX_MODELS));
-        setDefaultInitialized(true);
-      }
-    },
-    [llmOptions]
-  );
-
-  const selectSingleModel = useCallback(
-    (modelName: string) => {
-      if (llmOptions.length === 0) return;
-      const match = llmOptions.find(
-        (opt) =>
-          opt.modelName === modelName ||
-          opt.displayName === modelName ||
-          opt.name === modelName
-      );
-      if (match) {
-        setSelectedModels([
-          {
-            name: match.name,
-            provider: match.provider,
-            modelName: match.modelName,
-            displayName: match.displayName,
-          },
-        ]);
-      }
-    },
-    [llmOptions]
-  );
-
-  const buildLlmOverrides = useCallback((): LLMOverride[] => {
-    return selectedModels.map((m) => ({
-      model_provider: m.provider,
-      model_version: m.modelName,
-      display_name: m.displayName,
-    }));
-  }, [selectedModels]);
-
-  return {
-    selectedModels,
-    isMultiModelActive,
-    addModel,
-    removeModel,
-    replaceModel,
-    clearModels,
-    buildLlmOverrides,
-    restoreFromModelNames,
-    selectSingleModel,
-  };
-}

web/src/hooks/useShowOnboarding.ts

@@ -9,6 +9,7 @@ import {
   OnboardingState,
   OnboardingStep,
 } from "@/interfaces/onboarding";
+import { WellKnownLLMProviderDescriptor } from "@/interfaces/llm";
 import { updateUserPersonalization } from "@/lib/userSettings";
 import { useUser } from "@/providers/UserProvider";
 import { MinimalPersonaSnapshot } from "@/app/admin/agents/interfaces";
@@ -21,6 +22,7 @@ function getOnboardingCompletedKey(userId: string): string {
 
 function useOnboardingState(liveAgent?: MinimalPersonaSnapshot): {
   state: OnboardingState;
+  llmDescriptors: WellKnownLLMProviderDescriptor[];
   actions: OnboardingActions;
   isLoading: boolean;
   hasProviders: boolean;
@@ -33,6 +35,7 @@ function useOnboardingState(liveAgent?: MinimalPersonaSnapshot): {
     llmProviders,
     isLoadingProviders,
     hasProviders: hasLlmProviders,
+    providerOptions,
     refreshProviderInfo,
   } = useProviderStatus();
 
@@ -40,6 +43,7 @@ function useOnboardingState(liveAgent?: MinimalPersonaSnapshot): {
   const { refetch: refreshPersonaProviders } = useLLMProviders(liveAgent?.id);
 
   const userName = user?.personalization?.name;
+  const llmDescriptors = providerOptions;
 
   const nameUpdateTimeoutRef = useRef<ReturnType<typeof setTimeout> | null>(
     null
@@ -231,6 +235,7 @@ function useOnboardingState(liveAgent?: MinimalPersonaSnapshot): {
 
   return {
     state,
+    llmDescriptors,
     actions: {
       nextStep,
       prevStep,
@@ -275,6 +280,7 @@ export function useShowOnboarding({
   const {
     state: onboardingState,
     actions: onboardingActions,
+    llmDescriptors,
     isLoading: isLoadingOnboarding,
     hasProviders: hasAnyProvider,
   } = useOnboardingState(liveAgent);
@@ -344,6 +350,7 @@ export function useShowOnboarding({
     onboardingDismissed,
     onboardingState,
     onboardingActions,
+    llmDescriptors,
     isLoadingOnboarding,
     hideOnboarding,
     finishOnboarding,

web/src/refresh-components/popovers/LLMPopover.tsx

@@ -1,7 +1,7 @@
 "use client";
 
 import { useState, useEffect, useCallback, useMemo, useRef } from "react";
-import Popover from "@/refresh-components/Popover";
+import Popover, { PopoverMenu } from "@/refresh-components/Popover";
 import { LlmDescriptor, LlmManager } from "@/lib/hooks";
 import { structureValue } from "@/lib/llmConfig/utils";
 import {
@@ -11,11 +11,25 @@ import {
 import { LLMProviderDescriptor } from "@/interfaces/llm";
 import { Slider } from "@/components/ui/slider";
 import { useUser } from "@/providers/UserProvider";
+import LineItem from "@/refresh-components/buttons/LineItem";
+import InputTypeIn from "@/refresh-components/inputs/InputTypeIn";
 import Text from "@/refresh-components/texts/Text";
-import { SvgRefreshCw } from "@opal/icons";
+import SimpleLoader from "@/refresh-components/loaders/SimpleLoader";
+import {
+  Accordion,
+  AccordionContent,
+  AccordionItem,
+  AccordionTrigger,
+} from "@/components/ui/accordion";
+import {
+  SvgCheck,
+  SvgChevronDown,
+  SvgChevronRight,
+  SvgRefreshCw,
+} from "@opal/icons";
+import { Section } from "@/layouts/general-layouts";
 import { OpenButton } from "@opal/components";
 import { LLMOption, LLMOptionGroup } from "./interfaces";
-import ModelListContent from "./ModelListContent";
 
 export interface LLMPopoverProps {
   llmManager: LlmManager;
@@ -136,6 +150,7 @@ export default function LLMPopover({
   const isLoadingProviders = llmManager.isLoadingProviders;
 
   const [open, setOpen] = useState(false);
+  const [searchQuery, setSearchQuery] = useState("");
   const { user } = useUser();
 
   const [localTemperature, setLocalTemperature] = useState(
@@ -146,7 +161,9 @@ export default function LLMPopover({
     setLocalTemperature(llmManager.temperature ?? 0.5);
   }, [llmManager.temperature]);
 
+  const searchInputRef = useRef<HTMLInputElement>(null);
   const scrollContainerRef = useRef<HTMLDivElement>(null);
+  const selectedItemRef = useRef<HTMLDivElement>(null);
 
   const handleGlobalTemperatureChange = useCallback((value: number[]) => {
     const value_0 = value[0];
@@ -165,28 +182,39 @@ export default function LLMPopover({
     [llmManager]
   );
 
-  const isSelected = useCallback(
-    (option: LLMOption) =>
-      option.modelName === llmManager.currentLlm.modelName &&
-      option.provider === llmManager.currentLlm.provider,
-    [llmManager.currentLlm.modelName, llmManager.currentLlm.provider]
+  const llmOptions = useMemo(
+    () => buildLlmOptions(llmProviders, currentModelName),
+    [llmProviders, currentModelName]
   );
 
-  const handleSelectModel = useCallback(
-    (option: LLMOption) => {
-      llmManager.updateCurrentLlm({
-        modelName: option.modelName,
-        provider: option.provider,
-        name: option.name,
-      } as LlmDescriptor);
-      onSelect?.(
-        structureValue(option.name, option.provider, option.modelName)
+  // Filter options by vision capability (when images are uploaded) and search query
+  const filteredOptions = useMemo(() => {
+    let result = llmOptions;
+    if (requiresImageInput) {
+      result = result.filter((opt) => opt.supportsImageInput);
+    }
+    if (searchQuery.trim()) {
+      const query = searchQuery.toLowerCase();
+      result = result.filter(
+        (opt) =>
+          opt.displayName.toLowerCase().includes(query) ||
+          opt.modelName.toLowerCase().includes(query) ||
+          (opt.vendor && opt.vendor.toLowerCase().includes(query))
       );
-      setOpen(false);
-    },
-    [llmManager, onSelect]
+    }
+    return result;
+  }, [llmOptions, searchQuery, requiresImageInput]);
+
+  // Group options by provider using backend-provided display names and ordering
+  // For aggregator providers (bedrock, openrouter, vertex_ai), flatten to "Provider/Vendor" format
+  const groupedOptions = useMemo(
+    () => groupLlmOptions(filteredOptions),
+    [filteredOptions]
   );
 
+  // Get display name for the model to show in the button
+  // Use currentModelName prop if provided (e.g., for regenerate showing the model used),
+  // otherwise fall back to the globally selected model
   const currentLlmDisplayName = useMemo(() => {
     // Only use currentModelName if it's a non-empty string
     const currentModel =
@@ -206,30 +234,122 @@ export default function LLMPopover({
     return currentModel;
   }, [llmProviders, currentModelName, llmManager.currentLlm.modelName]);
 
-  const temperatureFooter = user?.preferences?.temperature_override_enabled ? (
-    <>
-      <div className="border-t border-border-02 mx-2" />
-      <div className="flex flex-col w-full py-2 gap-2">
-        <Slider
-          value={[localTemperature]}
-          max={llmManager.maxTemperature}
-          min={0}
-          step={0.01}
-          onValueChange={handleGlobalTemperatureChange}
-          onValueCommit={handleGlobalTemperatureCommit}
-          className="w-full"
-        />
-        <div className="flex flex-row items-center justify-between">
-          <Text secondaryBody text03>
-            Temperature (creativity)
-          </Text>
-          <Text secondaryBody text03>
-            {localTemperature.toFixed(1)}
-          </Text>
-        </div>
+  // Determine which group the current model belongs to (for auto-expand)
+  const currentGroupKey = useMemo(() => {
+    const currentModel = llmManager.currentLlm.modelName;
+    const currentProvider = llmManager.currentLlm.provider;
+    // Match by both modelName AND provider to handle same model name across providers
+    const option = llmOptions.find(
+      (o) => o.modelName === currentModel && o.provider === currentProvider
+    );
+    if (!option) return "openai";
+
+    const provider = option.provider.toLowerCase();
+    const isAggregator = AGGREGATOR_PROVIDERS.has(provider);
+
+    if (isAggregator && option.vendor) {
+      return `${provider}/${option.vendor.toLowerCase()}`;
+    }
+    return provider;
+  }, [
+    llmOptions,
+    llmManager.currentLlm.modelName,
+    llmManager.currentLlm.provider,
+  ]);
+
+  // Track expanded groups - initialize with current model's group
+  const [expandedGroups, setExpandedGroups] = useState<string[]>([
+    currentGroupKey,
+  ]);
+
+  // Reset state when popover closes/opens
+  useEffect(() => {
+    if (!open) {
+      setSearchQuery("");
+    } else {
+      // Reset expanded groups to only show the selected model's group
+      setExpandedGroups([currentGroupKey]);
+    }
+  }, [open, currentGroupKey]);
+
+  // Auto-scroll to selected model when popover opens
+  useEffect(() => {
+    if (open) {
+      // Small delay to let accordion content render
+      const timer = setTimeout(() => {
+        selectedItemRef.current?.scrollIntoView({
+          behavior: "instant",
+          block: "center",
+        });
+      }, 50);
+      return () => clearTimeout(timer);
+    }
+  }, [open]);
+
+  const isSearching = searchQuery.trim().length > 0;
+
+  // Compute final expanded groups
+  const effectiveExpandedGroups = useMemo(() => {
+    if (isSearching) {
+      // Force expand all when searching
+      return groupedOptions.map((g) => g.key);
+    }
+    return expandedGroups;
+  }, [isSearching, groupedOptions, expandedGroups]);
+
+  // Handler for accordion changes
+  const handleAccordionChange = (value: string[]) => {
+    // Only update state when not searching (force-expanding)
+    if (!isSearching) {
+      setExpandedGroups(value);
+    }
+  };
+
+  const handleSelectModel = (option: LLMOption) => {
+    llmManager.updateCurrentLlm({
+      modelName: option.modelName,
+      provider: option.provider,
+      name: option.name,
+    } as LlmDescriptor);
+    onSelect?.(structureValue(option.name, option.provider, option.modelName));
+    setOpen(false);
+  };
+
+  const renderModelItem = (option: LLMOption) => {
+    const isSelected =
+      option.modelName === llmManager.currentLlm.modelName &&
+      option.provider === llmManager.currentLlm.provider;
+
+    const capabilities: string[] = [];
+    if (option.supportsReasoning) {
+      capabilities.push("Reasoning");
+    }
+    if (option.supportsImageInput) {
+      capabilities.push("Vision");
+    }
+    const description =
+      capabilities.length > 0 ? capabilities.join(", ") : undefined;
+
+    return (
+      <div
+        key={`${option.name}-${option.modelName}`}
+        ref={isSelected ? selectedItemRef : undefined}
+      >
+        <LineItem
+          selected={isSelected}
+          description={description}
+          onClick={() => handleSelectModel(option)}
+          rightChildren={
+            isSelected ? (
+              <SvgCheck className="h-4 w-4 stroke-action-link-05 shrink-0" />
+            ) : null
+          }
+        >
+          {option.displayName}
+        </LineItem>
       </div>
-    </>
-  ) : undefined;
+    );
+  };
 
   return (
     <Popover open={open} onOpenChange={setOpen}>
@@ -253,16 +373,129 @@ export default function LLMPopover({
       </div>
 
       <Popover.Content side="top" align="end" width="xl">
-        <ModelListContent
-          llmProviders={llmProviders}
-          currentModelName={currentModelName}
-          requiresImageInput={requiresImageInput}
-          isLoading={isLoadingProviders}
-          onSelect={handleSelectModel}
-          isSelected={isSelected}
-          scrollContainerRef={scrollContainerRef}
-          footer={temperatureFooter}
-        />
+        <Section gap={0.5}>
+          {/* Search Input */}
+          <InputTypeIn
+            ref={searchInputRef}
+            leftSearchIcon
+            variant="internal"
+            value={searchQuery}
+            onChange={(e) => setSearchQuery(e.target.value)}
+            placeholder="Search models..."
+          />
+
+          {/* Model List with Vendor Groups */}
+          <PopoverMenu scrollContainerRef={scrollContainerRef}>
+            {isLoadingProviders
+              ? [
+                  <div key="loading" className="flex items-center gap-2 py-3">
+                    <SimpleLoader />
+                    <Text secondaryBody text03>
+                      Loading models...
+                    </Text>
+                  </div>,
+                ]
+              : groupedOptions.length === 0
+                ? [
+                    <div key="empty" className="py-3">
+                      <Text secondaryBody text03>
+                        No models found
+                      </Text>
+                    </div>,
+                  ]
+                : groupedOptions.length === 1
+                  ? // Single provider - show models directly without accordion
+                    [
+                      <div
+                        key="single-provider"
+                        className="flex flex-col gap-1"
+                      >
+                        {groupedOptions[0]!.options.map(renderModelItem)}
+                      </div>,
+                    ]
+                  : // Multiple providers - show accordion with groups
+                    [
+                      <Accordion
+                        key="accordion"
+                        type="multiple"
+                        value={effectiveExpandedGroups}
+                        onValueChange={handleAccordionChange}
+                        className="w-full flex flex-col"
+                      >
+                        {groupedOptions.map((group) => {
+                          const isExpanded = effectiveExpandedGroups.includes(
+                            group.key
+                          );
+                          return (
+                            <AccordionItem
+                              key={group.key}
+                              value={group.key}
+                              className="border-none pt-1"
+                            >
+                              {/* Group Header */}
+                              <AccordionTrigger className="flex items-center rounded-08 hover:no-underline hover:bg-background-tint-02 group [&>svg]:hidden w-full py-1">
+                                <div className="flex items-center gap-1 shrink-0">
+                                  <div className="flex items-center justify-center size-5 shrink-0">
+                                    <group.Icon size={16} />
+                                  </div>
+                                  <Text
+                                    secondaryBody
+                                    text03
+                                    nowrap
+                                    className="px-0.5"
+                                  >
+                                    {group.displayName}
+                                  </Text>
+                                </div>
+                                <div className="flex-1" />
+                                <div className="flex items-center justify-center size-6 shrink-0">
+                                  {isExpanded ? (
+                                    <SvgChevronDown className="h-4 w-4 stroke-text-04 shrink-0" />
+                                  ) : (
+                                    <SvgChevronRight className="h-4 w-4 stroke-text-04 shrink-0" />
+                                  )}
+                                </div>
+                              </AccordionTrigger>
+
+                              {/* Model Items - full width highlight */}
+                              <AccordionContent className="pb-0 pt-0">
+                                <div className="flex flex-col gap-1">
+                                  {group.options.map(renderModelItem)}
+                                </div>
+                              </AccordionContent>
+                            </AccordionItem>
+                          );
+                        })}
+                      </Accordion>,
+                    ]}
+          </PopoverMenu>
+
+          {/* Global Temperature Slider (shown if enabled in user prefs) */}
+          {user?.preferences?.temperature_override_enabled && (
+            <>
+              <div className="border-t border-border-02 mx-2" />
+              <div className="flex flex-col w-full py-2 gap-2">
+                <Slider
+                  value={[localTemperature]}
+                  max={llmManager.maxTemperature}
+                  min={0}
+                  step={0.01}
+                  onValueChange={handleGlobalTemperatureChange}
+                  onValueCommit={handleGlobalTemperatureCommit}
+                  className="w-full"
+                />
+                <div className="flex flex-row items-center justify-between">
+                  <Text secondaryBody text03>
+                    Temperature (creativity)
+                  </Text>
+                  <Text secondaryBody text03>
+                    {localTemperature.toFixed(1)}
+                  </Text>
+                </div>
+              </div>
+            </>
+          )}
+        </Section>
       </Popover.Content>
     </Popover>
   );

web/src/refresh-components/popovers/ModelListContent.tsx

@@ -1,200 +0,0 @@
-"use client";
-
-import { useState, useMemo, useRef, useEffect } from "react";
-import { PopoverMenu } from "@/refresh-components/Popover";
-import InputTypeIn from "@/refresh-components/inputs/InputTypeIn";
-import { Text } from "@opal/components";
-import { SvgCheck, SvgChevronDown, SvgChevronRight } from "@opal/icons";
-import { Section } from "@/layouts/general-layouts";
-import { LLMOption } from "./interfaces";
-import { buildLlmOptions, groupLlmOptions } from "./LLMPopover";
-import LineItem from "@/refresh-components/buttons/LineItem";
-import { LLMProviderDescriptor } from "@/interfaces/llm";
-import {
-  Collapsible,
-  CollapsibleContent,
-  CollapsibleTrigger,
-} from "@/refresh-components/Collapsible";
-
-export interface ModelListContentProps {
-  llmProviders: LLMProviderDescriptor[] | undefined;
-  currentModelName?: string;
-  requiresImageInput?: boolean;
-  onSelect: (option: LLMOption) => void;
-  isSelected: (option: LLMOption) => boolean;
-  isDisabled?: (option: LLMOption) => boolean;
-  scrollContainerRef?: React.RefObject<HTMLDivElement | null>;
-  isLoading?: boolean;
-  footer?: React.ReactNode;
-}
-
-export default function ModelListContent({
-  llmProviders,
-  currentModelName,
-  requiresImageInput,
-  onSelect,
-  isSelected,
-  isDisabled,
-  scrollContainerRef: externalScrollRef,
-  isLoading,
-  footer,
-}: ModelListContentProps) {
-  const [searchQuery, setSearchQuery] = useState("");
-  const internalScrollRef = useRef<HTMLDivElement>(null);
-  const scrollContainerRef = externalScrollRef ?? internalScrollRef;
-
-  const llmOptions = useMemo(
-    () => buildLlmOptions(llmProviders, currentModelName),
-    [llmProviders, currentModelName]
-  );
-
-  const filteredOptions = useMemo(() => {
-    let result = llmOptions;
-    if (requiresImageInput) {
-      result = result.filter((opt) => opt.supportsImageInput);
-    }
-    if (searchQuery.trim()) {
-      const query = searchQuery.toLowerCase();
-      result = result.filter(
-        (opt) =>
-          opt.displayName.toLowerCase().includes(query) ||
-          opt.modelName.toLowerCase().includes(query) ||
-          (opt.vendor && opt.vendor.toLowerCase().includes(query))
-      );
-    }
-    return result;
-  }, [llmOptions, searchQuery, requiresImageInput]);
-
-  const groupedOptions = useMemo(
-    () => groupLlmOptions(filteredOptions),
-    [filteredOptions]
-  );
-
-  // Find which group contains a currently-selected model (for auto-expand)
-  const defaultGroupKey = useMemo(() => {
-    for (const group of groupedOptions) {
-      if (group.options.some((opt) => isSelected(opt))) {
-        return group.key;
-      }
-    }
-    return groupedOptions[0]?.key ?? "";
-  }, [groupedOptions, isSelected]);
-
-  const [expandedGroups, setExpandedGroups] = useState<Set<string>>(
-    new Set([defaultGroupKey])
-  );
-
-  // Reset expanded groups when default changes (e.g. popover re-opens)
-  useEffect(() => {
-    setExpandedGroups(new Set([defaultGroupKey]));
-  }, [defaultGroupKey]);
-
-  const isSearching = searchQuery.trim().length > 0;
-
-  const toggleGroup = (key: string) => {
-    if (isSearching) return;
-    setExpandedGroups((prev) => {
-      const next = new Set(prev);
-      if (next.has(key)) next.delete(key);
-      else next.add(key);
-      return next;
-    });
-  };
-
-  const isGroupOpen = (key: string) => isSearching || expandedGroups.has(key);
-
-  const renderModelItem = (option: LLMOption) => {
-    const selected = isSelected(option);
-    const disabled = isDisabled?.(option) ?? false;
-
-    const capabilities: string[] = [];
-    if (option.supportsReasoning) capabilities.push("Reasoning");
-    if (option.supportsImageInput) capabilities.push("Vision");
-    const description =
-      capabilities.length > 0 ? capabilities.join(", ") : undefined;
-
-    return (
-      <LineItem
-        key={`${option.provider}:${option.modelName}`}
-        selected={selected}
-        disabled={disabled}
-        description={description}
-        onClick={() => onSelect(option)}
-        rightChildren={
-          selected ? (
-            <SvgCheck className="h-4 w-4 stroke-action-link-05 shrink-0" />
-          ) : null
-        }
-      >
-        {option.displayName}
-      </LineItem>
-    );
-  };
-
-  return (
-    <Section gap={0.5}>
-      <InputTypeIn
-        leftSearchIcon
-        variant="internal"
-        value={searchQuery}
-        onChange={(e) => setSearchQuery(e.target.value)}
-        placeholder="Search models..."
-      />
-
-      <PopoverMenu scrollContainerRef={scrollContainerRef}>
-        {isLoading
-          ? [
-              <Text key="loading" font="secondary-body" color="text-03">
-                Loading models...
-              </Text>,
-            ]
-          : groupedOptions.length === 0
-            ? [
-                <Text key="empty" font="secondary-body" color="text-03">
-                  No models found
-                </Text>,
-              ]
-            : groupedOptions.length === 1
-              ? [
-                  <Section key="single-provider" gap={0.25}>
-                    {groupedOptions[0]!.options.map(renderModelItem)}
-                  </Section>,
-                ]
-              : groupedOptions.map((group) => {
-                  const open = isGroupOpen(group.key);
-                  return (
-                    <Collapsible
-                      key={group.key}
-                      open={open}
-                      onOpenChange={() => toggleGroup(group.key)}
-                    >
-                      <CollapsibleTrigger asChild>
-                        <LineItem
-                          muted
-                          icon={group.Icon}
-                          rightChildren={
-                            open ? (
-                              <SvgChevronDown className="h-4 w-4 stroke-text-04 shrink-0" />
-                            ) : (
-                              <SvgChevronRight className="h-4 w-4 stroke-text-04 shrink-0" />
-                            )
-                          }
-                        >
-                          {group.displayName}
-                        </LineItem>
-                      </CollapsibleTrigger>
-
-                      <CollapsibleContent>
-                        <Section gap={0.25}>
-                          {group.options.map(renderModelItem)}
-                        </Section>
-                      </CollapsibleContent>
-                    </Collapsible>
-                  );
-                })}
-      </PopoverMenu>
-
-      {footer}
-    </Section>
-  );
-}

web/src/refresh-components/popovers/ModelSelector.tsx

@@ -1,230 +0,0 @@
-"use client";
-
-import { useState, useMemo, useRef } from "react";
-import Popover from "@/refresh-components/Popover";
-import { LlmManager } from "@/lib/hooks";
-import { getProviderIcon } from "@/app/admin/configuration/llm/utils";
-import { Button, SelectButton, OpenButton } from "@opal/components";
-import { SvgPlusCircle, SvgX } from "@opal/icons";
-import { LLMOption } from "@/refresh-components/popovers/interfaces";
-import ModelListContent from "@/refresh-components/popovers/ModelListContent";
-import Separator from "@/refresh-components/Separator";
-
-export const MAX_MODELS = 3;
-
-export interface SelectedModel {
-  name: string;
-  provider: string;
-  modelName: string;
-  displayName: string;
-}
-
-export interface ModelSelectorProps {
-  llmManager: LlmManager;
-  selectedModels: SelectedModel[];
-  onAdd: (model: SelectedModel) => void;
-  onRemove: (index: number) => void;
-  onReplace: (index: number, model: SelectedModel) => void;
-}
-
-function modelKey(provider: string, modelName: string): string {
-  return `${provider}:${modelName}`;
-}
-
-export default function ModelSelector({
-  llmManager,
-  selectedModels,
-  onAdd,
-  onRemove,
-  onReplace,
-}: ModelSelectorProps) {
-  const [open, setOpen] = useState(false);
-  // null = add mode (via + button), number = replace mode (via pill click)
-  const [replacingIndex, setReplacingIndex] = useState<number | null>(null);
-  // Virtual anchor ref — points to the clicked pill so the popover positions above it
-  const anchorRef = useRef<HTMLElement | null>(null);
-
-  const isMultiModel = selectedModels.length > 1;
-  const atMax = selectedModels.length >= MAX_MODELS;
-
-  const selectedKeys = useMemo(
-    () => new Set(selectedModels.map((m) => modelKey(m.provider, m.modelName))),
-    [selectedModels]
-  );
-
-  const otherSelectedKeys = useMemo(() => {
-    if (replacingIndex === null) return new Set<string>();
-    return new Set(
-      selectedModels
-        .filter((_, i) => i !== replacingIndex)
-        .map((m) => modelKey(m.provider, m.modelName))
-    );
-  }, [selectedModels, replacingIndex]);
-
-  const replacingKey =
-    replacingIndex !== null
-      ? (() => {
-          const m = selectedModels[replacingIndex];
-          return m ? modelKey(m.provider, m.modelName) : null;
-        })()
-      : null;
-
-  const isSelected = (option: LLMOption) => {
-    const key = modelKey(option.provider, option.modelName);
-    if (replacingIndex !== null) return key === replacingKey;
-    return selectedKeys.has(key);
-  };
-
-  const isDisabled = (option: LLMOption) => {
-    const key = modelKey(option.provider, option.modelName);
-    if (replacingIndex !== null) return otherSelectedKeys.has(key);
-    return !selectedKeys.has(key) && atMax;
-  };
-
-  const handleSelect = (option: LLMOption) => {
-    const model: SelectedModel = {
-      name: option.name,
-      provider: option.provider,
-      modelName: option.modelName,
-      displayName: option.displayName,
-    };
-
-    if (replacingIndex !== null) {
-      onReplace(replacingIndex, model);
-      setOpen(false);
-      setReplacingIndex(null);
-      return;
-    }
-
-    const key = modelKey(option.provider, option.modelName);
-    const existingIndex = selectedModels.findIndex(
-      (m) => modelKey(m.provider, m.modelName) === key
-    );
-    if (existingIndex >= 0) {
-      onRemove(existingIndex);
-    } else if (!atMax) {
-      onAdd(model);
-    }
-  };
-
-  const handleOpenChange = (nextOpen: boolean) => {
-    setOpen(nextOpen);
-    if (!nextOpen) setReplacingIndex(null);
-  };
-
-  const handlePillClick = (index: number, element: HTMLElement) => {
-    anchorRef.current = element;
-    setReplacingIndex(index);
-    setOpen(true);
-  };
-
-  return (
-    <Popover open={open} onOpenChange={handleOpenChange}>
-      <div className="flex items-center justify-end gap-1 p-1">
-        {!atMax && (
-          <Button
-            prominence="tertiary"
-            icon={SvgPlusCircle}
-            size="sm"
-            tooltip="Add Model"
-            onClick={(e: React.MouseEvent) => {
-              anchorRef.current = e.currentTarget as HTMLElement;
-              setReplacingIndex(null);
-              setOpen(true);
-            }}
-          />
-        )}
-
-        <Popover.Anchor
-          virtualRef={anchorRef as React.RefObject<HTMLElement>}
-        />
-        {selectedModels.length > 0 && (
-          <>
-            {!atMax && (
-              <Separator
-                orientation="vertical"
-                paddingXRem={0.5}
-                paddingYRem={0.5}
-              />
-            )}
-            <div className="flex items-center">
-              {selectedModels.map((model, index) => {
-                const ProviderIcon = getProviderIcon(
-                  model.provider,
-                  model.modelName
-                );
-
-                if (!isMultiModel) {
-                  return (
-                    <OpenButton
-                      key={modelKey(model.provider, model.modelName)}
-                      icon={ProviderIcon}
-                      onClick={(e: React.MouseEvent) =>
-                        handlePillClick(index, e.currentTarget as HTMLElement)
-                      }
-                    >
-                      {model.displayName}
-                    </OpenButton>
-                  );
-                }
-
-                return (
-                  <div
-                    key={modelKey(model.provider, model.modelName)}
-                    className="flex items-center"
-                  >
-                    {index > 0 && (
-                      <Separator
-                        orientation="vertical"
-                        paddingXRem={0.5}
-                        className="h-5"
-                      />
-                    )}
-                    <SelectButton
-                      icon={ProviderIcon}
-                      rightIcon={SvgX}
-                      state="empty"
-                      variant="select-tinted"
-                      interaction="hover"
-                      size="lg"
-                      onClick={(e: React.MouseEvent) => {
-                        const target = e.target as HTMLElement;
-                        const btn = e.currentTarget as HTMLElement;
-                        const icons = btn.querySelectorAll(
-                          ".interactive-foreground-icon"
-                        );
-                        const lastIcon = icons[icons.length - 1];
-                        if (lastIcon && lastIcon.contains(target)) {
-                          onRemove(index);
-                        } else {
-                          handlePillClick(index, btn);
-                        }
-                      }}
-                    >
-                      {model.displayName}
-                    </SelectButton>
-                  </div>
-                );
-              })}
-            </div>
-          </>
-        )}
-      </div>
-
-      <Popover.Content
-        side="top"
-        align="start"
-        width="lg"
-        avoidCollisions={false}
-      >
-        <ModelListContent
-          llmProviders={llmManager.llmProviders}
-          isLoading={llmManager.isLoadingProviders}
-          onSelect={handleSelect}
-          isSelected={isSelected}
-          isDisabled={isDisabled}
-        />
-      </Popover.Content>
-    </Popover>
-  );
-}

web/src/refresh-pages/AppPage.tsx

@@ -232,6 +232,7 @@ export default function AppPage({ firstMessage }: ChatPageProps) {
     onboardingDismissed,
     onboardingState,
     onboardingActions,
+    llmDescriptors,
     isLoadingOnboarding,
     finishOnboarding,
     hideOnboarding,
@@ -811,6 +812,7 @@ export default function AppPage({ firstMessage }: ChatPageProps) {
                           handleFinishOnboarding={finishOnboarding}
                           state={onboardingState}
                           actions={onboardingActions}
+                          llmDescriptors={llmDescriptors}
                         />
                       )}
 

web/src/sections/onboarding/OnboardingFlow.tsx

@@ -10,6 +10,7 @@ import {
   OnboardingState,
   OnboardingStep,
 } from "@/interfaces/onboarding";
+import { WellKnownLLMProviderDescriptor } from "@/interfaces/llm";
 import { useUser } from "@/providers/UserProvider";
 import { UserRole } from "@/lib/types";
 import NonAdminStep from "./components/NonAdminStep";
@@ -20,6 +21,7 @@ type OnboardingFlowProps = {
   handleFinishOnboarding: () => void;
   state: OnboardingState;
   actions: OnboardingActions;
+  llmDescriptors: WellKnownLLMProviderDescriptor[];
 };
 
 const OnboardingFlowInner = ({
@@ -28,6 +30,7 @@ const OnboardingFlowInner = ({
   handleFinishOnboarding,
   state: onboardingState,
   actions: onboardingActions,
+  llmDescriptors,
 }: OnboardingFlowProps) => {
   const { user } = useUser();
 
@@ -54,6 +57,7 @@ const OnboardingFlowInner = ({
               <LLMStep
                 state={onboardingState}
                 actions={onboardingActions}
+                llmDescriptors={llmDescriptors}
                 disabled={
                   onboardingState.currentStep !== OnboardingStep.LlmSetup
                 }

web/src/sections/onboarding/steps/LLMStep.tsx

@@ -19,11 +19,11 @@ import { Disabled } from "@opal/core";
 import { ProviderIcon } from "@/app/admin/configuration/llm/ProviderIcon";
 import { SvgCheckCircle, SvgCpu, SvgExternalLink } from "@opal/icons";
 import { ContentAction } from "@opal/layouts";
-import { useLLMProviderOptions } from "@/lib/hooks/useLLMProviderOptions";
 
 type LLMStepProps = {
   state: OnboardingState;
   actions: OnboardingActions;
+  llmDescriptors: WellKnownLLMProviderDescriptor[];
   disabled?: boolean;
 };
 
@@ -92,10 +92,10 @@ const StackedProviderIcons = ({ providers }: StackedProviderIconsProps) => {
 const LLMStepInner = ({
   state: onboardingState,
   actions: onboardingActions,
+  llmDescriptors,
   disabled,
 }: LLMStepProps) => {
-  const { llmProviderOptions, isLoading } = useLLMProviderOptions();
-  const llmDescriptors = llmProviderOptions ?? [];
+  const isLoading = !llmDescriptors || llmDescriptors.length === 0;
 
   const [selectedProvider, setSelectedProvider] =
     useState<SelectedProvider | null>(null);

web/tests/e2e/chat/llm_runtime_selection.spec.ts

@@ -411,8 +411,7 @@ test.describe("LLM Runtime Selection", () => {
     const sharedModelOptions = dialog.locator("[data-selected]");
     await expect(sharedModelOptions).toHaveCount(2);
     const openAiModelOption = dialog
-      .getByRole("button", { name: /openai/i })
-      .locator("..")
+      .getByRole("region", { name: /openai/i })
       .locator("[data-selected]")
       .first();
     await expect(openAiModelOption).toBeVisible();
@@ -437,8 +436,7 @@ test.describe("LLM Runtime Selection", () => {
     const secondSharedModelOptions = secondDialog.locator("[data-selected]");
     await expect(secondSharedModelOptions).toHaveCount(2);
     const anthropicModelOption = secondDialog
-      .getByRole("button", { name: /anthropic/i })
-      .locator("..")
+      .getByRole("region", { name: /anthropic/i })
       .locator("[data-selected]")
       .first();
     await expect(anthropicModelOption).toBeVisible();
@@ -449,8 +447,7 @@ test.describe("LLM Runtime Selection", () => {
     await page.waitForSelector('[role="dialog"]', { state: "visible" });
     const verifyDialog = page.locator('[role="dialog"]');
     const selectedAnthropicOption = verifyDialog
-      .getByRole("button", { name: /anthropic/i })
-      .locator("..")
+      .getByRole("region", { name: /anthropic/i })
       .locator('[data-selected="true"]');
     await expect(selectedAnthropicOption).toHaveCount(1);
     await page.keyboard.press("Escape");