k

.github/workflows/docker-build-backend-container-on-merge-group.yml

@@ -1,8 +1,6 @@
 name: Build Backend Image on Merge Group
 
 on:
-  pull_request:
-    branches: [ "main" ]
   merge_group:
     types: [checks_requested]
 

.github/workflows/docker-build-web-container-on-merge-group.yml

@@ -1,8 +1,6 @@
 name: Build Web Image on Merge Group
 
 on:
-  pull_request:
-    branches: [ "main" ]
   merge_group:
     types: [checks_requested]
 

backend/alembic/versions/05c07bf07c00_add_search_doc_relevance_details.py

@@ -0,0 +1,32 @@
+"""add search doc relevance details
+
+Revision ID: 05c07bf07c00
+Revises: b896bbd0d5a7
+Create Date: 2024-07-10 17:48:15.886653
+
+"""
+
+from alembic import op
+import sqlalchemy as sa
+
+# revision identifiers, used by Alembic.
+revision = "05c07bf07c00"
+down_revision = "b896bbd0d5a7"
+branch_labels: None = None
+depends_on: None = None
+
+
+def upgrade() -> None:
+    op.add_column(
+        "search_doc",
+        sa.Column("is_relevant", sa.Boolean(), nullable=True),
+    )
+    op.add_column(
+        "search_doc",
+        sa.Column("relevance_explanation", sa.String(), nullable=True),
+    )
+
+
+def downgrade() -> None:
+    op.drop_column("search_doc", "relevance_explanation")
+    op.drop_column("search_doc", "is_relevant")

backend/alembic/versions/b896bbd0d5a7_backfill_is_internet_data_to_false.py

@@ -0,0 +1,23 @@
+"""backfill is_internet data to False
+
+Revision ID: b896bbd0d5a7
+Revises: 44f856ae2a4a
+Create Date: 2024-07-16 15:21:05.718571
+
+"""
+from alembic import op
+
+
+# revision identifiers, used by Alembic.
+revision = "b896bbd0d5a7"
+down_revision = "44f856ae2a4a"
+branch_labels: None = None
+depends_on: None = None
+
+
+def upgrade() -> None:
+    op.execute("UPDATE search_doc SET is_internet = FALSE WHERE is_internet IS NULL")
+
+
+def downgrade() -> None:
+    pass

backend/danswer/chat/models.py

@@ -42,11 +42,21 @@ class QADocsResponse(RetrievalDocs):
         return initial_dict
 
 
-# Second chunk of info for streaming QA
 class LLMRelevanceFilterResponse(BaseModel):
     relevant_chunk_indices: list[int]
 
 
+class RelevanceChunk(BaseModel):
+    # TODO make this document level. Also slight misnomer here as this is actually
+    # done at the section level currently rather than the chunk
+    relevant: bool | None = None
+    content: str | None = None
+
+
+class LLMRelevanceSummaryResponse(BaseModel):
+    relevance_summaries: dict[str, RelevanceChunk]
+
+
 class DanswerAnswerPiece(BaseModel):
     # A small piece of a complete answer. Used for streaming back answers.
     answer_piece: str | None  # if None, specifies the end of an Answer

backend/danswer/configs/app_configs.py

@@ -223,6 +223,11 @@ MAX_PRUNING_DOCUMENT_RETRIEVAL_PER_MINUTE = int(
     os.environ.get("MAX_PRUNING_DOCUMENT_RETRIEVAL_PER_MINUTE", 0)
 )
 
+# comma delimited list of zendesk article labels to skip indexing for
+ZENDESK_CONNECTOR_SKIP_ARTICLE_LABELS = os.environ.get(
+    "ZENDESK_CONNECTOR_SKIP_ARTICLE_LABELS", ""
+).split(",")
+
 
 #####
 # Indexing Configs

backend/danswer/configs/chat_configs.py

@@ -75,6 +75,17 @@ LANGUAGE_CHAT_NAMING_HINT = (
     or "The name of the conversation must be in the same language as the user query."
 )
 
+
+# Agentic search takes significantly more tokens and therefore has much higher cost.
+# This configuration allows users to get a search-only experience with instant results
+# and no involvement from the LLM.
+# Additionally, some LLM providers have strict rate limits which may prohibit
+# sending many API requests at once (as is done in agentic search).
+DISABLE_AGENTIC_SEARCH = (
+    os.environ.get("DISABLE_AGENTIC_SEARCH") or "false"
+).lower() == "true"
+
+
 # Stops streaming answers back to the UI if this pattern is seen:
 STOP_STREAM_PAT = os.environ.get("STOP_STREAM_PAT") or None
 

backend/danswer/connectors/confluence/rate_limit_handler.py

@@ -38,7 +38,7 @@ def make_confluence_call_handle_rate_limit(confluence_call: F) -> F:
                     retry_after = None
                     try:
                         retry_after = int(e.response.headers.get("Retry-After"))
-                    except ValueError:
+                    except (ValueError, TypeError):
                         pass
 
                     if retry_after:

backend/danswer/connectors/discourse/connector.py

@@ -11,6 +11,9 @@ from requests import Response
 from danswer.configs.app_configs import INDEX_BATCH_SIZE
 from danswer.configs.constants import DocumentSource
 from danswer.connectors.cross_connector_utils.miscellaneous_utils import time_str_to_utc
+from danswer.connectors.cross_connector_utils.rate_limit_wrapper import (
+    rate_limit_builder,
+)
 from danswer.connectors.cross_connector_utils.retry_wrapper import retry_builder
 from danswer.connectors.interfaces import GenerateDocumentsOutput
 from danswer.connectors.interfaces import PollConnector
@@ -58,67 +61,36 @@ class DiscourseConnector(PollConnector):
         self.category_id_map: dict[int, str] = {}
 
         self.batch_size = batch_size
-
         self.permissions: DiscoursePerms | None = None
+        self.active_categories: set | None = None
+
+    @rate_limit_builder(max_calls=100, period=60)
+    def _make_request(self, endpoint: str, params: dict | None = None) -> Response:
+        if not self.permissions:
+            raise ConnectorMissingCredentialError("Discourse")
+        return discourse_request(endpoint, self.permissions, params)
 
     def _get_categories_map(
         self,
     ) -> None:
         assert self.permissions is not None
         categories_endpoint = urllib.parse.urljoin(self.base_url, "categories.json")
-        response = discourse_request(
+        response = self._make_request(
             endpoint=categories_endpoint,
-            perms=self.permissions,
             params={"include_subcategories": True},
         )
         categories = response.json()["category_list"]["categories"]
-
         self.category_id_map = {
-            category["id"]: category["name"]
-            for category in categories
-            if not self.categories or category["name"].lower() in self.categories
+            cat["id"]: cat["name"]
+            for cat in categories
+            if not self.categories or cat["name"].lower() in self.categories
         }
-
-    def _get_latest_topics(
-        self, start: datetime | None, end: datetime | None
-    ) -> list[int]:
-        assert self.permissions is not None
-        topic_ids = []
-
-        valid_categories = set(self.category_id_map.keys())
-
-        latest_endpoint = urllib.parse.urljoin(self.base_url, "latest.json")
-        response = discourse_request(endpoint=latest_endpoint, perms=self.permissions)
-        topics = response.json()["topic_list"]["topics"]
-        for topic in topics:
-            last_time = topic.get("last_posted_at")
-            if not last_time:
-                continue
-            last_time_dt = time_str_to_utc(last_time)
-
-            if start and start > last_time_dt:
-                continue
-            if end and end < last_time_dt:
-                continue
-
-            if (
-                self.categories
-                and valid_categories
-                and topic.get("category_id") not in valid_categories
-            ):
-                continue
-
-            topic_ids.append(topic["id"])
-
-        return topic_ids
+        self.active_categories = set(self.category_id_map)
 
     def _get_doc_from_topic(self, topic_id: int) -> Document:
         assert self.permissions is not None
         topic_endpoint = urllib.parse.urljoin(self.base_url, f"t/{topic_id}.json")
-        response = discourse_request(
-            endpoint=topic_endpoint,
-            perms=self.permissions,
-        )
+        response = self._make_request(endpoint=topic_endpoint)
         topic = response.json()
 
         topic_url = urllib.parse.urljoin(self.base_url, f"t/{topic['slug']}")
@@ -167,26 +139,78 @@ class DiscourseConnector(PollConnector):
         )
         return doc
 
+    def _get_latest_topics(
+        self, start: datetime | None, end: datetime | None, page: int
+    ) -> list[int]:
+        assert self.permissions is not None
+        topic_ids = []
+
+        if not self.categories:
+            latest_endpoint = urllib.parse.urljoin(
+                self.base_url, f"latest.json?page={page}"
+            )
+            response = self._make_request(endpoint=latest_endpoint)
+            topics = response.json()["topic_list"]["topics"]
+
+        else:
+            topics = []
+            empty_categories = []
+
+            for category_id in self.category_id_map.keys():
+                category_endpoint = urllib.parse.urljoin(
+                    self.base_url, f"c/{category_id}.json?page={page}&sys=latest"
+                )
+                response = self._make_request(endpoint=category_endpoint)
+                new_topics = response.json()["topic_list"]["topics"]
+
+                if len(new_topics) == 0:
+                    empty_categories.append(category_id)
+                topics.extend(new_topics)
+
+            for empty_category in empty_categories:
+                self.category_id_map.pop(empty_category)
+
+        for topic in topics:
+            last_time = topic.get("last_posted_at")
+            if not last_time:
+                continue
+
+            last_time_dt = time_str_to_utc(last_time)
+            if (start and start > last_time_dt) or (end and end < last_time_dt):
+                continue
+
+            topic_ids.append(topic["id"])
+            if len(topic_ids) >= self.batch_size:
+                break
+
+        return topic_ids
+
     def _yield_discourse_documents(
-        self, topic_ids: list[int]
+        self,
+        start: datetime,
+        end: datetime,
     ) -> GenerateDocumentsOutput:
-        doc_batch: list[Document] = []
-        for topic_id in topic_ids:
-            doc_batch.append(self._get_doc_from_topic(topic_id))
+        page = 1
+        while topic_ids := self._get_latest_topics(start, end, page):
+            doc_batch: list[Document] = []
+            for topic_id in topic_ids:
+                doc_batch.append(self._get_doc_from_topic(topic_id))
+                if len(doc_batch) >= self.batch_size:
+                    yield doc_batch
+                    doc_batch = []
 
-            if len(doc_batch) >= self.batch_size:
+            if doc_batch:
                 yield doc_batch
-                doc_batch = []
+            page += 1
 
-        if doc_batch:
-            yield doc_batch
-
-    def load_credentials(self, credentials: dict[str, Any]) -> dict[str, Any] | None:
+    def load_credentials(
+        self,
+        credentials: dict[str, Any],
+    ) -> dict[str, Any] | None:
         self.permissions = DiscoursePerms(
             api_key=credentials["discourse_api_key"],
             api_username=credentials["discourse_api_username"],
         )
-
         return None
 
     def poll_source(
@@ -194,16 +218,13 @@ class DiscourseConnector(PollConnector):
     ) -> GenerateDocumentsOutput:
         if self.permissions is None:
             raise ConnectorMissingCredentialError("Discourse")
+
         start_datetime = datetime.utcfromtimestamp(start).replace(tzinfo=timezone.utc)
         end_datetime = datetime.utcfromtimestamp(end).replace(tzinfo=timezone.utc)
 
         self._get_categories_map()
 
-        latest_topic_ids = self._get_latest_topics(
-            start=start_datetime, end=end_datetime
-        )
-
-        yield from self._yield_discourse_documents(latest_topic_ids)
+        yield from self._yield_discourse_documents(start_datetime, end_datetime)
 
 
 if __name__ == "__main__":
@@ -219,7 +240,5 @@ if __name__ == "__main__":
 
     current = time.time()
     one_year_ago = current - 24 * 60 * 60 * 360
-
     latest_docs = connector.poll_source(one_year_ago, current)
-
     print(next(latest_docs))

backend/danswer/connectors/file/connector.py

@@ -88,7 +88,7 @@ def _process_file(
     # add a prefix to avoid conflicts with other connectors
     doc_id = f"FILE_CONNECTOR__{file_name}"
     if metadata:
-        doc_id = metadata.get("id") or doc_id
+        doc_id = metadata.get("document_id") or doc_id
 
     # If this is set, we will show this in the UI as the "name" of the file
     file_display_name = all_metadata.get("file_display_name") or os.path.basename(
@@ -111,6 +111,7 @@ def _process_file(
         for k, v in all_metadata.items()
         if k
         not in [
+            "document_id",
             "time_updated",
             "doc_updated_at",
             "link",

backend/danswer/connectors/zendesk/connector.py

@@ -4,6 +4,7 @@ from zenpy import Zenpy  # type: ignore
 from zenpy.lib.api_objects.help_centre_objects import Article  # type: ignore
 
 from danswer.configs.app_configs import INDEX_BATCH_SIZE
+from danswer.configs.app_configs import ZENDESK_CONNECTOR_SKIP_ARTICLE_LABELS
 from danswer.configs.constants import DocumentSource
 from danswer.connectors.cross_connector_utils.miscellaneous_utils import (
     time_str_to_utc,
@@ -81,7 +82,14 @@ class ZendeskConnector(LoadConnector, PollConnector):
         )
         doc_batch = []
         for article in articles:
-            if article.body is None or article.draft:
+            if (
+                article.body is None
+                or article.draft
+                or any(
+                    label in ZENDESK_CONNECTOR_SKIP_ARTICLE_LABELS
+                    for label in article.label_names
+                )
+            ):
                 continue
 
             doc_batch.append(_article_to_document(article))

backend/danswer/db/chat.py

@@ -3,15 +3,20 @@ from datetime import datetime
 from datetime import timedelta
 from uuid import UUID
 
+from sqlalchemy import and_
 from sqlalchemy import delete
+from sqlalchemy import desc
+from sqlalchemy import func
 from sqlalchemy import nullsfirst
 from sqlalchemy import or_
 from sqlalchemy import select
+from sqlalchemy import update
 from sqlalchemy.exc import MultipleResultsFound
 from sqlalchemy.orm import joinedload
 from sqlalchemy.orm import Session
 
 from danswer.auth.schemas import UserRole
+from danswer.chat.models import LLMRelevanceSummaryResponse
 from danswer.configs.chat_configs import HARD_DELETE_CHATS
 from danswer.configs.constants import MessageType
 from danswer.db.models import ChatMessage
@@ -34,6 +39,7 @@ from danswer.server.query_and_chat.models import ChatMessageDetail
 from danswer.tools.tool_runner import ToolCallFinalResult
 from danswer.utils.logger import setup_logger
 
+
 logger = setup_logger()
 
 
@@ -81,17 +87,46 @@ def get_chat_sessions_by_slack_thread_id(
     return db_session.scalars(stmt).all()
 
 
+def get_first_messages_for_chat_sessions(
+    chat_session_ids: list[int], db_session: Session
+) -> dict[int, str]:
+    subquery = (
+        select(ChatMessage.chat_session_id, func.min(ChatMessage.id).label("min_id"))
+        .where(
+            and_(
+                ChatMessage.chat_session_id.in_(chat_session_ids),
+                ChatMessage.message_type == MessageType.USER,  # Select USER messages
+            )
+        )
+        .group_by(ChatMessage.chat_session_id)
+        .subquery()
+    )
+
+    query = select(ChatMessage.chat_session_id, ChatMessage.message).join(
+        subquery,
+        (ChatMessage.chat_session_id == subquery.c.chat_session_id)
+        & (ChatMessage.id == subquery.c.min_id),
+    )
+
+    first_messages = db_session.execute(query).all()
+    return dict([(row.chat_session_id, row.message) for row in first_messages])
+
+
 def get_chat_sessions_by_user(
     user_id: UUID | None,
     deleted: bool | None,
     db_session: Session,
-    include_one_shot: bool = False,
+    only_one_shot: bool = False,
 ) -> list[ChatSession]:
     stmt = select(ChatSession).where(ChatSession.user_id == user_id)
 
-    if not include_one_shot:
+    if only_one_shot:
+        stmt = stmt.where(ChatSession.one_shot.is_(True))
+    else:
         stmt = stmt.where(ChatSession.one_shot.is_(False))
 
+    stmt = stmt.order_by(desc(ChatSession.time_created))
+
     if deleted is not None:
         stmt = stmt.where(ChatSession.deleted == deleted)
 
@@ -111,6 +146,12 @@ def delete_search_doc_message_relationship(
     db_session.commit()
 
 
+def delete_tool_call_for_message_id(message_id: int, db_session: Session) -> None:
+    stmt = delete(ToolCall).where(ToolCall.message_id == message_id)
+    db_session.execute(stmt)
+    db_session.commit()
+
+
 def delete_orphaned_search_docs(db_session: Session) -> None:
     orphaned_docs = (
         db_session.query(SearchDoc)
@@ -134,6 +175,7 @@ def delete_messages_and_files_from_chat_session(
     ).fetchall()
 
     for id, files in messages_with_files:
+        delete_tool_call_for_message_id(message_id=id, db_session=db_session)
         delete_search_doc_message_relationship(message_id=id, db_session=db_session)
         for file_info in files or {}:
             lobj_name = file_info.get("id")
@@ -275,6 +317,20 @@ def get_chat_messages_by_sessions(
     return db_session.execute(stmt).scalars().all()
 
 
+def get_search_docs_for_chat_message(
+    chat_message_id: int, db_session: Session
+) -> list[SearchDoc]:
+    stmt = (
+        select(SearchDoc)
+        .join(
+            ChatMessage__SearchDoc, ChatMessage__SearchDoc.search_doc_id == SearchDoc.id
+        )
+        .where(ChatMessage__SearchDoc.chat_message_id == chat_message_id)
+    )
+
+    return list(db_session.scalars(stmt).all())
+
+
 def get_chat_messages_by_session(
     chat_session_id: int,
     user_id: UUID | None,
@@ -295,8 +351,6 @@ def get_chat_messages_by_session(
 
     if prefetch_tool_calls:
         stmt = stmt.options(joinedload(ChatMessage.tool_calls))
-
-    if prefetch_tool_calls:
         result = db_session.scalars(stmt).unique().all()
     else:
         result = db_session.scalars(stmt).all()
@@ -484,6 +538,27 @@ def get_doc_query_identifiers_from_model(
     return doc_query_identifiers
 
 
+def update_search_docs_table_with_relevance(
+    db_session: Session,
+    reference_db_search_docs: list[SearchDoc],
+    relevance_summary: LLMRelevanceSummaryResponse,
+) -> None:
+    for search_doc in reference_db_search_docs:
+        relevance_data = relevance_summary.relevance_summaries.get(
+            f"{search_doc.document_id}-{search_doc.chunk_ind}"
+        )
+        if relevance_data is not None:
+            db_session.execute(
+                update(SearchDoc)
+                .where(SearchDoc.id == search_doc.id)
+                .values(
+                    is_relevant=relevance_data.relevant,
+                    relevance_explanation=relevance_data.content,
+                )
+            )
+    db_session.commit()
+
+
 def create_db_search_doc(
     server_search_doc: ServerSearchDoc,
     db_session: Session,
@@ -498,6 +573,8 @@ def create_db_search_doc(
         boost=server_search_doc.boost,
         hidden=server_search_doc.hidden,
         doc_metadata=server_search_doc.metadata,
+        is_relevant=server_search_doc.is_relevant,
+        relevance_explanation=server_search_doc.relevance_explanation,
         # For docs further down that aren't reranked, we can't use the retrieval score
         score=server_search_doc.score or 0.0,
         match_highlights=server_search_doc.match_highlights,
@@ -509,7 +586,6 @@ def create_db_search_doc(
 
     db_session.add(db_search_doc)
     db_session.commit()
-
     return db_search_doc
 
 
@@ -538,6 +614,8 @@ def translate_db_search_doc_to_server_search_doc(
         match_highlights=(
             db_search_doc.match_highlights if not remove_doc_content else []
         ),
+        relevance_explanation=db_search_doc.relevance_explanation,
+        is_relevant=db_search_doc.is_relevant,
         updated_at=db_search_doc.updated_at if not remove_doc_content else None,
         primary_owners=db_search_doc.primary_owners if not remove_doc_content else [],
         secondary_owners=(
@@ -561,9 +639,11 @@ def get_retrieval_docs_from_chat_message(
 
 
 def translate_db_message_to_chat_message_detail(
-    chat_message: ChatMessage, remove_doc_content: bool = False
+    chat_message: ChatMessage,
+    remove_doc_content: bool = False,
 ) -> ChatMessageDetail:
     chat_msg_detail = ChatMessageDetail(
+        chat_session_id=chat_message.chat_session_id,
         message_id=chat_message.id,
         parent_message=chat_message.parent_message,
         latest_child_message=chat_message.latest_child_message,

backend/danswer/db/models.py

@@ -671,6 +671,9 @@ class SearchDoc(Base):
     )
     is_internet: Mapped[bool] = mapped_column(Boolean, default=False, nullable=True)
 
+    is_relevant: Mapped[bool | None] = mapped_column(Boolean, nullable=True)
+    relevance_explanation: Mapped[str | None] = mapped_column(String, nullable=True)
+
     chat_messages = relationship(
         "ChatMessage",
         secondary="chat_message__search_doc",

backend/danswer/document_index/vespa/app_config/schemas/danswer_chunk.sd

@@ -153,43 +153,41 @@ schema DANSWER_CHUNK_NAME {
             query(query_embedding) tensor<float>(x[VARIABLE_DIM])
         }
 
-        # This must be separate function for normalize_linear to work
-        function vector_score() {
+        function title_vector_score() {
             expression {
                 # If no title, the full vector score comes from the content embedding
-                (query(title_content_ratio) * if(attribute(skip_title), closeness(field, embeddings), closeness(field, title_embedding))) +
-                ((1 - query(title_content_ratio)) * closeness(field, embeddings))
-            }
-        }
-
-        # This must be separate function for normalize_linear to work
-        function keyword_score() {
-            expression {
-                (query(title_content_ratio) * bm25(title)) +
-                ((1 - query(title_content_ratio)) * bm25(content))
+                #query(title_content_ratio) * if(attribute(skip_title), closeness(field, embeddings), closeness(field, title_embedding))
+                if(attribute(skip_title), closeness(field, embeddings), closeness(field, title_embedding))
             }
         }
 
         first-phase {
-            expression: vector_score
+            expression: closeness(field, embeddings)
         }
 
         # Weighted average between Vector Search and BM-25
-        # Each is a weighted average between the Title and Content fields
-        # Finally each doc is boosted by it's user feedback based boost and recency
-        # If any embedding or index field is missing, it just receives a score of 0
-        # Assumptions:
-        # - For a given query + corpus, the BM-25 scores will be relatively similar in distribution
-        #   therefore not normalizing before combining.
-        # - For documents without title, it gets a score of 0 for that and this is ok as documents
-        #   without any title match should be penalized.
         global-phase {
             expression {
                 (
                     # Weighted Vector Similarity Score
-                    (query(alpha) * normalize_linear(vector_score)) +
+                    (
+                        query(alpha) * (
+                            (query(title_content_ratio) * normalize_linear(title_vector_score))
+                            +
+                            ((1 - query(title_content_ratio)) * normalize_linear(closeness(field, embeddings)))
+                        )
+                    )
+
+                    +
+
                     # Weighted Keyword Similarity Score
-                    ((1 - query(alpha)) * normalize_linear(keyword_score))
+                    (
+                        (1 - query(alpha)) * (
+                            (query(title_content_ratio) * normalize_linear(bm25(title)))
+                            +
+                            ((1 - query(title_content_ratio)) * normalize_linear(bm25(content)))
+                        )
+                    )
                 )
                 # Boost based on user feedback
                 * document_boost
@@ -204,8 +202,6 @@ schema DANSWER_CHUNK_NAME {
             bm25(content)
             closeness(field, title_embedding)
             closeness(field, embeddings)
-            keyword_score
-            vector_score
             document_boost
             recency_bias
             closest(embeddings)

backend/danswer/llm/answering/answer.py

@@ -89,6 +89,9 @@ def _get_answer_stream_processor(
 AnswerStream = Iterator[AnswerQuestionPossibleReturn | ToolCallKickoff | ToolResponse]
 
 
+logger = setup_logger()
+
+
 class Answer:
     def __init__(
         self,
@@ -112,6 +115,7 @@ class Answer:
         skip_explicit_tool_calling: bool = False,
         # Returns the full document sections text from the search tool
         return_contexts: bool = False,
+        skip_gen_ai_answer_generation: bool = False,
     ) -> None:
         if single_message_history and message_history:
             raise ValueError(
@@ -140,11 +144,12 @@ class Answer:
         self._final_prompt: list[BaseMessage] | None = None
 
         self._streamed_output: list[str] | None = None
-        self._processed_stream: list[
-            AnswerQuestionPossibleReturn | ToolResponse | ToolCallKickoff
-        ] | None = None
+        self._processed_stream: (
+            list[AnswerQuestionPossibleReturn | ToolResponse | ToolCallKickoff] | None
+        ) = None
 
         self._return_contexts = return_contexts
+        self.skip_gen_ai_answer_generation = skip_gen_ai_answer_generation
 
     def _update_prompt_builder_for_search_tool(
         self, prompt_builder: AnswerPromptBuilder, final_context_documents: list[LlmDoc]
@@ -403,8 +408,9 @@ class Answer:
                     )
                 )
             )
+        final = tool_runner.tool_final_result()
 
-        yield tool_runner.tool_final_result()
+        yield final
 
         prompt = prompt_builder.build()
         yield from message_generator_to_string_generator(self.llm.stream(prompt=prompt))
@@ -467,22 +473,23 @@ class Answer:
                     # assumes all tool responses will come first, then the final answer
                     break
 
-            process_answer_stream_fn = _get_answer_stream_processor(
-                context_docs=final_context_docs or [],
-                # if doc selection is enabled, then search_results will be None,
-                # so we need to use the final_context_docs
-                doc_id_to_rank_map=map_document_id_order(
-                    search_results or final_context_docs or []
-                ),
-                answer_style_configs=self.answer_style_config,
-            )
+            if not self.skip_gen_ai_answer_generation:
+                process_answer_stream_fn = _get_answer_stream_processor(
+                    context_docs=final_context_docs or [],
+                    # if doc selection is enabled, then search_results will be None,
+                    # so we need to use the final_context_docs
+                    doc_id_to_rank_map=map_document_id_order(
+                        search_results or final_context_docs or []
+                    ),
+                    answer_style_configs=self.answer_style_config,
+                )
 
-            def _stream() -> Iterator[str]:
-                if message:
-                    yield cast(str, message)
-                yield from cast(Iterator[str], stream)
+                def _stream() -> Iterator[str]:
+                    if message:
+                        yield cast(str, message)
+                    yield from cast(Iterator[str], stream)
 
-            yield from process_answer_stream_fn(_stream())
+                yield from process_answer_stream_fn(_stream())
 
         processed_stream = []
         for processed_packet in _process_stream(output_generator):

backend/danswer/llm/answering/prune_and_merge.py

@@ -265,6 +265,7 @@ def prune_sections(
         max_tokens=document_pruning_config.max_tokens,
         tool_token_count=document_pruning_config.tool_num_tokens,
     )
+
     return _apply_pruning(
         sections=sections,
         section_relevance_list=section_relevance_list,

backend/danswer/llm/chat_llm.py

@@ -232,32 +232,6 @@ class DefaultMultiLLM(LLM):
 
         self._model_kwargs = model_kwargs
 
-    @staticmethod
-    def _log_prompt(prompt: LanguageModelInput) -> None:
-        if isinstance(prompt, list):
-            for ind, msg in enumerate(prompt):
-                if isinstance(msg, AIMessageChunk):
-                    if msg.content:
-                        log_msg = msg.content
-                    elif msg.tool_call_chunks:
-                        log_msg = "Tool Calls: " + str(
-                            [
-                                {
-                                    key: value
-                                    for key, value in tool_call.items()
-                                    if key != "index"
-                                }
-                                for tool_call in msg.tool_call_chunks
-                            ]
-                        )
-                    else:
-                        log_msg = ""
-                    logger.debug(f"Message {ind}:\n{log_msg}")
-                else:
-                    logger.debug(f"Message {ind}:\n{msg.content}")
-        if isinstance(prompt, str):
-            logger.debug(f"Prompt:\n{prompt}")
-
     def log_model_configs(self) -> None:
         logger.info(f"Config: {self.config}")
 
@@ -311,7 +285,7 @@ class DefaultMultiLLM(LLM):
             api_version=self._api_version,
         )
 
-    def invoke(
+    def _invoke_implementation(
         self,
         prompt: LanguageModelInput,
         tools: list[dict] | None = None,
@@ -319,7 +293,6 @@ class DefaultMultiLLM(LLM):
     ) -> BaseMessage:
         if LOG_DANSWER_MODEL_INTERACTIONS:
             self.log_model_configs()
-            self._log_prompt(prompt)
 
         response = cast(
             litellm.ModelResponse, self._completion(prompt, tools, tool_choice, False)
@@ -328,7 +301,7 @@ class DefaultMultiLLM(LLM):
             response.choices[0].message
         )
 
-    def stream(
+    def _stream_implementation(
         self,
         prompt: LanguageModelInput,
         tools: list[dict] | None = None,
@@ -336,7 +309,6 @@ class DefaultMultiLLM(LLM):
     ) -> Iterator[BaseMessage]:
         if LOG_DANSWER_MODEL_INTERACTIONS:
             self.log_model_configs()
-            self._log_prompt(prompt)
 
         if DISABLE_LITELLM_STREAMING:
             yield self.invoke(prompt)

backend/danswer/llm/custom_llm.py

@@ -76,7 +76,7 @@ class CustomModelServer(LLM):
     def log_model_configs(self) -> None:
         logger.debug(f"Custom model at: {self._endpoint}")
 
-    def invoke(
+    def _invoke_implementation(
         self,
         prompt: LanguageModelInput,
         tools: list[dict] | None = None,
@@ -84,7 +84,7 @@ class CustomModelServer(LLM):
     ) -> BaseMessage:
         return self._execute(prompt)
 
-    def stream(
+    def _stream_implementation(
         self,
         prompt: LanguageModelInput,
         tools: list[dict] | None = None,

backend/danswer/llm/interfaces.py

@@ -3,9 +3,12 @@ from collections.abc import Iterator
 from typing import Literal
 
 from langchain.schema.language_model import LanguageModelInput
+from langchain_core.messages import AIMessageChunk
 from langchain_core.messages import BaseMessage
 from pydantic import BaseModel
 
+from danswer.configs.app_configs import DISABLE_GENERATIVE_AI
+from danswer.configs.app_configs import LOG_DANSWER_MODEL_INTERACTIONS
 from danswer.utils.logger import setup_logger
 
 
@@ -23,6 +26,32 @@ class LLMConfig(BaseModel):
     api_version: str | None
 
 
+def log_prompt(prompt: LanguageModelInput) -> None:
+    if isinstance(prompt, list):
+        for ind, msg in enumerate(prompt):
+            if isinstance(msg, AIMessageChunk):
+                if msg.content:
+                    log_msg = msg.content
+                elif msg.tool_call_chunks:
+                    log_msg = "Tool Calls: " + str(
+                        [
+                            {
+                                key: value
+                                for key, value in tool_call.items()
+                                if key != "index"
+                            }
+                            for tool_call in msg.tool_call_chunks
+                        ]
+                    )
+                else:
+                    log_msg = ""
+                logger.debug(f"Message {ind}:\n{log_msg}")
+            else:
+                logger.debug(f"Message {ind}:\n{msg.content}")
+    if isinstance(prompt, str):
+        logger.debug(f"Prompt:\n{prompt}")
+
+
 class LLM(abc.ABC):
     """Mimics the LangChain LLM / BaseChatModel interfaces to make it easy
     to use these implementations to connect to a variety of LLM providers."""
@@ -45,20 +74,48 @@ class LLM(abc.ABC):
     def log_model_configs(self) -> None:
         raise NotImplementedError
 
-    @abc.abstractmethod
+    def _precall(self, prompt: LanguageModelInput) -> None:
+        if DISABLE_GENERATIVE_AI:
+            raise Exception("Generative AI is disabled")
+        if LOG_DANSWER_MODEL_INTERACTIONS:
+            log_prompt(prompt)
+
     def invoke(
         self,
         prompt: LanguageModelInput,
         tools: list[dict] | None = None,
         tool_choice: ToolChoiceOptions | None = None,
+    ) -> BaseMessage:
+        self._precall(prompt)
+        # TODO add a postcall to log model outputs independent of concrete class
+        # implementation
+        return self._invoke_implementation(prompt, tools, tool_choice)
+
+    @abc.abstractmethod
+    def _invoke_implementation(
+        self,
+        prompt: LanguageModelInput,
+        tools: list[dict] | None = None,
+        tool_choice: ToolChoiceOptions | None = None,
     ) -> BaseMessage:
         raise NotImplementedError
 
-    @abc.abstractmethod
     def stream(
         self,
         prompt: LanguageModelInput,
         tools: list[dict] | None = None,
         tool_choice: ToolChoiceOptions | None = None,
+    ) -> Iterator[BaseMessage]:
+        self._precall(prompt)
+        # TODO add a postcall to log model outputs independent of concrete class
+        # implementation
+        return self._stream_implementation(prompt, tools, tool_choice)
+
+    @abc.abstractmethod
+    def _stream_implementation(
+        self,
+        prompt: LanguageModelInput,
+        tools: list[dict] | None = None,
+        tool_choice: ToolChoiceOptions | None = None,
     ) -> Iterator[BaseMessage]:
         raise NotImplementedError

backend/danswer/llm/llm_provider_options.py

@@ -26,6 +26,7 @@ OPENAI_PROVIDER_NAME = "openai"
 OPEN_AI_MODEL_NAMES = [
     "gpt-4",
     "gpt-4o",
+    "gpt-4o-mini",
     "gpt-4-turbo",
     "gpt-4-turbo-preview",
     "gpt-4-1106-preview",

backend/danswer/one_shot_answer/answer_question.py

@@ -10,6 +10,7 @@ from danswer.chat.models import DanswerAnswerPiece
 from danswer.chat.models import DanswerContexts
 from danswer.chat.models import DanswerQuotes
 from danswer.chat.models import LLMRelevanceFilterResponse
+from danswer.chat.models import LLMRelevanceSummaryResponse
 from danswer.chat.models import QADocsResponse
 from danswer.chat.models import StreamingError
 from danswer.configs.chat_configs import MAX_CHUNKS_FED_TO_CHAT
@@ -21,6 +22,7 @@ from danswer.db.chat import create_new_chat_message
 from danswer.db.chat import get_or_create_root_message
 from danswer.db.chat import translate_db_message_to_chat_message_detail
 from danswer.db.chat import translate_db_search_doc_to_server_search_doc
+from danswer.db.chat import update_search_docs_table_with_relevance
 from danswer.db.engine import get_session_context_manager
 from danswer.db.models import User
 from danswer.db.persona import get_prompt_by_id
@@ -48,6 +50,7 @@ from danswer.server.query_and_chat.models import ChatMessageDetail
 from danswer.server.utils import get_json_line
 from danswer.tools.force import ForceUseTool
 from danswer.tools.search.search_tool import SEARCH_DOC_CONTENT_ID
+from danswer.tools.search.search_tool import SEARCH_EVALUATION_ID
 from danswer.tools.search.search_tool import SEARCH_RESPONSE_SUMMARY_ID
 from danswer.tools.search.search_tool import SearchResponseSummary
 from danswer.tools.search.search_tool import SearchTool
@@ -57,6 +60,7 @@ from danswer.tools.tool_runner import ToolCallKickoff
 from danswer.utils.logger import setup_logger
 from danswer.utils.timing import log_generator_function_time
 
+
 logger = setup_logger()
 
 AnswerObjectIterator = Iterator[
@@ -70,6 +74,7 @@ AnswerObjectIterator = Iterator[
     | ChatMessageDetail
     | CitationInfo
     | ToolCallKickoff
+    | LLMRelevanceSummaryResponse
 ]
 
 
@@ -88,8 +93,9 @@ def stream_answer_objects(
     bypass_acl: bool = False,
     use_citations: bool = False,
     danswerbot_flow: bool = False,
-    retrieval_metrics_callback: Callable[[RetrievalMetricsContainer], None]
-    | None = None,
+    retrieval_metrics_callback: (
+        Callable[[RetrievalMetricsContainer], None] | None
+    ) = None,
     rerank_metrics_callback: Callable[[RerankMetricsContainer], None] | None = None,
 ) -> AnswerObjectIterator:
     """Streams in order:
@@ -127,6 +133,7 @@ def stream_answer_objects(
         user_query=query_msg.message,
         history_str=history_str,
     )
+
     # Given back ahead of the documents for latency reasons
     # In chat flow it's given back along with the documents
     yield QueryRephrase(rephrased_query=rephrased_query)
@@ -182,6 +189,7 @@ def stream_answer_objects(
         chunks_below=query_req.chunks_below,
         full_doc=query_req.full_doc,
         bypass_acl=bypass_acl,
+        llm_doc_eval=query_req.llm_doc_eval,
     )
 
     answer_config = AnswerStyleConfig(
@@ -189,6 +197,7 @@ def stream_answer_objects(
         quotes_config=QuotesConfig() if not use_citations else None,
         document_pruning_config=document_pruning_config,
     )
+
     answer = Answer(
         question=query_msg.message,
         answer_style_config=answer_config,
@@ -204,12 +213,16 @@ def stream_answer_objects(
         # tested quotes with tool calling too much yet
         skip_explicit_tool_calling=True,
         return_contexts=query_req.return_contexts,
+        skip_gen_ai_answer_generation=query_req.skip_gen_ai_answer_generation,
     )
+
     # won't be any ImageGenerationDisplay responses since that tool is never passed in
     dropped_inds: list[int] = []
+
     for packet in cast(AnswerObjectIterator, answer.processed_streamed_output):
         # for one-shot flow, don't currently do anything with these
         if isinstance(packet, ToolResponse):
+            # (likely fine that it comes after the initial creation of the search docs)
             if packet.id == SEARCH_RESPONSE_SUMMARY_ID:
                 search_response_summary = cast(SearchResponseSummary, packet.response)
 
@@ -242,6 +255,7 @@ def stream_answer_objects(
                     recency_bias_multiplier=search_response_summary.recency_bias_multiplier,
                 )
                 yield initial_response
+
             elif packet.id == SECTION_RELEVANCE_LIST_ID:
                 chunk_indices = packet.response
 
@@ -253,8 +267,21 @@ def stream_answer_objects(
                     )
 
                 yield LLMRelevanceFilterResponse(relevant_chunk_indices=packet.response)
+
             elif packet.id == SEARCH_DOC_CONTENT_ID:
                 yield packet.response
+
+            elif packet.id == SEARCH_EVALUATION_ID:
+                evaluation_response = LLMRelevanceSummaryResponse(
+                    relevance_summaries=packet.response
+                )
+                if reference_db_search_docs is not None:
+                    update_search_docs_table_with_relevance(
+                        db_session=db_session,
+                        reference_db_search_docs=reference_db_search_docs,
+                        relevance_summary=evaluation_response,
+                    )
+                yield evaluation_response
         else:
             yield packet
 
@@ -275,7 +302,6 @@ def stream_answer_objects(
     msg_detail_response = translate_db_message_to_chat_message_detail(
         gen_ai_response_message
     )
-
     yield msg_detail_response
 
 
@@ -309,8 +335,9 @@ def get_search_answer(
     bypass_acl: bool = False,
     use_citations: bool = False,
     danswerbot_flow: bool = False,
-    retrieval_metrics_callback: Callable[[RetrievalMetricsContainer], None]
-    | None = None,
+    retrieval_metrics_callback: (
+        Callable[[RetrievalMetricsContainer], None] | None
+    ) = None,
     rerank_metrics_callback: Callable[[RerankMetricsContainer], None] | None = None,
 ) -> OneShotQAResponse:
     """Collects the streamed one shot answer responses into a single object"""

backend/danswer/one_shot_answer/models.py

@@ -27,12 +27,19 @@ class DirectQARequest(ChunkContext):
     messages: list[ThreadMessage]
     prompt_id: int | None
     persona_id: int
+    agentic: bool | None = None
     retrieval_options: RetrievalDetails = Field(default_factory=RetrievalDetails)
     # This is to forcibly skip (or run) the step, if None it uses the system defaults
     skip_rerank: bool | None = None
     skip_llm_chunk_filter: bool | None = None
     chain_of_thought: bool = False
     return_contexts: bool = False
+    # This is to toggle agentic evaluation:
+    # 1. Evaluates whether each response is relevant or not
+    # 2. Provides a summary of the document's relevance in the resulsts
+    llm_doc_eval: bool = False
+    # If True, skips generative an AI response to the search query
+    skip_gen_ai_answer_generation: bool = False
 
     @root_validator
     def check_chain_of_thought_and_prompt_id(

backend/danswer/prompts/agentic_evaluation.py

@@ -0,0 +1,43 @@
+AGENTIC_SEARCH_SYSTEM_PROMPT = """
+You are an expert at evaluating the relevance of a document to a search query.
+Provided a document and a search query, you determine if the document is relevant to the user query.
+You ALWAYS output the 3 sections described below and every section always begins with the same header line.
+The "Chain of Thought" is to help you understand the document and query and their relevance to one another.
+The "Useful Analysis" is shown to the user to help them understand why the document is or is not useful for them.
+The "Final Relevance Determination" is always a single True or False.
+
+You always output your response following these 3 sections:
+
+1. Chain of Thought:
+Provide a chain of thought analysis considering:
+- The main purpose and content of the document
+- What the user is searching for
+- How the document relates to the query
+- Potential uses of the document for the given query
+Be thorough, but avoid unnecessary repetition. Think step by step.
+
+2. Useful Analysis:
+Summarize the contents of the document as it relates to the user query.
+BE ABSOLUTELY AS CONCISE AS POSSIBLE.
+If the document is not useful, briefly mention the what the document is about.
+Do NOT say whether this document is useful or not useful, ONLY provide the summary.
+If referring to the document, prefer using "this" document over "the" document.
+
+3. Final Relevance Determination:
+True or False
+"""
+
+AGENTIC_SEARCH_USER_PROMPT = """
+Document:
+```
+{content}
+```
+
+Query:
+{query}
+
+Be sure to run through the 3 steps of evaluation:
+1. Chain of Thought
+2. Useful Analysis
+3. Final Relevance Determination
+""".strip()

backend/danswer/search/models.py

@@ -130,11 +130,14 @@ class InferenceChunk(BaseChunk):
     recency_bias: float
     score: float | None
     hidden: bool
+    is_relevant: bool | None = None
+    relevance_explanation: str | None = None
     metadata: dict[str, str | list[str]]
     # Matched sections in the chunk. Uses Vespa syntax e.g. <hi>TEXT</hi>
     # to specify that a set of words should be highlighted. For example:
     # ["<hi>the</hi> <hi>answer</hi> is 42", "he couldn't find an <hi>answer</hi>"]
     match_highlights: list[str]
+
     # when the doc was last updated
     updated_at: datetime | None
     primary_owners: list[str] | None = None
@@ -227,6 +230,8 @@ class SearchDoc(BaseModel):
     hidden: bool
     metadata: dict[str, str | list[str]]
     score: float | None
+    is_relevant: bool | None = None
+    relevance_explanation: str | None = None
     # Matched sections in the doc. Uses Vespa syntax e.g. <hi>TEXT</hi>
     # to specify that a set of words should be highlighted. For example:
     # ["<hi>the</hi> <hi>answer</hi> is 42", "the answer is <hi>42</hi>""]

backend/danswer/search/pipeline.py

@@ -5,10 +5,14 @@ from typing import cast
 
 from sqlalchemy.orm import Session
 
+from danswer.chat.models import RelevanceChunk
+from danswer.configs.chat_configs import DISABLE_AGENTIC_SEARCH
 from danswer.configs.chat_configs import MULTILINGUAL_QUERY_EXPANSION
 from danswer.db.embedding_model import get_current_db_embedding_model
 from danswer.db.models import User
 from danswer.document_index.factory import get_default_document_index
+from danswer.llm.answering.models import DocumentPruningConfig
+from danswer.llm.answering.models import PromptConfig
 from danswer.llm.answering.prune_and_merge import ChunkRange
 from danswer.llm.answering.prune_and_merge import merge_chunk_intervals
 from danswer.llm.interfaces import LLM
@@ -25,7 +29,10 @@ from danswer.search.postprocessing.postprocessing import search_postprocessing
 from danswer.search.preprocessing.preprocessing import retrieval_preprocessing
 from danswer.search.retrieval.search_runner import retrieve_chunks
 from danswer.search.utils import inference_section_from_chunks
+from danswer.secondary_llm_flows.agentic_evaluation import evaluate_inference_section
 from danswer.utils.logger import setup_logger
+from danswer.utils.threadpool_concurrency import FunctionCall
+from danswer.utils.threadpool_concurrency import run_functions_in_parallel
 from danswer.utils.threadpool_concurrency import run_functions_tuples_in_parallel
 
 logger = setup_logger()
@@ -40,9 +47,12 @@ class SearchPipeline:
         fast_llm: LLM,
         db_session: Session,
         bypass_acl: bool = False,  # NOTE: VERY DANGEROUS, USE WITH CAUTION
-        retrieval_metrics_callback: Callable[[RetrievalMetricsContainer], None]
-        | None = None,
+        retrieval_metrics_callback: (
+            Callable[[RetrievalMetricsContainer], None] | None
+        ) = None,
         rerank_metrics_callback: Callable[[RerankMetricsContainer], None] | None = None,
+        prompt_config: PromptConfig | None = None,
+        pruning_config: DocumentPruningConfig | None = None,
     ):
         self.search_request = search_request
         self.user = user
@@ -58,6 +68,8 @@ class SearchPipeline:
             primary_index_name=self.embedding_model.index_name,
             secondary_index_name=None,
         )
+        self.prompt_config: PromptConfig | None = prompt_config
+        self.pruning_config: DocumentPruningConfig | None = pruning_config
 
         # Preprocessing steps generate this
         self._search_query: SearchQuery | None = None
@@ -74,9 +86,9 @@ class SearchPipeline:
         self._relevant_section_indices: list[int] | None = None
 
         # Generates reranked chunks and LLM selections
-        self._postprocessing_generator: Iterator[
-            list[InferenceSection] | list[int]
-        ] | None = None
+        self._postprocessing_generator: (
+            Iterator[list[InferenceSection] | list[int]] | None
+        ) = None
 
     """Pre-processing"""
 
@@ -323,6 +335,32 @@ class SearchPipeline:
         )
         return self._relevant_section_indices
 
+    @property
+    def relevance_summaries(self) -> dict[str, RelevanceChunk]:
+        if DISABLE_AGENTIC_SEARCH:
+            raise ValueError(
+                "Agentic saerch operation called while DISABLE_AGENTIC_SEARCH is toggled"
+            )
+        if len(self.reranked_sections) == 0:
+            logger.warning(
+                "No sections found in agentic search evalution. Returning empty dict."
+            )
+            return {}
+
+        sections = self.reranked_sections
+        functions = [
+            FunctionCall(
+                evaluate_inference_section, (section, self.search_query.query, self.llm)
+            )
+            for section in sections
+        ]
+
+        results = run_functions_in_parallel(function_calls=functions)
+
+        return {
+            next(iter(value)): value[next(iter(value))] for value in results.values()
+        }
+
     @property
     def section_relevance_list(self) -> list[bool]:
         return [

backend/danswer/search/postprocessing/postprocessing.py

@@ -278,17 +278,17 @@ def search_postprocessing(
             _log_top_section_links(search_query.search_type.value, reranked_sections)
             yield reranked_sections
 
-    llm_section_selection = cast(
-        list[str] | None,
-        post_processing_results.get(str(llm_filter_task_id))
-        if llm_filter_task_id
-        else None,
-    )
-    if llm_section_selection is not None:
-        yield [
-            index
-            for index, section in enumerate(reranked_sections or retrieved_sections)
-            if section.center_chunk.unique_id in llm_section_selection
+    llm_selected_section_ids = (
+        [
+            section.center_chunk.unique_id
+            for section in post_processing_results.get(str(llm_filter_task_id), [])
         ]
-    else:
-        yield cast(list[int], [])
+        if llm_filter_task_id
+        else []
+    )
+
+    yield [
+        index
+        for index, section in enumerate(reranked_sections or retrieved_sections)
+        if section.center_chunk.unique_id in llm_selected_section_ids
+    ]

backend/danswer/secondary_llm_flows/agentic_evaluation.py

@@ -0,0 +1,70 @@
+import re
+
+from danswer.chat.models import RelevanceChunk
+from danswer.llm.interfaces import LLM
+from danswer.llm.utils import dict_based_prompt_to_langchain_prompt
+from danswer.llm.utils import message_to_string
+from danswer.prompts.agentic_evaluation import AGENTIC_SEARCH_SYSTEM_PROMPT
+from danswer.prompts.agentic_evaluation import AGENTIC_SEARCH_USER_PROMPT
+from danswer.search.models import InferenceSection
+from danswer.utils.logger import setup_logger
+
+logger = setup_logger()
+
+
+def _get_agent_eval_messages(
+    title: str, content: str, query: str
+) -> list[dict[str, str]]:
+    messages = [
+        {
+            "role": "system",
+            "content": AGENTIC_SEARCH_SYSTEM_PROMPT,
+        },
+        {
+            "role": "user",
+            "content": AGENTIC_SEARCH_USER_PROMPT.format(
+                title=title, content=content, query=query
+            ),
+        },
+    ]
+    return messages
+
+
+def evaluate_inference_section(
+    document: InferenceSection, query: str, llm: LLM
+) -> dict[str, RelevanceChunk]:
+    results = {}
+
+    document_id = document.center_chunk.document_id
+    semantic_id = document.center_chunk.semantic_identifier
+    contents = document.combined_content
+    chunk_id = document.center_chunk.chunk_id
+
+    messages = _get_agent_eval_messages(
+        title=semantic_id, content=contents, query=query
+    )
+    filled_llm_prompt = dict_based_prompt_to_langchain_prompt(messages)
+    model_output = message_to_string(llm.invoke(filled_llm_prompt))
+
+    # Search for the "Useful Analysis" section in the model output
+    # This regex looks for "2. Useful Analysis" (case-insensitive) followed by an optional colon,
+    # then any text up to "3. Final Relevance"
+    # The (?i) flag makes it case-insensitive, and re.DOTALL allows the dot to match newlines
+    # If no match is found, the entire model output is used as the analysis
+    analysis_match = re.search(
+        r"(?i)2\.\s*useful analysis:?\s*(.+?)\n\n3\.\s*final relevance",
+        model_output,
+        re.DOTALL,
+    )
+    analysis = analysis_match.group(1).strip() if analysis_match else model_output
+
+    # Get the last non-empty line
+    last_line = next(
+        (line for line in reversed(model_output.split("\n")) if line.strip()), ""
+    )
+    relevant = last_line.strip().lower().startswith("true")
+
+    results[f"{document_id}-{chunk_id}"] = RelevanceChunk(
+        relevant=relevant, content=analysis
+    )
+    return results

backend/danswer/server/query_and_chat/chat_backend.py

@@ -303,7 +303,6 @@ def handle_new_chat_message(
             request.headers
         ),
     )
-
     return StreamingResponse(packets, media_type="application/json")
 
 
@@ -484,6 +483,7 @@ def upload_files_for_chat(
         "text/tab-separated-values",
         "application/json",
         "application/xml",
+        "text/xml",
         "application/x-yaml",
     }
     document_content_types = {

backend/danswer/server/query_and_chat/models.py

@@ -171,7 +171,6 @@ class SearchFeedbackRequest(BaseModel):
 
         if click is False and feedback is None:
             raise ValueError("Empty feedback received.")
-
         return values
 
 
@@ -186,6 +185,7 @@ class ChatMessageDetail(BaseModel):
     time_sent: datetime
     alternate_assistant_id: str | None
     # Dict mapping citation number to db_doc_id
+    chat_session_id: int | None = None
     citations: dict[int, int] | None
     files: list[FileDescriptor]
     tool_calls: list[ToolCallFinalResult]
@@ -196,6 +196,13 @@ class ChatMessageDetail(BaseModel):
         return initial_dict
 
 
+class SearchSessionDetailResponse(BaseModel):
+    search_session_id: int
+    description: str
+    documents: list[SearchDoc]
+    messages: list[ChatMessageDetail]
+
+
 class ChatSessionDetailResponse(BaseModel):
     chat_session_id: int
     description: str

backend/danswer/server/query_and_chat/query_backend.py

@@ -7,6 +7,14 @@ from sqlalchemy.orm import Session
 from danswer.auth.users import current_admin_user
 from danswer.auth.users import current_user
 from danswer.configs.constants import DocumentSource
+from danswer.configs.constants import MessageType
+from danswer.db.chat import get_chat_messages_by_session
+from danswer.db.chat import get_chat_session_by_id
+from danswer.db.chat import get_chat_sessions_by_user
+from danswer.db.chat import get_first_messages_for_chat_sessions
+from danswer.db.chat import get_search_docs_for_chat_message
+from danswer.db.chat import translate_db_message_to_chat_message_detail
+from danswer.db.chat import translate_db_search_doc_to_server_search_doc
 from danswer.db.embedding_model import get_current_db_embedding_model
 from danswer.db.engine import get_session
 from danswer.db.models import User
@@ -24,8 +32,11 @@ from danswer.secondary_llm_flows.query_validation import get_query_answerability
 from danswer.secondary_llm_flows.query_validation import stream_query_answerability
 from danswer.server.query_and_chat.models import AdminSearchRequest
 from danswer.server.query_and_chat.models import AdminSearchResponse
+from danswer.server.query_and_chat.models import ChatSessionDetails
+from danswer.server.query_and_chat.models import ChatSessionsResponse
 from danswer.server.query_and_chat.models import HelperResponse
 from danswer.server.query_and_chat.models import QueryValidationResponse
+from danswer.server.query_and_chat.models import SearchSessionDetailResponse
 from danswer.server.query_and_chat.models import SimpleQueryRequest
 from danswer.server.query_and_chat.models import SourceTag
 from danswer.server.query_and_chat.models import TagResponse
@@ -46,7 +57,6 @@ def admin_search(
 ) -> AdminSearchResponse:
     query = question.query
     logger.info(f"Received admin search query: {query}")
-
     user_acl_filters = build_access_filters_for_user(user, db_session)
     final_filters = IndexFilters(
         source_type=question.filters.source_type,
@@ -55,19 +65,15 @@ def admin_search(
         tags=question.filters.tags,
         access_control_list=user_acl_filters,
     )
-
     embedding_model = get_current_db_embedding_model(db_session)
-
     document_index = get_default_document_index(
         primary_index_name=embedding_model.index_name, secondary_index_name=None
     )
-
     if not isinstance(document_index, VespaIndex):
         raise HTTPException(
             status_code=400,
             detail="Cannot use admin-search when using a non-Vespa document index",
         )
-
     matching_chunks = document_index.admin_retrieval(query=query, filters=final_filters)
 
     documents = chunks_or_sections_to_search_docs(matching_chunks)
@@ -136,6 +142,103 @@ def query_validation(
     return QueryValidationResponse(reasoning=reasoning, answerable=answerable)
 
 
+@basic_router.get("/user-searches")
+def get_user_search_sessions(
+    user: User | None = Depends(current_user),
+    db_session: Session = Depends(get_session),
+) -> ChatSessionsResponse:
+    user_id = user.id if user is not None else None
+
+    try:
+        search_sessions = get_chat_sessions_by_user(
+            user_id=user_id, deleted=False, db_session=db_session, only_one_shot=True
+        )
+    except ValueError:
+        raise HTTPException(
+            status_code=404, detail="Chat session does not exist or has been deleted"
+        )
+
+    search_session_ids = [chat.id for chat in search_sessions]
+    first_messages = get_first_messages_for_chat_sessions(
+        search_session_ids, db_session
+    )
+    first_messages_dict = dict(first_messages)
+
+    response = ChatSessionsResponse(
+        sessions=[
+            ChatSessionDetails(
+                id=search.id,
+                name=first_messages_dict.get(search.id, search.description),
+                persona_id=search.persona_id,
+                time_created=search.time_created.isoformat(),
+                shared_status=search.shared_status,
+                folder_id=search.folder_id,
+                current_alternate_model=search.current_alternate_model,
+            )
+            for search in search_sessions
+        ]
+    )
+    return response
+
+
+@basic_router.get("/search-session/{session_id}")
+def get_search_session(
+    session_id: int,
+    is_shared: bool = False,
+    user: User | None = Depends(current_user),
+    db_session: Session = Depends(get_session),
+) -> SearchSessionDetailResponse:
+    user_id = user.id if user is not None else None
+
+    try:
+        search_session = get_chat_session_by_id(
+            chat_session_id=session_id,
+            user_id=user_id,
+            db_session=db_session,
+            is_shared=is_shared,
+        )
+    except ValueError:
+        raise ValueError("Search session does not exist or has been deleted")
+
+    session_messages = get_chat_messages_by_session(
+        chat_session_id=session_id,
+        user_id=user_id,
+        db_session=db_session,
+        # we already did a permission check above with the call to
+        # `get_chat_session_by_id`, so we can skip it here
+        skip_permission_check=True,
+        # we need the tool call objs anyways, so just fetch them in a single call
+        prefetch_tool_calls=True,
+    )
+    docs_response: list[SearchDoc] = []
+    for message in session_messages:
+        if (
+            message.message_type == MessageType.ASSISTANT
+            or message.message_type == MessageType.SYSTEM
+        ):
+            docs = get_search_docs_for_chat_message(
+                db_session=db_session, chat_message_id=message.id
+            )
+            for doc in docs:
+                server_doc = translate_db_search_doc_to_server_search_doc(doc)
+                docs_response.append(server_doc)
+
+    response = SearchSessionDetailResponse(
+        search_session_id=session_id,
+        description=search_session.description,
+        documents=docs_response,
+        messages=[
+            translate_db_message_to_chat_message_detail(
+                msg, remove_doc_content=is_shared  # if shared, don't leak doc content
+            )
+            for msg in session_messages
+        ],
+    )
+    return response
+
+
+# NOTE No longer used, after search/chat redesign.
+# No search responses are answered with a conversational generative AI response
 @basic_router.post("/stream-query-validation")
 def stream_query_validation(
     simple_query: SimpleQueryRequest, _: User = Depends(current_user)
@@ -156,6 +259,7 @@ def get_answer_with_quote(
     _: None = Depends(check_token_rate_limits),
 ) -> StreamingResponse:
     query = query_request.messages[0].message
+
     logger.info(f"Received query for one shot answer with quotes: {query}")
     packets = stream_search_answer(
         query_req=query_request,

backend/danswer/tools/search/search_tool.py

@@ -10,6 +10,7 @@ from danswer.chat.chat_utils import llm_doc_from_inference_section
 from danswer.chat.models import DanswerContext
 from danswer.chat.models import DanswerContexts
 from danswer.chat.models import LlmDoc
+from danswer.configs.chat_configs import DISABLE_AGENTIC_SEARCH
 from danswer.db.models import Persona
 from danswer.db.models import User
 from danswer.dynamic_configs.interface import JSON_ro
@@ -30,11 +31,15 @@ from danswer.secondary_llm_flows.query_expansion import history_based_query_reph
 from danswer.tools.search.search_utils import llm_doc_to_dict
 from danswer.tools.tool import Tool
 from danswer.tools.tool import ToolResponse
+from danswer.utils.logger import setup_logger
+
+logger = setup_logger()
 
 SEARCH_RESPONSE_SUMMARY_ID = "search_response_summary"
 SEARCH_DOC_CONTENT_ID = "search_doc_content"
 SECTION_RELEVANCE_LIST_ID = "section_relevance_list"
 FINAL_CONTEXT_DOCUMENTS = "final_context_documents"
+SEARCH_EVALUATION_ID = "llm_doc_eval"
 
 
 class SearchResponseSummary(BaseModel):
@@ -80,6 +85,7 @@ class SearchTool(Tool):
         chunks_below: int = 0,
         full_doc: bool = False,
         bypass_acl: bool = False,
+        llm_doc_eval: bool = False,
     ) -> None:
         self.user = user
         self.persona = persona
@@ -96,6 +102,7 @@ class SearchTool(Tool):
         self.full_doc = full_doc
         self.bypass_acl = bypass_acl
         self.db_session = db_session
+        self.llm_doc_eval = llm_doc_eval
 
     @property
     def name(self) -> str:
@@ -218,23 +225,28 @@ class SearchTool(Tool):
                     self.retrieval_options.filters if self.retrieval_options else None
                 ),
                 persona=self.persona,
-                offset=self.retrieval_options.offset
-                if self.retrieval_options
-                else None,
+                offset=(
+                    self.retrieval_options.offset if self.retrieval_options else None
+                ),
                 limit=self.retrieval_options.limit if self.retrieval_options else None,
                 chunks_above=self.chunks_above,
                 chunks_below=self.chunks_below,
                 full_doc=self.full_doc,
-                enable_auto_detect_filters=self.retrieval_options.enable_auto_detect_filters
-                if self.retrieval_options
-                else None,
+                enable_auto_detect_filters=(
+                    self.retrieval_options.enable_auto_detect_filters
+                    if self.retrieval_options
+                    else None
+                ),
             ),
             user=self.user,
             llm=self.llm,
             fast_llm=self.fast_llm,
             bypass_acl=self.bypass_acl,
             db_session=self.db_session,
+            prompt_config=self.prompt_config,
+            pruning_config=self.pruning_config,
         )
+
         yield ToolResponse(
             id=SEARCH_RESPONSE_SUMMARY_ID,
             response=SearchResponseSummary(
@@ -246,6 +258,7 @@ class SearchTool(Tool):
                 recency_bias_multiplier=search_pipeline.search_query.recency_bias_multiplier,
             ),
         )
+
         yield ToolResponse(
             id=SEARCH_DOC_CONTENT_ID,
             response=DanswerContexts(
@@ -260,6 +273,7 @@ class SearchTool(Tool):
                 ]
             ),
         )
+
         yield ToolResponse(
             id=SECTION_RELEVANCE_LIST_ID,
             response=search_pipeline.relevant_section_indices,
@@ -281,6 +295,11 @@ class SearchTool(Tool):
 
         yield ToolResponse(id=FINAL_CONTEXT_DOCUMENTS, response=llm_docs)
 
+        if self.llm_doc_eval and not DISABLE_AGENTIC_SEARCH:
+            yield ToolResponse(
+                id=SEARCH_EVALUATION_ID, response=search_pipeline.relevance_summaries
+            )
+
     def final_result(self, *args: ToolResponse) -> JSON_ro:
         final_docs = cast(
             list[LlmDoc],

backend/danswer/tools/utils.py

@@ -6,7 +6,13 @@ from danswer.llm.utils import get_default_llm_tokenizer
 from danswer.tools.tool import Tool
 
 
-OPEN_AI_TOOL_CALLING_MODELS = {"gpt-3.5-turbo", "gpt-4-turbo", "gpt-4"}
+OPEN_AI_TOOL_CALLING_MODELS = {
+    "gpt-3.5-turbo",
+    "gpt-4-turbo",
+    "gpt-4",
+    "gpt-4o",
+    "gpt-4o-mini",
+}
 
 
 def explicit_tool_calling_supported(model_provider: str, model_name: str) -> bool:

backend/danswer/utils/threadpool_concurrency.py

@@ -87,6 +87,7 @@ def run_functions_in_parallel(
     are the result_id of the FunctionCall and the values are the results of the call.
     """
     results = {}
+
     with ThreadPoolExecutor(max_workers=len(function_calls)) as executor:
         future_to_id = {
             executor.submit(func_call.execute): func_call.result_id

backend/ee/danswer/server/query_and_chat/chat_backend.py

@@ -33,6 +33,7 @@ def translate_doc_response_to_simple_doc(
 ) -> list[SimpleDoc]:
     return [
         SimpleDoc(
+            id=doc.document_id,
             semantic_identifier=doc.semantic_identifier,
             link=doc.link,
             blurb=doc.blurb,

backend/ee/danswer/server/query_and_chat/models.py

@@ -44,6 +44,7 @@ class BasicCreateChatMessageRequest(ChunkContext):
 
 
 class SimpleDoc(BaseModel):
+    id: str
     semantic_identifier: str
     link: str | None
     blurb: str

backend/scripts/api_inference_sample.py

@@ -33,6 +33,7 @@ def process_question(danswer_url: str, question: str, api_key: str | None) -> No
         "message": question,
         "chat_session_id": chat_session_id,
         "parent_message_id": None,
+        "file_descriptors": [],
         # Default Question Answer prompt
         "prompt_id": 0,
         # Not specifying any specific docs to chat to, we want to run a search

backend/scripts/test-openapi-key.py

@@ -2,6 +2,8 @@ from openai import OpenAI
 
 
 VALID_MODEL_LIST = [
+    "gpt-4o-mini",
+    "gpt-4o",
     "gpt-4-1106-preview",
     "gpt-4-vision-preview",
     "gpt-4",

backend/shared_configs/enums.py

@@ -0,0 +1,6 @@
+from enum import Enum
+
+
+class EmbedTextType(str, Enum):
+    QUERY = "query"
+    PASSAGE = "passage"

backend/tests/regression/answer_quality/README.md

@@ -53,8 +53,10 @@ Edit `search_test_config.yaml` to set:
     - The path to the zip file containing the files you'd like to test against
 - questions_file
     - The path to the yaml containing the questions you'd like to test with 
-- branch
-    - Set the branch to null if you want it to just use the code as is
+- commit_sha
+    - Set this to the SHA of the commit you want to run the test against
+    - You must clear all local changes if you want to use this option
+    - Set this to null if you want it to just use the code as is
 - clean_up_docker_containers
     - Set this to true to automatically delete all docker containers, networks and volumes after the test
 - launch_web_ui
@@ -68,10 +70,10 @@ Edit `search_test_config.yaml` to set:
 - model_server_port
     - This is the port of the remote model server
     - Only need to set this if use_cloud_gpu is true
-- existing_test_suffix
+- existing_test_suffix (THIS IS NOT A SUFFIX ANYMORE, TODO UPDATE THE DOCS HERE)
     - Use this if you would like to relaunch a previous test instance
     - Input the suffix of the test you'd like to re-launch 
-    - (E.g. to use the data from folder "test_1234_5678" put "_1234_5678")
+    - (E.g. to use the data from folder "test-1234-5678" put "-1234-5678")
     - No new files will automatically be uploaded
     - Leave empty to run a new test
 - limit

backend/tests/regression/answer_quality/api_utils.py

@@ -2,25 +2,44 @@ import requests
 from retry import retry
 
 from danswer.configs.constants import DocumentSource
-from danswer.configs.constants import MessageType
 from danswer.connectors.models import InputType
 from danswer.db.enums import IndexingStatus
-from danswer.one_shot_answer.models import DirectQARequest
-from danswer.one_shot_answer.models import ThreadMessage
 from danswer.search.models import IndexFilters
 from danswer.search.models import OptionalSearchSetting
 from danswer.search.models import RetrievalDetails
 from danswer.server.documents.models import ConnectorBase
-from tests.regression.answer_quality.cli_utils import (
-    get_api_server_host_port,
-)
+from danswer.server.query_and_chat.models import ChatSessionCreationRequest
+from ee.danswer.server.query_and_chat.models import BasicCreateChatMessageRequest
+from tests.regression.answer_quality.cli_utils import get_api_server_host_port
+
+GENERAL_HEADERS = {"Content-Type": "application/json"}
 
 
 def _api_url_builder(run_suffix: str, api_path: str) -> str:
     return f"http://localhost:{get_api_server_host_port(run_suffix)}" + api_path
 
 
-@retry(tries=5, delay=2, backoff=2)
+def _create_new_chat_session(run_suffix: str) -> int:
+    create_chat_request = ChatSessionCreationRequest(
+        persona_id=0,
+        description=None,
+    )
+    body = create_chat_request.dict()
+
+    create_chat_url = _api_url_builder(run_suffix, "/chat/create-chat-session/")
+
+    response_json = requests.post(
+        create_chat_url, headers=GENERAL_HEADERS, json=body
+    ).json()
+    chat_session_id = response_json.get("chat_session_id")
+
+    if isinstance(chat_session_id, int):
+        return chat_session_id
+    else:
+        raise RuntimeError(response_json)
+
+
+@retry(tries=10, delay=10)
 def get_answer_from_query(query: str, run_suffix: str) -> tuple[list[str], str]:
     filters = IndexFilters(
         source_type=None,
@@ -29,49 +48,49 @@ def get_answer_from_query(query: str, run_suffix: str) -> tuple[list[str], str]:
         tags=None,
         access_control_list=None,
     )
-
-    messages = [ThreadMessage(message=query, sender=None, role=MessageType.USER)]
-
-    new_message_request = DirectQARequest(
-        messages=messages,
-        prompt_id=0,
-        persona_id=0,
-        retrieval_options=RetrievalDetails(
-            run_search=OptionalSearchSetting.ALWAYS,
-            real_time=True,
-            filters=filters,
-            enable_auto_detect_filters=False,
-        ),
-        chain_of_thought=False,
-        return_contexts=True,
+    retrieval_options = RetrievalDetails(
+        run_search=OptionalSearchSetting.ALWAYS,
+        real_time=True,
+        filters=filters,
+        enable_auto_detect_filters=False,
     )
 
-    url = _api_url_builder(run_suffix, "/query/answer-with-quote/")
-    headers = {
-        "Content-Type": "application/json",
-    }
+    chat_session_id = _create_new_chat_session(run_suffix)
+
+    url = _api_url_builder(run_suffix, "/chat/send-message-simple-api/")
+
+    new_message_request = BasicCreateChatMessageRequest(
+        chat_session_id=chat_session_id,
+        message=query,
+        retrieval_options=retrieval_options,
+        query_override=query,
+    )
 
     body = new_message_request.dict()
     body["user"] = None
     try:
-        response_json = requests.post(url, headers=headers, json=body).json()
-        context_data_list = response_json.get("contexts", {}).get("contexts", [])
+        response_json = requests.post(url, headers=GENERAL_HEADERS, json=body).json()
+        simple_search_docs = response_json.get("simple_search_docs", [])
         answer = response_json.get("answer", "")
     except Exception as e:
-        print("Failed to answer the questions, trying again")
-        print(f"error: {str(e)}")
+        print("Failed to answer the questions:")
+        print(f"\t {str(e)}")
+        print("trying again")
         raise e
 
-    return context_data_list, answer
+    return simple_search_docs, answer
 
 
+@retry(tries=10, delay=10)
 def check_if_query_ready(run_suffix: str) -> bool:
     url = _api_url_builder(run_suffix, "/manage/admin/connector/indexing-status/")
-    headers = {
-        "Content-Type": "application/json",
-    }
-
-    indexing_status_dict = requests.get(url, headers=headers).json()
+    try:
+        indexing_status_dict = requests.get(url, headers=GENERAL_HEADERS).json()
+    except Exception as e:
+        print("Failed to check indexing status, API server is likely starting up:")
+        print(f"\t {str(e)}")
+        print("trying again")
+        raise e
 
     ongoing_index_attempts = False
     doc_count = 0
@@ -93,17 +112,13 @@ def check_if_query_ready(run_suffix: str) -> bool:
 
 def run_cc_once(run_suffix: str, connector_id: int, credential_id: int) -> None:
     url = _api_url_builder(run_suffix, "/manage/admin/connector/run-once/")
-    headers = {
-        "Content-Type": "application/json",
-    }
-
     body = {
         "connector_id": connector_id,
         "credential_ids": [credential_id],
         "from_beginning": True,
     }
     print("body:", body)
-    response = requests.post(url, headers=headers, json=body)
+    response = requests.post(url, headers=GENERAL_HEADERS, json=body)
     if response.status_code == 200:
         print("Connector created successfully:", response.json())
     else:
@@ -115,13 +130,10 @@ def create_cc_pair(run_suffix: str, connector_id: int, credential_id: int) -> No
     url = _api_url_builder(
         run_suffix, f"/manage/connector/{connector_id}/credential/{credential_id}"
     )
-    headers = {
-        "Content-Type": "application/json",
-    }
 
     body = {"name": "zip_folder_contents", "is_public": True}
     print("body:", body)
-    response = requests.put(url, headers=headers, json=body)
+    response = requests.put(url, headers=GENERAL_HEADERS, json=body)
     if response.status_code == 200:
         print("Connector created successfully:", response.json())
     else:
@@ -131,14 +143,12 @@ def create_cc_pair(run_suffix: str, connector_id: int, credential_id: int) -> No
 
 def _get_existing_connector_names(run_suffix: str) -> list[str]:
     url = _api_url_builder(run_suffix, "/manage/connector")
-    headers = {
-        "Content-Type": "application/json",
-    }
+
     body = {
         "credential_json": {},
         "admin_public": True,
     }
-    response = requests.get(url, headers=headers, json=body)
+    response = requests.get(url, headers=GENERAL_HEADERS, json=body)
     if response.status_code == 200:
         connectors = response.json()
         return [connector["name"] for connector in connectors]
@@ -148,9 +158,6 @@ def _get_existing_connector_names(run_suffix: str) -> list[str]:
 
 def create_connector(run_suffix: str, file_paths: list[str]) -> int:
     url = _api_url_builder(run_suffix, "/manage/admin/connector")
-    headers = {
-        "Content-Type": "application/json",
-    }
     connector_name = base_connector_name = "search_eval_connector"
     existing_connector_names = _get_existing_connector_names(run_suffix)
 
@@ -171,7 +178,7 @@ def create_connector(run_suffix: str, file_paths: list[str]) -> int:
 
     body = connector.dict()
     print("body:", body)
-    response = requests.post(url, headers=headers, json=body)
+    response = requests.post(url, headers=GENERAL_HEADERS, json=body)
     if response.status_code == 200:
         print("Connector created successfully:", response.json())
         return response.json()["id"]
@@ -181,14 +188,11 @@ def create_connector(run_suffix: str, file_paths: list[str]) -> int:
 
 def create_credential(run_suffix: str) -> int:
     url = _api_url_builder(run_suffix, "/manage/credential")
-    headers = {
-        "Content-Type": "application/json",
-    }
     body = {
         "credential_json": {},
         "admin_public": True,
     }
-    response = requests.post(url, headers=headers, json=body)
+    response = requests.post(url, headers=GENERAL_HEADERS, json=body)
     if response.status_code == 200:
         print("credential created successfully:", response.json())
         return response.json()["id"]

backend/tests/regression/answer_quality/cli_utils.py

@@ -3,9 +3,12 @@ import os
 import socket
 import subprocess
 import sys
+import time
+from datetime import datetime
 from threading import Thread
 from typing import IO
 
+import yaml
 from retry import retry
 
 
@@ -57,17 +60,44 @@ def get_current_commit_sha() -> str:
     return sha
 
 
-def switch_to_branch(branch: str) -> None:
-    print(f"Switching to branch: {branch}...")
-    _run_command(f"git checkout {branch}")
-    _run_command("git pull")
-    print(f"Successfully switched to branch: {branch}")
+def switch_to_commit(commit_sha: str) -> None:
+    print(f"Switching to commit: {commit_sha}...")
+    _run_command(f"git checkout {commit_sha}")
+    print(f"Successfully switched to commit: {commit_sha}")
     print("Repository updated successfully.")
 
 
-def manage_data_directories(suffix: str, base_path: str, use_cloud_gpu: bool) -> str:
+def get_docker_container_env_vars(suffix: str) -> dict:
+    """
+    Retrieves environment variables from "background" and "api_server" Docker containers.
+    """
+    print(f"Getting environment variables for containers with suffix: {suffix}")
+
+    combined_env_vars = {}
+    for container_type in ["background", "api_server"]:
+        container_name = _run_command(
+            f"docker ps -a --format '{{{{.Names}}}}' | awk '/{container_type}/ && /{suffix}/'"
+        )[0].strip()
+        if not container_name:
+            raise RuntimeError(
+                f"No {container_type} container found with suffix: {suffix}"
+            )
+
+        env_vars_json = _run_command(
+            f"docker inspect --format='{{{{json .Config.Env}}}}' {container_name}"
+        )[0]
+        env_vars_list = json.loads(env_vars_json.strip())
+
+        for env_var in env_vars_list:
+            key, value = env_var.split("=", 1)
+            combined_env_vars[key] = value
+
+    return combined_env_vars
+
+
+def manage_data_directories(suffix: str, base_path: str, use_cloud_gpu: bool) -> None:
     # Use the user's home directory as the base path
-    target_path = os.path.join(os.path.expanduser(base_path), f"test{suffix}")
+    target_path = os.path.join(os.path.expanduser(base_path), suffix)
     directories = {
         "DANSWER_POSTGRES_DATA_DIR": os.path.join(target_path, "postgres/"),
         "DANSWER_VESPA_DATA_DIR": os.path.join(target_path, "vespa/"),
@@ -85,9 +115,8 @@ def manage_data_directories(suffix: str, base_path: str, use_cloud_gpu: bool) ->
         os.makedirs(directory, exist_ok=True)
         os.environ[env_var] = directory
         print(f"Set {env_var} to: {directory}")
-    relari_output_path = os.path.join(target_path, "relari_output/")
-    os.makedirs(relari_output_path, exist_ok=True)
-    return relari_output_path
+    results_output_path = os.path.join(target_path, "evaluations_output/")
+    os.makedirs(results_output_path, exist_ok=True)
 
 
 def set_env_variables(
@@ -115,26 +144,30 @@ def _is_port_in_use(port: int) -> bool:
 
 
 def start_docker_compose(
-    run_suffix: str, launch_web_ui: bool, use_cloud_gpu: bool
+    run_suffix: str, launch_web_ui: bool, use_cloud_gpu: bool, only_state: bool = False
 ) -> None:
     print("Starting Docker Compose...")
     os.chdir(os.path.dirname(__file__))
     os.chdir("../../../../deployment/docker_compose/")
-    command = f"docker compose -f docker-compose.search-testing.yml -p danswer-stack{run_suffix} up -d"
+    command = f"docker compose -f docker-compose.search-testing.yml -p danswer-stack-{run_suffix} up -d"
     command += " --build"
     command += " --force-recreate"
-    if use_cloud_gpu:
-        command += " --scale indexing_model_server=0"
-        command += " --scale inference_model_server=0"
-    if launch_web_ui:
-        web_ui_port = 3000
-        while _is_port_in_use(web_ui_port):
-            web_ui_port += 1
-        print(f"UI will be launched at http://localhost:{web_ui_port}")
-        os.environ["NGINX_PORT"] = str(web_ui_port)
+
+    if only_state:
+        command += " index relational_db"
     else:
-        command += " --scale web_server=0"
-        command += " --scale nginx=0"
+        if use_cloud_gpu:
+            command += " --scale indexing_model_server=0"
+            command += " --scale inference_model_server=0"
+        if launch_web_ui:
+            web_ui_port = 3000
+            while _is_port_in_use(web_ui_port):
+                web_ui_port += 1
+            print(f"UI will be launched at http://localhost:{web_ui_port}")
+            os.environ["NGINX_PORT"] = str(web_ui_port)
+        else:
+            command += " --scale web_server=0"
+            command += " --scale nginx=0"
 
     print("Docker Command:\n", command)
 
@@ -150,7 +183,8 @@ def cleanup_docker(run_suffix: str) -> None:
     stdout, _ = _run_command("docker ps -a --format '{{json .}}'")
 
     containers = [json.loads(line) for line in stdout.splitlines()]
-
+    if not run_suffix:
+        run_suffix = datetime.now().strftime("-%Y")
     project_name = f"danswer-stack{run_suffix}"
     containers_to_delete = [
         c for c in containers if c["Names"].startswith(project_name)
@@ -247,3 +281,71 @@ def get_api_server_host_port(suffix: str) -> str:
             f"No port found containing: {client_port} for container: {container_name} and suffix: {suffix}"
         )
     return matching_ports[0]
+
+
+# Added function to check Vespa container health status
+def is_vespa_container_healthy(suffix: str) -> bool:
+    print(f"Checking health status of Vespa container for suffix: {suffix}")
+
+    # Find the Vespa container
+    stdout, _ = _run_command(
+        f"docker ps -a --format '{{{{.Names}}}}' | awk /vespa/ && /{suffix}/"
+    )
+    container_name = stdout.strip()
+
+    if not container_name:
+        print(f"No Vespa container found with suffix: {suffix}")
+        return False
+
+    # Get the health status
+    stdout, _ = _run_command(
+        f"docker inspect --format='{{{{.State.Health.Status}}}}' {container_name}"
+    )
+    health_status = stdout.strip()
+
+    is_healthy = health_status.lower() == "healthy"
+    print(f"Vespa container '{container_name}' health status: {health_status}")
+
+    return is_healthy
+
+
+# Added function to restart Vespa container
+def restart_vespa_container(suffix: str) -> None:
+    print(f"Restarting Vespa container for suffix: {suffix}")
+
+    # Find the Vespa container
+    stdout, _ = _run_command(
+        f"docker ps -a --format '{{{{.Names}}}}' | awk /vespa/ && /{suffix}/"
+    )
+    container_name = stdout.strip()
+
+    if not container_name:
+        raise RuntimeError(f"No Vespa container found with suffix: {suffix}")
+
+    # Restart the container
+    _run_command(f"docker restart {container_name}")
+
+    print(f"Vespa container '{container_name}' has begun restarting")
+
+    time_to_wait = 5
+    while not is_vespa_container_healthy(suffix):
+        print(f"Waiting {time_to_wait} seconds for vespa container to restart")
+        time.sleep(5)
+
+    print(f"Vespa container '{container_name}' has been restarted")
+
+
+if __name__ == "__main__":
+    """
+    Running this just cleans up the docker environment for the container indicated by existing_test_suffix
+    If no existing_test_suffix is indicated, will just clean up all danswer docker containers/volumes/networks
+    Note: vespa/postgres mounts are not deleted
+    """
+    current_dir = os.path.dirname(os.path.abspath(__file__))
+    config_path = os.path.join(current_dir, "search_test_config.yaml")
+    with open(config_path, "r") as file:
+        config = yaml.safe_load(file)
+
+    if not isinstance(config, dict):
+        raise TypeError("config must be a dictionary")
+    cleanup_docker(config["existing_test_suffix"])

backend/tests/regression/answer_quality/run_eval_pipeline.py

@@ -8,7 +8,7 @@ from tests.regression.answer_quality.cli_utils import cleanup_docker
 from tests.regression.answer_quality.cli_utils import manage_data_directories
 from tests.regression.answer_quality.cli_utils import set_env_variables
 from tests.regression.answer_quality.cli_utils import start_docker_compose
-from tests.regression.answer_quality.cli_utils import switch_to_branch
+from tests.regression.answer_quality.cli_utils import switch_to_commit
 from tests.regression.answer_quality.file_uploader import upload_test_files
 from tests.regression.answer_quality.run_qa import run_qa_test_and_save_results
 
@@ -26,7 +26,7 @@ def main() -> None:
         run_suffix = config.existing_test_suffix
         print("launching danswer with existing data suffix:", run_suffix)
     else:
-        run_suffix = datetime.now().strftime("_%Y%m%d_%H%M%S")
+        run_suffix = datetime.now().strftime("-%Y%m%d-%H%M%S")
         print("run_suffix:", run_suffix)
 
     set_env_variables(
@@ -35,20 +35,18 @@ def main() -> None:
         config.use_cloud_gpu,
         config.llm,
     )
-    relari_output_folder_path = manage_data_directories(
-        run_suffix, config.output_folder, config.use_cloud_gpu
+    manage_data_directories(run_suffix, config.output_folder, config.use_cloud_gpu)
+    if config.commit_sha:
+        switch_to_commit(config.commit_sha)
+
+    start_docker_compose(
+        run_suffix, config.launch_web_ui, config.use_cloud_gpu, config.only_state
     )
-    if config.branch:
-        switch_to_branch(config.branch)
 
-    start_docker_compose(run_suffix, config.launch_web_ui, config.use_cloud_gpu)
-
-    if not config.existing_test_suffix:
+    if not config.existing_test_suffix and not config.only_state:
         upload_test_files(config.zipped_documents_file, run_suffix)
 
-        run_qa_test_and_save_results(
-            config.questions_file, relari_output_folder_path, run_suffix, config.limit
-        )
+        run_qa_test_and_save_results(run_suffix)
 
         if config.clean_up_docker_containers:
             cleanup_docker(run_suffix)

backend/tests/regression/answer_quality/run_qa.py

@@ -1,55 +1,35 @@
 import json
+import multiprocessing
 import os
+import shutil
 import time
-from types import SimpleNamespace
 
 import yaml
 
 from tests.regression.answer_quality.api_utils import check_if_query_ready
 from tests.regression.answer_quality.api_utils import get_answer_from_query
 from tests.regression.answer_quality.cli_utils import get_current_commit_sha
+from tests.regression.answer_quality.cli_utils import get_docker_container_env_vars
+
+RESULTS_FILENAME = "results.jsonl"
+METADATA_FILENAME = "metadata.yaml"
 
 
-def _process_and_write_query_results(
-    samples: list[dict], run_suffix: str, output_file_path: str
-) -> None:
-    while not check_if_query_ready(run_suffix):
-        time.sleep(5)
-
-    count = 0
-    with open(output_file_path, "w", encoding="utf-8") as file:
-        for sample in samples:
-            print(f"On question number {count}")
-            query = sample["question"]
-            print(f"query: {query}")
-            context_data_list, answer = get_answer_from_query(
-                query=query,
-                run_suffix=run_suffix,
-            )
-
-            print(f"answer: {answer[:50]}...")
-            if not context_data_list:
-                print("No context found")
-            else:
-                print(f"{len(context_data_list)} context docs found")
-            print("\n")
-
-            output = {
-                "question_data": sample,
-                "answer": answer,
-                "context_data_list": context_data_list,
-            }
-
-            file.write(json.dumps(output) + "\n")
+def _populate_results_file(output_folder_path: str, all_qa_output: list[dict]) -> None:
+    output_file_path = os.path.join(output_folder_path, RESULTS_FILENAME)
+    with open(output_file_path, "a", encoding="utf-8") as file:
+        for qa_output in all_qa_output:
+            file.write(json.dumps(qa_output) + "\n")
             file.flush()
-            count += 1
 
 
-def _write_metadata_file(run_suffix: str, metadata_file_path: str) -> None:
-    metadata = {"commit_sha": get_current_commit_sha(), "run_suffix": run_suffix}
+def _update_metadata_file(test_output_folder: str, invalid_answer_count: int) -> None:
+    metadata_path = os.path.join(test_output_folder, METADATA_FILENAME)
+    with open(metadata_path, "r", encoding="utf-8") as file:
+        metadata = yaml.safe_load(file)
 
-    print("saving metadata to:", metadata_file_path)
-    with open(metadata_file_path, "w", encoding="utf-8") as yaml_file:
+    metadata["number_of_failed_questions"] = invalid_answer_count
+    with open(metadata_path, "w", encoding="utf-8") as yaml_file:
         yaml.dump(metadata, yaml_file)
 
 
@@ -62,60 +42,155 @@ def _read_questions_jsonl(questions_file_path: str) -> list[dict]:
     return questions
 
 
-def run_qa_test_and_save_results(
-    questions_file_path: str,
-    results_folder_path: str,
-    run_suffix: str,
-    limit: int | None = None,
-) -> None:
-    results_file = "run_results.jsonl"
-    metadata_file = "run_metadata.yaml"
-    samples = _read_questions_jsonl(questions_file_path)
-
-    if limit is not None:
-        samples = samples[:limit]
+def _get_test_output_folder(config: dict) -> str:
+    base_output_folder = os.path.expanduser(config["output_folder"])
+    if config["run_suffix"]:
+        base_output_folder = os.path.join(
+            base_output_folder, config["run_suffix"], "evaluations_output"
+        )
+    else:
+        base_output_folder = os.path.join(base_output_folder, "no_defined_suffix")
 
     counter = 1
-    output_file_path = os.path.join(results_folder_path, results_file)
-    metadata_file_path = os.path.join(results_folder_path, metadata_file)
-    while os.path.exists(output_file_path):
-        output_file_path = os.path.join(
-            results_folder_path,
-            results_file.replace("run_results", f"run_results_{counter}"),
-        )
-        metadata_file_path = os.path.join(
-            results_folder_path,
-            metadata_file.replace("run_metadata", f"run_metadata_{counter}"),
+    output_folder_path = os.path.join(base_output_folder, "run_1")
+    while os.path.exists(output_folder_path):
+        output_folder_path = os.path.join(
+            output_folder_path.replace(f"run_{counter-1}", f"run_{counter}"),
         )
         counter += 1
 
-    print("saving question results to:", output_file_path)
-    _write_metadata_file(run_suffix, metadata_file_path)
-    _process_and_write_query_results(
-        samples=samples, run_suffix=run_suffix, output_file_path=output_file_path
+    os.makedirs(output_folder_path, exist_ok=True)
+
+    return output_folder_path
+
+
+def _initialize_files(config: dict) -> tuple[str, list[dict]]:
+    test_output_folder = _get_test_output_folder(config)
+
+    questions_file_path = config["questions_file"]
+
+    questions = _read_questions_jsonl(questions_file_path)
+
+    metadata = {
+        "commit_sha": get_current_commit_sha(),
+        "run_suffix": config["run_suffix"],
+        "test_config": config,
+        "number_of_questions_in_dataset": len(questions),
+    }
+
+    env_vars = get_docker_container_env_vars(config["run_suffix"])
+    if env_vars["ENV_SEED_CONFIGURATION"]:
+        del env_vars["ENV_SEED_CONFIGURATION"]
+    if env_vars["GPG_KEY"]:
+        del env_vars["GPG_KEY"]
+    if metadata["test_config"]["llm"]["api_key"]:
+        del metadata["test_config"]["llm"]["api_key"]
+    metadata.update(env_vars)
+    metadata_path = os.path.join(test_output_folder, METADATA_FILENAME)
+    print("saving metadata to:", metadata_path)
+    with open(metadata_path, "w", encoding="utf-8") as yaml_file:
+        yaml.dump(metadata, yaml_file)
+
+    copied_questions_file_path = os.path.join(
+        test_output_folder, os.path.basename(questions_file_path)
+    )
+    shutil.copy2(questions_file_path, copied_questions_file_path)
+
+    zipped_files_path = config["zipped_documents_file"]
+    copied_zipped_documents_path = os.path.join(
+        test_output_folder, os.path.basename(zipped_files_path)
+    )
+    shutil.copy2(zipped_files_path, copied_zipped_documents_path)
+
+    zipped_files_folder = os.path.dirname(zipped_files_path)
+    jsonl_file_path = os.path.join(zipped_files_folder, "target_docs.jsonl")
+    if os.path.exists(jsonl_file_path):
+        copied_jsonl_path = os.path.join(test_output_folder, "target_docs.jsonl")
+        shutil.copy2(jsonl_file_path, copied_jsonl_path)
+
+    return test_output_folder, questions
+
+
+def _process_question(question_data: dict, config: dict, question_number: int) -> dict:
+    print(f"On question number {question_number}")
+
+    query = question_data["question"]
+    print(f"query: {query}")
+    context_data_list, answer = get_answer_from_query(
+        query=query,
+        run_suffix=config["run_suffix"],
     )
 
+    if not context_data_list:
+        print("No answer or context found")
+    else:
+        print(f"answer: {answer[:50]}...")
+        print(f"{len(context_data_list)} context docs found")
+    print("\n")
 
-def main() -> None:
+    output = {
+        "question_data": question_data,
+        "answer": answer,
+        "context_data_list": context_data_list,
+    }
+
+    return output
+
+
+def _process_and_write_query_results(config: dict) -> None:
+    start_time = time.time()
+    test_output_folder, questions = _initialize_files(config)
+    print("saving test results to folder:", test_output_folder)
+
+    while not check_if_query_ready(config["run_suffix"]):
+        time.sleep(5)
+
+    if config["limit"] is not None:
+        questions = questions[: config["limit"]]
+
+    with multiprocessing.Pool(processes=multiprocessing.cpu_count() * 2) as pool:
+        results = pool.starmap(
+            _process_question, [(q, config, i + 1) for i, q in enumerate(questions)]
+        )
+
+    _populate_results_file(test_output_folder, results)
+
+    invalid_answer_count = 0
+    for result in results:
+        if not result.get("answer"):
+            invalid_answer_count += 1
+
+        if not result.get("context_data_list"):
+            raise RuntimeError("Search failed, this is a critical failure!")
+
+    _update_metadata_file(test_output_folder, invalid_answer_count)
+
+    if invalid_answer_count:
+        print(f"Warning: {invalid_answer_count} questions failed!")
+        print("Suggest restarting the vespa container and rerunning")
+
+    time_to_finish = time.time() - start_time
+    minutes, seconds = divmod(int(time_to_finish), 60)
+    print(
+        f"Took {minutes:02d}:{seconds:02d} to ask and answer {len(results)} questions"
+    )
+    print("saved test results to folder:", test_output_folder)
+
+
+def run_qa_test_and_save_results(run_suffix: str = "") -> None:
     current_dir = os.path.dirname(os.path.abspath(__file__))
     config_path = os.path.join(current_dir, "search_test_config.yaml")
     with open(config_path, "r") as file:
-        config = SimpleNamespace(**yaml.safe_load(file))
+        config = yaml.safe_load(file)
 
-    current_output_folder = os.path.expanduser(config.output_folder)
-    if config.existing_test_suffix:
-        current_output_folder = os.path.join(
-            current_output_folder, "test" + config.existing_test_suffix, "relari_output"
-        )
-    else:
-        current_output_folder = os.path.join(current_output_folder, "no_defined_suffix")
+    if not isinstance(config, dict):
+        raise TypeError("config must be a dictionary")
 
-    run_qa_test_and_save_results(
-        config.questions_file,
-        current_output_folder,
-        config.existing_test_suffix,
-        config.limit,
-    )
+    if not run_suffix:
+        run_suffix = config["existing_test_suffix"]
+
+    config["run_suffix"] = run_suffix
+    _process_and_write_query_results(config)
 
 
 if __name__ == "__main__":
@@ -123,4 +198,4 @@ if __name__ == "__main__":
     To run a different set of questions, update the questions_file in search_test_config.yaml
     If there is more than one instance of Danswer running, specify the suffix in search_test_config.yaml
     """
-    main()
+    run_qa_test_and_save_results()

backend/tests/regression/answer_quality/search_test_config.yaml.template

@@ -10,8 +10,8 @@ zipped_documents_file: "~/sampledocs.zip"
 # Path to the YAML file containing sample questions
 questions_file: "~/sample_questions.yaml"
 
-# Git branch to use (null means use current branch as is)
-branch: null
+# Git commit SHA to use (null means use current code as is)
+commit_sha: null
 
 # Whether to remove Docker containers after the test
 clean_up_docker_containers: true
@@ -19,6 +19,9 @@ clean_up_docker_containers: true
 # Whether to launch a web UI for the test
 launch_web_ui: false
 
+# Whether to only run Vespa and Postgres
+only_state: false
+
 # Whether to use a cloud GPU for processing
 use_cloud_gpu: false
 
@@ -28,7 +31,8 @@ model_server_ip: "PUT_PUBLIC_CLOUD_IP_HERE"
 # Port of the model server (placeholder)
 model_server_port: "PUT_PUBLIC_CLOUD_PORT_HERE"
 
-# Suffix for existing test results (empty string means no suffix)
+# Suffix for existing test results (E.g. -1234-5678)
+# empty string means no suffix
 existing_test_suffix: ""
 
 # Limit on number of tests to run (null means no limit)

backend/throttle.ctrl

@@ -1 +1 @@
-f1f2 1 1718910083.03085 wikipedia:en
+f1f2 2 1721064549.902656 wikipedia:en

deployment/docker_compose/docker-compose.dev.yml

@@ -19,7 +19,7 @@ services:
     environment:
       # Auth Settings
       - AUTH_TYPE=${AUTH_TYPE:-disabled}
-      - SESSION_EXPIRE_TIME_SECONDS=${SESSION_EXPIRE_TIME_SECONDS:-86400}
+      - SESSION_EXPIRE_TIME_SECONDS=${SESSION_EXPIRE_TIME_SECONDS:-}
       - ENCRYPTION_KEY_SECRET=${ENCRYPTION_KEY_SECRET:-}
       - VALID_EMAIL_DOMAINS=${VALID_EMAIL_DOMAINS:-}
       - GOOGLE_OAUTH_CLIENT_ID=${GOOGLE_OAUTH_CLIENT_ID:-}
@@ -51,6 +51,7 @@ services:
       - DISABLE_LITELLM_STREAMING=${DISABLE_LITELLM_STREAMING:-}
       - LITELLM_EXTRA_HEADERS=${LITELLM_EXTRA_HEADERS:-}
       - BING_API_KEY=${BING_API_KEY:-}
+      - DISABLE_AGENTIC_SEARCH=${DISABLE_AGENTIC_SEARCH:-}
       # if set, allows for the use of the token budget system
       - TOKEN_BUDGET_GLOBALLY_ENABLED=${TOKEN_BUDGET_GLOBALLY_ENABLED:-}
       # Enables the use of bedrock models
@@ -230,6 +231,7 @@ services:
       - INTERNAL_URL=http://api_server:8080
       - WEB_DOMAIN=${WEB_DOMAIN:-}
       - THEME_IS_DARK=${THEME_IS_DARK:-}
+      - DISABLE_AGENTIC_SEARCH=${DISABLE_AGENTIC_SEARCH:-}
 
       # Enterprise Edition only
       - ENABLE_PAID_ENTERPRISE_EDITION_FEATURES=${ENABLE_PAID_ENTERPRISE_EDITION_FEATURES:-false}
@@ -282,7 +284,7 @@ services:
       - LOG_LEVEL=${LOG_LEVEL:-info}
     volumes:
       # Not necessary, this is just to reduce download time during startup
-      - model_cache_huggingface:/root/.cache/huggingface/
+      - indexing_huggingface_model_cache:/root/.cache/huggingface/
     logging:
       driver: json-file
       options:
@@ -353,3 +355,4 @@ volumes:
   vespa_volume:
   # Created by the container itself
   model_cache_huggingface:
+  indexing_huggingface_model_cache:

deployment/docker_compose/docker-compose.gpu-dev.yml

@@ -19,7 +19,7 @@ services:
     environment:
       # Auth Settings
       - AUTH_TYPE=${AUTH_TYPE:-disabled}
-      - SESSION_EXPIRE_TIME_SECONDS=${SESSION_EXPIRE_TIME_SECONDS:-86400}
+      - SESSION_EXPIRE_TIME_SECONDS=${SESSION_EXPIRE_TIME_SECONDS:-}
       - ENCRYPTION_KEY_SECRET=${ENCRYPTION_KEY_SECRET:-}
       - VALID_EMAIL_DOMAINS=${VALID_EMAIL_DOMAINS:-}
       - GOOGLE_OAUTH_CLIENT_ID=${GOOGLE_OAUTH_CLIENT_ID:-}
@@ -289,7 +289,7 @@ services:
       - LOG_LEVEL=${LOG_LEVEL:-info}
     volumes:
       # Not necessary, this is just to reduce download time during startup
-      - model_cache_huggingface:/root/.cache/huggingface/
+      - indexing_huggingface_model_cache:/root/.cache/huggingface/
     logging:
       driver: json-file
       options:
@@ -360,3 +360,4 @@ volumes:
   vespa_volume:
   # Created by the container itself
   model_cache_huggingface:
+  indexing_huggingface_model_cache:

deployment/docker_compose/docker-compose.prod-no-letsencrypt.yml

@@ -132,7 +132,7 @@ services:
       - LOG_LEVEL=${LOG_LEVEL:-info}
     volumes:
       # Not necessary, this is just to reduce download time during startup
-      - model_cache_huggingface:/root/.cache/huggingface/
+      - indexing_huggingface_model_cache:/root/.cache/huggingface/
     logging:
       driver: json-file
       options:
@@ -207,3 +207,4 @@ volumes:
   vespa_volume:
   # Created by the container itself
   model_cache_huggingface:
+  indexing_huggingface_model_cache:

deployment/docker_compose/docker-compose.prod.yml

@@ -146,7 +146,7 @@ services:
       - LOG_LEVEL=${LOG_LEVEL:-info}
     volumes:
       # Not necessary, this is just to reduce download time during startup
-      - model_cache_huggingface:/root/.cache/huggingface/
+      - indexing_huggingface_model_cache:/root/.cache/huggingface/
     logging:
       driver: json-file
       options:
@@ -225,3 +225,4 @@ volumes:
   vespa_volume:
   # Created by the container itself
   model_cache_huggingface:
+  indexing_huggingface_model_cache:

deployment/docker_compose/env.prod.template

@@ -57,8 +57,8 @@ SECRET=
 #SAML_CONF_DIR=
 
 
-# How long before user needs to reauthenticate, default to 1 day. (cookie expiration time)
-SESSION_EXPIRE_TIME_SECONDS=86400
+# How long before user needs to reauthenticate, default to 7 days. (cookie expiration time)
+SESSION_EXPIRE_TIME_SECONDS=604800
 
 
 # Use the below to specify a list of allowed user domains, only checked if user Auth is turned on

examples/widget/.eslintrc.json

@@ -0,0 +1,3 @@
+{
+  "extends": "next/core-web-vitals"
+}

examples/widget/.gitignore

@@ -0,0 +1,36 @@
+# See https://help.github.com/articles/ignoring-files/ for more about ignoring files.
+
+# dependencies
+/node_modules
+/.pnp
+.pnp.js
+.yarn/install-state.gz
+
+# testing
+/coverage
+
+# next.js
+/.next/
+/out/
+
+# production
+/build
+
+# misc
+.DS_Store
+*.pem
+
+# debug
+npm-debug.log*
+yarn-debug.log*
+yarn-error.log*
+
+# local env files
+.env*.local
+
+# vercel
+.vercel
+
+# typescript
+*.tsbuildinfo
+next-env.d.ts

examples/widget/README.md

@@ -0,0 +1,19 @@
+This is a code example for how you can use Danswer's APIs to build a chat bot widget for a website! The main code to look at can be found in `src/app/widget/Widget.tsx`.
+
+If you want to get fancier, then take a peek at the Chat implementation within Danswer itself [here](https://github.com/danswer-ai/danswer/blob/main/web/src/app/chat/ChatPage.tsx#L82).
+
+## Getting Started
+
+First, install the requirements:
+
+```bash
+npm i
+```
+
+Then run the development server:
+
+```bash
+npm run dev
+```
+
+Open [http://localhost:3000](http://localhost:3000) with your browser to see the result.

examples/widget/next.config.mjs

@@ -0,0 +1,4 @@
+/** @type {import('next').NextConfig} */
+const nextConfig = {};
+
+export default nextConfig;

examples/widget/package.json

@@ -0,0 +1,28 @@
+{
+  "name": "widget",
+  "version": "0.1.0",
+  "private": true,
+  "scripts": {
+    "dev": "next dev",
+    "build": "next build",
+    "start": "next start",
+    "lint": "next lint"
+  },
+  "dependencies": {
+    "next": "14.2.5",
+    "react": "^18",
+    "react-dom": "^18",
+    "react-markdown": "^8.0.6"
+  },
+  "devDependencies": {
+    "@types/node": "^20",
+    "@types/react": "^18",
+    "@types/react-dom": "^18",
+    "autoprefixer": "^10.4.19",
+    "eslint": "^8",
+    "eslint-config-next": "14.2.5",
+    "postcss": "^8.4.39",
+    "tailwindcss": "^3.4.6",
+    "typescript": "^5"
+  }
+}

examples/widget/postcss.config.mjs

@@ -0,0 +1,8 @@
+/** @type {import('postcss-load-config').Config} */
+const config = {
+  plugins: {
+    tailwindcss: {},
+  },
+};
+
+export default config;

examples/widget/src/app/globals.css

@@ -0,0 +1,3 @@
+@tailwind base;
+@tailwind components;
+@tailwind utilities;

examples/widget/src/app/layout.tsx

@@ -0,0 +1,23 @@
+import type { Metadata } from "next";
+import { Inter } from "next/font/google";
+
+import "./globals.css";
+
+const inter = Inter({ subsets: ["latin"] });
+
+export const metadata: Metadata = {
+  title: "Example Danswer Widget",
+  description: "Example Danswer Widget",
+};
+
+export default function RootLayout({
+  children,
+}: Readonly<{
+  children: React.ReactNode;
+}>) {
+  return (
+    <html lang="en">
+      <body className={inter.className}>{children}</body>
+    </html>
+  );
+}

examples/widget/src/app/page.tsx

@@ -0,0 +1,9 @@
+import { ChatWidget } from "./widget/Widget";
+
+export default function Home() {
+  return (
+    <main className="flex min-h-screen flex-col items-center justify-between p-24">
+      <ChatWidget />
+    </main>
+  );
+}

examples/widget/src/app/widget/Widget.tsx

@@ -0,0 +1,344 @@
+"use client";
+
+import React, { useState } from "react";
+import ReactMarkdown from "react-markdown";
+
+const API_URL = process.env.NEXT_PUBLIC_API_URL || "http://localhost:8080";
+const API_KEY = process.env.NEXT_PUBLIC_API_KEY || "";
+
+type NonEmptyObject = { [k: string]: any };
+
+const processSingleChunk = <T extends NonEmptyObject>(
+  chunk: string,
+  currPartialChunk: string | null,
+): [T | null, string | null] => {
+  const completeChunk = (currPartialChunk || "") + chunk;
+  try {
+    // every complete chunk should be valid JSON
+    const chunkJson = JSON.parse(completeChunk);
+    return [chunkJson, null];
+  } catch (err) {
+    // if it's not valid JSON, then it's probably an incomplete chunk
+    return [null, completeChunk];
+  }
+};
+
+const processRawChunkString = <T extends NonEmptyObject>(
+  rawChunkString: string,
+  previousPartialChunk: string | null,
+): [T[], string | null] => {
+  /* This is required because, in practice, we see that nginx does not send over
+  each chunk one at a time even with buffering turned off. Instead,
+  chunks are sometimes in batches or are sometimes incomplete */
+  if (!rawChunkString) {
+    return [[], null];
+  }
+  const chunkSections = rawChunkString
+    .split("\n")
+    .filter((chunk) => chunk.length > 0);
+  let parsedChunkSections: T[] = [];
+  let currPartialChunk = previousPartialChunk;
+  chunkSections.forEach((chunk) => {
+    const [processedChunk, partialChunk] = processSingleChunk<T>(
+      chunk,
+      currPartialChunk,
+    );
+    if (processedChunk) {
+      parsedChunkSections.push(processedChunk);
+      currPartialChunk = null;
+    } else {
+      currPartialChunk = partialChunk;
+    }
+  });
+
+  return [parsedChunkSections, currPartialChunk];
+};
+
+async function* handleStream<T extends NonEmptyObject>(
+  streamingResponse: Response,
+): AsyncGenerator<T[], void, unknown> {
+  const reader = streamingResponse.body?.getReader();
+  const decoder = new TextDecoder("utf-8");
+
+  let previousPartialChunk: string | null = null;
+  while (true) {
+    const rawChunk = await reader?.read();
+    if (!rawChunk) {
+      throw new Error("Unable to process chunk");
+    }
+    const { done, value } = rawChunk;
+    if (done) {
+      break;
+    }
+
+    const [completedChunks, partialChunk] = processRawChunkString<T>(
+      decoder.decode(value, { stream: true }),
+      previousPartialChunk,
+    );
+    if (!completedChunks.length && !partialChunk) {
+      break;
+    }
+    previousPartialChunk = partialChunk as string | null;
+
+    yield await Promise.resolve(completedChunks);
+  }
+}
+
+async function* sendMessage({
+  message,
+  chatSessionId,
+  parentMessageId,
+}: {
+  message: string;
+  chatSessionId?: number;
+  parentMessageId?: number;
+}) {
+  if (!chatSessionId || !parentMessageId) {
+    // Create a new chat session if one doesn't exist
+    const createSessionResponse = await fetch(
+      `${API_URL}/chat/create-chat-session`,
+      {
+        method: "POST",
+        headers: {
+          "Content-Type": "application/json",
+          Authorization: `Bearer ${API_KEY}`,
+        },
+        body: JSON.stringify({
+          // or specify an assistant you have defined
+          persona_id: 0,
+        }),
+      },
+    );
+
+    if (!createSessionResponse.ok) {
+      const errorJson = await createSessionResponse.json();
+      const errorMsg = errorJson.message || errorJson.detail || "";
+      throw Error(`Failed to create chat session - ${errorMsg}`);
+    }
+
+    const sessionData = await createSessionResponse.json();
+    chatSessionId = sessionData.chat_session_id;
+  }
+
+  const sendMessageResponse = await fetch(`${API_URL}/chat/send-message`, {
+    method: "POST",
+    headers: {
+      "Content-Type": "application/json",
+      Authorization: `Bearer ${API_KEY}`,
+    },
+    body: JSON.stringify({
+      chat_session_id: chatSessionId,
+      parent_message_id: parentMessageId,
+      message: message,
+      prompt_id: null,
+      search_doc_ids: null,
+      file_descriptors: [],
+      // checkout https://github.com/danswer-ai/danswer/blob/main/backend/danswer/search/models.py#L105 for
+      // all available options
+      retrieval_options: {
+        run_search: "always",
+        filters: null,
+      },
+      query_override: null,
+    }),
+  });
+  if (!sendMessageResponse.ok) {
+    const errorJson = await sendMessageResponse.json();
+    const errorMsg = errorJson.message || errorJson.detail || "";
+    throw Error(`Failed to send message - ${errorMsg}`);
+  }
+
+  yield* handleStream<NonEmptyObject>(sendMessageResponse);
+}
+
+export const ChatWidget = () => {
+  const [messages, setMessages] = useState<{ text: string; isUser: boolean }[]>(
+    [],
+  );
+  const [inputText, setInputText] = useState("");
+  const [isLoading, setIsLoading] = useState(false);
+
+  const handleSubmit = async (e: React.FormEvent) => {
+    e.preventDefault();
+    if (inputText.trim()) {
+      const initialPrevMessages = messages;
+      setMessages([...initialPrevMessages, { text: inputText, isUser: true }]);
+      setInputText("");
+      setIsLoading(true);
+
+      try {
+        const messageGenerator = sendMessage({
+          message: inputText,
+          chatSessionId: undefined,
+          parentMessageId: undefined,
+        });
+        let fullResponse = "";
+
+        for await (const chunks of messageGenerator) {
+          for (const chunk of chunks) {
+            if ("answer_piece" in chunk) {
+              fullResponse += chunk.answer_piece;
+              setMessages([
+                ...initialPrevMessages,
+                { text: inputText, isUser: true },
+                { text: fullResponse, isUser: false },
+              ]);
+            }
+          }
+        }
+      } catch (error) {
+        console.error("Error sending message:", error);
+        setMessages((prevMessages) => [
+          ...prevMessages,
+          { text: "An error occurred. Please try again.", isUser: false },
+        ]);
+      } finally {
+        setIsLoading(false);
+      }
+    }
+  };
+
+  return (
+    <div
+      className="
+      fixed
+      bottom-4
+      right-4
+      z-50
+      bg-white
+      rounded-lg
+      shadow-xl
+      w-96
+      h-[32rem]
+      flex
+      flex-col
+      overflow-hidden
+      transition-all
+      duration-300
+      ease-in-out
+    "
+    >
+      <div
+        className="
+        bg-gradient-to-r
+        from-blue-600
+        to-blue-800
+        text-white
+        p-4
+        font-bold
+        flex
+        justify-between
+        items-center
+      "
+      >
+        <span>Chat Support</span>
+      </div>
+      <div
+        className="
+        flex-grow
+        overflow-y-auto
+        p-4
+        space-y-4
+        bg-gray-50
+        border-b
+        border-gray-200
+      "
+      >
+        {messages.map((message, index) => (
+          <div
+            key={index}
+            className={`
+            flex
+            ${message.isUser ? "justify-end" : "justify-start"}
+          `}
+          >
+            <div
+              className={`
+              max-w-[75%]
+              p-3
+              rounded-lg
+              ${
+                message.isUser
+                  ? "bg-blue-500 text-white"
+                  : "bg-white text-black"
+              }
+              shadow
+            `}
+            >
+              <ReactMarkdown>{message.text}</ReactMarkdown>
+            </div>
+          </div>
+        ))}
+        {isLoading && (
+          <div className="flex justify-center">
+            <div className="animate-pulse flex space-x-2">
+              <div className="w-2 h-2 bg-gray-500 rounded-full"></div>
+              <div className="w-2 h-2 bg-gray-500 rounded-full"></div>
+              <div className="w-2 h-2 bg-gray-500 rounded-full"></div>
+            </div>
+          </div>
+        )}
+      </div>
+      <form
+        onSubmit={handleSubmit}
+        className="
+        p-4
+        bg-white
+        border-t
+        border-gray-200
+      "
+      >
+        <div className="relative">
+          <input
+            type="text"
+            value={inputText}
+            onChange={(e) => setInputText(e.target.value)}
+            placeholder="Type a message..."
+            className="
+              w-full
+              p-2
+              pr-10
+              border
+              border-gray-300
+              rounded-full
+              focus:outline-none
+              focus:ring-2
+              focus:ring-blue-500
+              focus:border-transparent
+            "
+            disabled={isLoading}
+          />
+          <button
+            type="submit"
+            disabled={isLoading}
+            className="
+              absolute
+              right-2
+              top-1/2
+              transform
+              -translate-y-1/2
+              text-blue-500
+              hover:text-blue-600
+              focus:outline-none
+            "
+          >
+            <svg
+              xmlns="http://www.w3.org/2000/svg"
+              className="h-6 w-6"
+              fill="none"
+              viewBox="0 0 24 24"
+              stroke="currentColor"
+            >
+              <path
+                strokeLinecap="round"
+                strokeLinejoin="round"
+                strokeWidth={2}
+                d="M12 19l9 2-9-18-9 18 9-2zm0 0v-8"
+              />
+            </svg>
+          </button>
+        </div>
+      </form>
+    </div>
+  );
+};

examples/widget/tailwind.config.ts

@@ -0,0 +1,20 @@
+import type { Config } from "tailwindcss";
+
+const config: Config = {
+  content: [
+    "./src/pages/**/*.{js,ts,jsx,tsx,mdx}",
+    "./src/components/**/*.{js,ts,jsx,tsx,mdx}",
+    "./src/app/**/*.{js,ts,jsx,tsx,mdx}",
+  ],
+  theme: {
+    extend: {
+      backgroundImage: {
+        "gradient-radial": "radial-gradient(var(--tw-gradient-stops))",
+        "gradient-conic":
+          "conic-gradient(from 180deg at 50% 50%, var(--tw-gradient-stops))",
+      },
+    },
+  },
+  plugins: [],
+};
+export default config;

examples/widget/tsconfig.json

@@ -0,0 +1,26 @@
+{
+  "compilerOptions": {
+    "lib": ["dom", "dom.iterable", "esnext"],
+    "allowJs": true,
+    "skipLibCheck": true,
+    "strict": true,
+    "noEmit": true,
+    "esModuleInterop": true,
+    "module": "esnext",
+    "moduleResolution": "bundler",
+    "resolveJsonModule": true,
+    "isolatedModules": true,
+    "jsx": "preserve",
+    "incremental": true,
+    "plugins": [
+      {
+        "name": "next"
+      }
+    ],
+    "paths": {
+      "@/*": ["./src/*"]
+    }
+  },
+  "include": ["next-env.d.ts", "**/*.ts", "**/*.tsx", ".next/types/**/*.ts"],
+  "exclude": ["node_modules"]
+}

web/src/app/admin/assistants/AssistantEditor.tsx

@@ -400,7 +400,7 @@ export function AssistantEditor({
           }
 
           return (
-            <Form>
+            <Form className="w-full">
               <div className="pb-6">
                 <TextFormField
                   name="name"
@@ -550,7 +550,7 @@ export function AssistantEditor({
 
                         {searchToolEnabled() && (
                           <CollapsibleSection prompt="Configure Search">
-                            <div className=" ">
+                            <div>
                               {ccPairs.length > 0 && (
                                 <>
                                   <Label small>Document Sets</Label>

web/src/app/admin/assistants/CollapsibleSection.tsx

@@ -39,8 +39,10 @@ const CollapsibleSection: React.FC<CollapsibleSectionProps> = ({
         `}
         onClick={toggleCollapse}
       >
+        {" "}
+        Great and also a
         {isCollapsed ? (
-          <span className="collapse-toggle text-lg absolute left-0 top-0  text-sm flex  items-center gap-x-3 cursor-pointer">
+          <span className="collapse-toggle text-lg absolute left-0 top-0 text-sm flex items-center gap-x-3 cursor-pointer">
             <FiSettings className="pointer-events-none my-auto" size={16} />
             {prompt}{" "}
           </span>

web/src/app/admin/assistants/new/page.tsx

@@ -28,7 +28,7 @@ export default async function Page() {
   }
 
   return (
-    <div>
+    <div className="w-full">
       <BackButton />
       <AdminPageTitle
         title="Create a New Assistant"

web/src/app/admin/assistants/page.tsx

@@ -5,7 +5,7 @@ import { Divider, Text, Title } from "@tremor/react";
 import { fetchSS } from "@/lib/utilsSS";
 import { ErrorCallout } from "@/components/ErrorCallout";
 import { Persona } from "./interfaces";
-import { RobotIcon } from "@/components/icons/icons";
+import { AssistantsIcon, RobotIcon } from "@/components/icons/icons";
 import { AdminPageTitle } from "@/components/admin/Title";
 
 export default async function Page() {
@@ -24,7 +24,7 @@ export default async function Page() {
 
   return (
     <div className="mx-auto container">
-      <AdminPageTitle icon={<RobotIcon size={32} />} title="Assistants" />
+      <AdminPageTitle icon={<AssistantsIcon size={32} />} title="Assistants" />
 
       <Text className="mb-2">
         Assistants are a way to build custom search/question-answering

web/src/app/admin/bot/page.tsx

@@ -301,7 +301,7 @@ const Page = () => {
   return (
     <div className="container mx-auto">
       <AdminPageTitle
-        icon={<FiSlack size={32} />}
+        icon={<SlackIcon size={32} />}
         title="Slack Bot Configuration"
       />
       <InstantSSRAutoRefresh />

web/src/app/admin/connectors/gmail/Credential.tsx

@@ -34,7 +34,7 @@ const DriveJsonUpload = ({
       <input
         className={
           "mr-3 text-sm text-gray-900 border border-gray-300 rounded-lg " +
-          "cursor-pointer bg-gray-50 dark:text-gray-400 focus:outline-none " +
+          "cursor-pointer bg-background dark:text-gray-400 focus:outline-none " +
           "dark:bg-gray-700 dark:border-gray-600 dark:placeholder-gray-400"
         }
         type="file"

web/src/app/admin/connectors/google-drive/Credential.tsx

@@ -34,7 +34,7 @@ const DriveJsonUpload = ({
       <input
         className={
           "mr-3 text-sm text-gray-900 border border-gray-300 rounded-lg " +
-          "cursor-pointer bg-gray-50 dark:text-gray-400 focus:outline-none " +
+          "cursor-pointer bg-background dark:text-gray-400 focus:outline-none " +
           "dark:bg-gray-700 dark:border-gray-600 dark:placeholder-gray-400"
         }
         type="file"

web/src/app/admin/models/embedding/CloudEmbeddingPage.tsx

@@ -116,7 +116,13 @@ export default function CloudEmbeddingPage({
                   <div
                     key={index}
                     className={`p-3 my-2 border-2 border-neutral-300 border-opacity-40 rounded-md rounded cursor-pointer  
-                    ${!provider.configured ? "opacity-80 hover:opacity-100" : enabled ? "bg-background-stronger" : "hover:bg-background-strong"}`}
+                    ${
+                      !provider.configured
+                        ? "opacity-80 hover:opacity-100"
+                        : enabled
+                          ? "bg-background-stronger"
+                          : "hover:bg-background-strong"
+                    }`}
                     onClick={() => {
                       if (enabled) {
                         setAlreadySelectedModel(model);

web/src/app/admin/models/embedding/page.tsx

@@ -39,6 +39,7 @@ export interface EmbeddingDetails {
   default_model_id?: number;
   name: string;
 }
+import { EmbeddingIcon, PackageIcon } from "@/components/icons/icons";
 
 function Main() {
   const [openToggle, setOpenToggle] = useState(true);
@@ -364,17 +365,25 @@ function Main() {
             monitor the progress of the re-indexing on this page.
           </Text>
 
-          <div className="mt-8 text-sm  mr-auto  mb-12 divide-x-2  flex   ">
+          <div className="mt-8 text-sm mr-auto mb-12 divide-x-2 flex">
             <button
               onClick={() => setOpenToggle(true)}
-              className={` mx-2 p-2 font-bold  ${openToggle ? "rounded bg-neutral-900 text-neutral-100 underline" : "hover:underline"}`}
+              className={` mx-2 p-2 font-bold  ${
+                openToggle
+                  ? "rounded bg-neutral-900 text-neutral-100 underline"
+                  : "hover:underline"
+              }`}
             >
               Self-hosted
             </button>
             <div className="px-2 ">
               <button
                 onClick={() => setOpenToggle(false)}
-                className={`mx-2 p-2 font-bold  ${!openToggle ? "rounded bg-neutral-900   text-neutral-100 underline" : " hover:underline"}`}
+                className={`mx-2 p-2 font-bold  ${
+                  !openToggle
+                    ? "rounded bg-neutral-900   text-neutral-100 underline"
+                    : " hover:underline"
+                }`}
               >
                 Cloud-based
               </button>
@@ -516,7 +525,7 @@ function Page() {
     <div className="mx-auto container">
       <AdminPageTitle
         title="Embedding"
-        icon={<FiPackage size={32} className="my-auto" />}
+        icon={<EmbeddingIcon size={32} className="my-auto" />}
       />
 
       <Main />

web/src/app/admin/models/llm/page.tsx

@@ -3,13 +3,14 @@
 import { AdminPageTitle } from "@/components/admin/Title";
 import { FiCpu } from "react-icons/fi";
 import { LLMConfiguration } from "./LLMConfiguration";
+import { CpuIcon } from "@/components/icons/icons";
 
 const Page = () => {
   return (
     <div className="mx-auto container">
       <AdminPageTitle
         title="LLM Setup"
-        icon={<FiCpu size={32} className="my-auto" />}
+        icon={<CpuIcon size={32} className="my-auto" />}
       />
 
       <LLMConfiguration />

web/src/app/admin/settings/page.tsx

@@ -2,13 +2,14 @@ import { AdminPageTitle } from "@/components/admin/Title";
 import { FiSettings } from "react-icons/fi";
 import { SettingsForm } from "./SettingsForm";
 import { Text } from "@tremor/react";
+import { SettingsIcon } from "@/components/icons/icons";
 
 export default async function Page() {
   return (
     <div className="mx-auto container">
       <AdminPageTitle
         title="Workspace Settings"
-        icon={<FiSettings size={32} className="my-auto" />}
+        icon={<SettingsIcon size={32} className="my-auto" />}
       />
 
       <Text className="mb-8">

web/src/app/admin/token-rate-limits/page.tsx

@@ -23,6 +23,7 @@ import { mutate } from "swr";
 import { usePopup } from "@/components/admin/connectors/Popup";
 import { CreateRateLimitModal } from "./CreateRateLimitModal";
 import { usePaidEnterpriseFeaturesEnabled } from "@/components/settings/usePaidEnterpriseFeaturesEnabled";
+import { ShieldIcon } from "@/components/icons/icons";
 
 const BASE_URL = "/api/admin/token-rate-limits";
 const GLOBAL_TOKEN_FETCH_URL = `${BASE_URL}/global`;
@@ -219,8 +220,10 @@ function Main() {
 export default function Page() {
   return (
     <div className="mx-auto container">
-      <AdminPageTitle title="Token Rate Limits" icon={<FiShield size={32} />} />
-
+      <AdminPageTitle
+        title="Token Rate Limits"
+        icon={<ShieldIcon size={32} />}
+      />
       <Main />
     </div>
   );

web/src/app/admin/tools/edit/[toolId]/page.tsx

@@ -6,6 +6,7 @@ import { DeleteToolButton } from "./DeleteToolButton";
 import { FiTool } from "react-icons/fi";
 import { AdminPageTitle } from "@/components/admin/Title";
 import { BackButton } from "@/components/BackButton";
+import { ToolIcon } from "@/components/icons/icons";
 
 export default async function Page({ params }: { params: { toolId: string } }) {
   const tool = await fetchToolByIdSS(params.toolId);
@@ -46,7 +47,7 @@ export default async function Page({ params }: { params: { toolId: string } }) {
 
       <AdminPageTitle
         title="Edit Tool"
-        icon={<FiTool size={32} className="my-auto" />}
+        icon={<ToolIcon size={32} className="my-auto" />}
       />
 
       {body}

web/src/app/admin/tools/new/page.tsx

@@ -3,6 +3,7 @@
 import { ToolEditor } from "@/app/admin/tools/ToolEditor";
 import { BackButton } from "@/components/BackButton";
 import { AdminPageTitle } from "@/components/admin/Title";
+import { ToolIcon } from "@/components/icons/icons";
 import { Card } from "@tremor/react";
 import { FiTool } from "react-icons/fi";
 
@@ -13,7 +14,7 @@ export default function NewToolPage() {
 
       <AdminPageTitle
         title="Create Tool"
-        icon={<FiTool size={32} className="my-auto" />}
+        icon={<ToolIcon size={32} className="my-auto" />}
       />
 
       <Card>

web/src/app/admin/tools/page.tsx

@@ -6,6 +6,7 @@ import { Divider, Text, Title } from "@tremor/react";
 import { fetchSS } from "@/lib/utilsSS";
 import { ErrorCallout } from "@/components/ErrorCallout";
 import { AdminPageTitle } from "@/components/admin/Title";
+import { ToolIcon } from "@/components/icons/icons";
 
 export default async function Page() {
   const toolResponse = await fetchSS("/tool");
@@ -24,7 +25,7 @@ export default async function Page() {
   return (
     <div className="mx-auto container">
       <AdminPageTitle
-        icon={<FiTool size={32} className="my-auto" />}
+        icon={<ToolIcon size={32} className="my-auto" />}
         title="Tools"
       />
 

web/src/app/assistants/SidebarWrapper.tsx

@@ -0,0 +1,149 @@
+"use client";
+
+import { HistorySidebar } from "@/app/chat/sessionSidebar/HistorySidebar";
+import { AssistantsGallery } from "./gallery/AssistantsGallery";
+import FixedLogo from "@/app/chat/shared_chat_search/FixedLogo";
+import { UserDropdown } from "@/components/UserDropdown";
+import { ChatSession } from "@/app/chat/interfaces";
+import { Folder } from "@/app/chat/folders/interfaces";
+import { User } from "@/lib/types";
+import { Persona } from "@/app/admin/assistants/interfaces";
+import Cookies from "js-cookie";
+import { SIDEBAR_TOGGLED_COOKIE_NAME } from "@/components/resizable/constants";
+import { ReactNode, useEffect, useRef, useState } from "react";
+import { useSidebarVisibility } from "@/components/chat_search/hooks";
+import FunctionalHeader from "@/components/chat_search/Header";
+import { useRouter } from "next/navigation";
+
+interface SidebarWrapperProps<T extends object> {
+  chatSessions: ChatSession[];
+  folders: Folder[];
+  initiallyToggled: boolean;
+  openedFolders?: { [key: number]: boolean };
+  content: (props: T) => ReactNode;
+  headerProps: {
+    page: string;
+    user: User | null;
+  };
+  contentProps: T;
+}
+
+export default function SidebarWrapper<T extends object>({
+  chatSessions,
+  initiallyToggled,
+  folders,
+  openedFolders,
+  headerProps,
+  contentProps,
+  content,
+}: SidebarWrapperProps<T>) {
+  const [toggledSidebar, setToggledSidebar] = useState(initiallyToggled);
+
+  const [showDocSidebar, setShowDocSidebar] = useState(false); // State to track if sidebar is open
+
+  const toggleSidebar = () => {
+    Cookies.set(
+      SIDEBAR_TOGGLED_COOKIE_NAME,
+      String(!toggledSidebar).toLocaleLowerCase()
+    ),
+      {
+        path: "/",
+      };
+    setToggledSidebar((toggledSidebar) => !toggledSidebar);
+  };
+
+  const sidebarElementRef = useRef<HTMLDivElement>(null);
+
+  useSidebarVisibility({
+    toggledSidebar,
+    sidebarElementRef,
+    showDocSidebar,
+    setShowDocSidebar,
+  });
+
+  const innerSidebarElementRef = useRef<HTMLDivElement>(null);
+  const router = useRouter();
+  useEffect(() => {
+    const handleKeyDown = (event: KeyboardEvent) => {
+      if (event.metaKey || event.ctrlKey) {
+        switch (event.key.toLowerCase()) {
+          case "e":
+            event.preventDefault();
+            toggleSidebar();
+            break;
+        }
+      }
+    };
+
+    window.addEventListener("keydown", handleKeyDown);
+    return () => {
+      window.removeEventListener("keydown", handleKeyDown);
+    };
+  }, [router]);
+
+  return (
+    <div className="flex relative overflow-x-hidden overscroll-contain flex-col w-full h-screen">
+      <div
+        ref={sidebarElementRef}
+        className={`
+            flex-none
+            fixed
+            left-0
+            z-20
+            overflow-y-hidden
+            sidebar
+            bg-background-100
+            h-screen
+            transition-all
+            bg-opacity-80
+            duration-300
+            ease-in-out
+            ${
+              showDocSidebar || toggledSidebar
+                ? "opacity-100 w-[300px] translate-x-0"
+                : "opacity-0 w-[200px] pointer-events-none -translate-x-10"
+            }`}
+      >
+        <div className="w-full relative">
+          <HistorySidebar
+            page="chat"
+            ref={innerSidebarElementRef}
+            toggleSidebar={toggleSidebar}
+            toggled={toggledSidebar}
+            existingChats={chatSessions}
+            currentChatSession={null}
+            folders={folders}
+            openedFolders={openedFolders}
+          />
+        </div>
+      </div>
+
+      <div className="absolute left-0 w-full top-0">
+        <FunctionalHeader
+          page="assistants"
+          showSidebar={showDocSidebar}
+          user={headerProps.user}
+        />
+        <div className="w-full flex">
+          <div
+            style={{ transition: "width 0.30s ease-out" }}
+            className={`
+                    flex-none
+                    overflow-y-hidden
+                    bg-background-100
+                    h-full
+                    transition-all
+                    bg-opacity-80
+                    duration-300 
+                    ease-in-out
+                    ${toggledSidebar ? "w-[300px]" : "w-[0px]"}
+                  `}
+          />
+
+          <div className="mt-4 mx-auto">{content(contentProps)}</div>
+        </div>
+      </div>
+      <FixedLogo />
+    </div>
+  );
+}

web/src/app/assistants/ToolsDisplay.tsx

@@ -5,6 +5,7 @@ import { FiImage, FiSearch, FiGlobe } from "react-icons/fi";
 export function ToolsDisplay({ tools }: { tools: ToolSnapshot[] }) {
   return (
     <div className="text-xs text-subtle flex flex-wrap gap-1 mt-2">
+      <p className="text-sm text-default my-auto">Tools:</p>
       {tools.map((tool) => {
         let toolName = tool.name;
         let toolIcon = null;

web/src/app/assistants/gallery/WrappedAssistantsGallery.tsx

@@ -0,0 +1,44 @@
+"use client";
+
+import SidebarWrapper from "../SidebarWrapper";
+import { ChatSession } from "@/app/chat/interfaces";
+import { Folder } from "@/app/chat/folders/interfaces";
+import { Persona } from "@/app/admin/assistants/interfaces";
+import { User } from "@/lib/types";
+import { AssistantsGallery } from "./AssistantsGallery";
+
+export default function WrappedAssistantsGallery({
+  chatSessions,
+  initiallyToggled,
+  folders,
+  openedFolders,
+  user,
+  assistants,
+}: {
+  chatSessions: ChatSession[];
+  folders: Folder[];
+  initiallyToggled: boolean;
+  openedFolders?: { [key: number]: boolean };
+  user: User | null;
+  assistants: Persona[];
+}) {
+  return (
+    <SidebarWrapper
+      initiallyToggled={initiallyToggled}
+      chatSessions={chatSessions}
+      folders={folders}
+      openedFolders={openedFolders}
+      headerProps={{ user, page: "" }}
+      contentProps={{
+        assistants: assistants,
+        user: user,
+      }}
+      content={(contentProps) => (
+        <AssistantsGallery
+          assistants={contentProps.assistants}
+          user={contentProps.user}
+        />
+      )}
+    />
+  );
+}

web/src/app/assistants/gallery/page.tsx

@@ -1,4 +1,4 @@
-import { ChatSidebar } from "@/app/chat/sessionSidebar/ChatSidebar";
+import { HistorySidebar } from "@/app/chat/sessionSidebar/HistorySidebar";
 import { InstantSSRAutoRefresh } from "@/components/SSRAutoRefresh";
 import { UserDropdown } from "@/components/UserDropdown";
 import { ChatProvider } from "@/components/context/ChatContext";
@@ -7,6 +7,9 @@ import { fetchChatData } from "@/lib/chat/fetchChatData";
 import { unstable_noStore as noStore } from "next/cache";
 import { redirect } from "next/navigation";
 import { AssistantsGallery } from "./AssistantsGallery";
+import FixedLogo from "@/app/chat/shared_chat_search/FixedLogo";
+import GalleryWrapper from "../SidebarWrapper";
+import WrappedAssistantsGallery from "./WrappedAssistantsGallery";
 
 export default async function GalleryPage({
   searchParams,
@@ -32,6 +35,7 @@ export default async function GalleryPage({
     folders,
     openedFolders,
     shouldShowWelcomeModal,
+    toggleSidebar,
   } = data;
 
   return (
@@ -53,28 +57,14 @@ export default async function GalleryPage({
           openedFolders,
         }}
       >
-        <div className="flex relative bg-background text-default overflow-x-hidden h-screen">
-          <ChatSidebar
-            existingChats={chatSessions}
-            currentChatSession={null}
-            folders={folders}
-            openedFolders={openedFolders}
-          />
-
-          <div
-            className={`w-full h-full flex flex-col overflow-y-auto overflow-x-hidden relative`}
-          >
-            <div className="sticky top-0 left-80 z-10 w-full bg-background flex h-fit">
-              <div className="ml-auto my-auto mt-4 mr-8">
-                <UserDropdown user={user} />
-              </div>
-            </div>
-
-            <div className="mt-4">
-              <AssistantsGallery assistants={assistants} user={user} />
-            </div>
-          </div>
-        </div>
+        <WrappedAssistantsGallery
+          initiallyToggled={toggleSidebar}
+          chatSessions={chatSessions}
+          folders={folders}
+          openedFolders={openedFolders}
+          user={user}
+          assistants={assistants}
+        />
       </ChatProvider>
     </>
   );

web/src/app/assistants/mine/WrappedAssistantsMine.tsx

@@ -0,0 +1,43 @@
+"use client";
+import { AssistantsList } from "./AssistantsList";
+import SidebarWrapper from "../SidebarWrapper";
+import { ChatSession } from "@/app/chat/interfaces";
+import { Folder } from "@/app/chat/folders/interfaces";
+import { Persona } from "@/app/admin/assistants/interfaces";
+import { User } from "@/lib/types";
+
+export default function WrappedAssistantsMine({
+  chatSessions,
+  initiallyToggled,
+  folders,
+  openedFolders,
+  user,
+  assistants,
+}: {
+  chatSessions: ChatSession[];
+  folders: Folder[];
+  initiallyToggled: boolean;
+  openedFolders?: { [key: number]: boolean };
+  user: User | null;
+  assistants: Persona[];
+}) {
+  return (
+    <SidebarWrapper
+      initiallyToggled={initiallyToggled}
+      chatSessions={chatSessions}
+      folders={folders}
+      openedFolders={openedFolders}
+      headerProps={{ user, page: "" }}
+      contentProps={{
+        assistants: assistants,
+        user: user,
+      }}
+      content={(contentProps) => (
+        <AssistantsList
+          assistants={contentProps.assistants}
+          user={contentProps.user}
+        />
+      )}
+    />
+  );
+}

web/src/app/assistants/mine/page.tsx

@@ -1,4 +1,4 @@
-import { ChatSidebar } from "@/app/chat/sessionSidebar/ChatSidebar";
+import { HistorySidebar } from "@/app/chat/sessionSidebar/HistorySidebar";
 import { InstantSSRAutoRefresh } from "@/components/SSRAutoRefresh";
 import { UserDropdown } from "@/components/UserDropdown";
 import { ChatProvider } from "@/components/context/ChatContext";
@@ -8,6 +8,10 @@ import { fetchChatData } from "@/lib/chat/fetchChatData";
 import { unstable_noStore as noStore } from "next/cache";
 import { redirect } from "next/navigation";
 import { AssistantsList } from "./AssistantsList";
+import { Logo } from "@/components/Logo";
+import FixedLogo from "@/app/chat/shared_chat_search/FixedLogo";
+import SidebarWrapper from "../SidebarWrapper";
+import WrappedAssistantsMine from "./WrappedAssistantsMine";
 
 export default async function GalleryPage({
   searchParams,
@@ -33,6 +37,7 @@ export default async function GalleryPage({
     folders,
     openedFolders,
     shouldShowWelcomeModal,
+    toggleSidebar,
   } = data;
 
   return (
@@ -54,28 +59,14 @@ export default async function GalleryPage({
           openedFolders,
         }}
       >
-        <div className="flex relative bg-background text-default overflow-x-hidden h-screen">
-          <ChatSidebar
-            existingChats={chatSessions}
-            currentChatSession={null}
-            folders={folders}
-            openedFolders={openedFolders}
-          />
-
-          <div
-            className={`w-full h-screen flex flex-col overflow-y-auto overflow-x-hidden relative`}
-          >
-            <div className="sticky top-0 left-80 z-10 w-full bg-background flex h-fit">
-              <div className="ml-auto my-auto mt-4 mr-8">
-                <UserDropdown user={user} />
-              </div>
-            </div>
-
-            <div className="mt-4">
-              <AssistantsList user={user} assistants={assistants} />
-            </div>
-          </div>
-        </div>
+        <WrappedAssistantsMine
+          initiallyToggled={toggleSidebar}
+          chatSessions={chatSessions}
+          folders={folders}
+          openedFolders={openedFolders}
+          user={user}
+          assistants={assistants}
+        />
       </ChatProvider>
     </>
   );

web/src/app/auth/login/SignInButton.tsx

@@ -42,7 +42,7 @@ export function SignInButton({
 
   return (
     <a
-      className="mt-6 py-3 w-72 text-gray-100 bg-accent flex rounded cursor-pointer hover:bg-indigo-800"
+      className="mt-6 py-3 w-72 text-text-100 bg-accent flex rounded cursor-pointer hover:bg-indigo-800"
       href={authorizeUrl}
     >
       {button}

web/src/app/chat/ChatBanner.tsx

@@ -16,7 +16,7 @@ export function ChatBanner() {
       className={`
       z-[39] 
       h-[30px]
-      bg-background-custom-header
+      bg-background-100
       shadow-sm
       m-2
       rounded

web/src/app/chat/ChatIntro.tsx

@@ -35,17 +35,13 @@ export function ChatIntro({
 }) {
   const availableSourceMetadata = getSourceMetadataForSources(availableSources);
 
-  const [displaySources, setDisplaySources] = useState(false);
-
   return (
     <>
       <div className="flex justify-center items-center h-full">
         <div className="w-message-xs 2xl:w-message-sm 3xl:w-message">
           <div className="flex">
             <div className="mx-auto">
-              <Logo height={80} width={80} className="m-auto" />
-
-              <div className="m-auto text-3xl font-bold text-strong mt-4 w-fit">
+              <div className="m-auto text-3xl font-strong font-bold text-strong w-fit">
                 {selectedPersona?.name || "How can I help you today?"}
               </div>
               {selectedPersona && (

web/src/app/chat/WrappedChat.tsx

@@ -0,0 +1,26 @@
+"use client";
+
+import { useState } from "react";
+import { ChatPage } from "./ChatPage";
+import FunctionalWrapper from "./shared_chat_search/FunctionalWrapper";
+
+export default function WrappedChat({
+  defaultPersonaId,
+  initiallyToggled,
+}: {
+  defaultPersonaId?: number;
+  initiallyToggled: boolean;
+}) {
+  return (
+    <FunctionalWrapper
+      initiallyToggled={initiallyToggled}
+      content={(toggledSidebar, toggle) => (
+        <ChatPage
+          toggle={toggle}
+          defaultSelectedPersonaId={defaultPersonaId}
+          toggledSidebar={toggledSidebar}
+        />
+      )}
+    />
+  );
+}

web/src/app/chat/documentSidebar/ChatDocumentDisplay.tsx

@@ -1,8 +1,6 @@
 import { HoverPopup } from "@/components/HoverPopup";
 import { SourceIcon } from "@/components/SourceIcon";
 import { PopupSpec } from "@/components/admin/connectors/Popup";
-import { DocumentFeedbackBlock } from "@/components/search/DocumentFeedbackBlock";
-import { DocumentUpdatedAtBadge } from "@/components/search/DocumentUpdatedAtBadge";
 import { DanswerDocument } from "@/lib/search/interfaces";
 import { FiInfo, FiRadio } from "react-icons/fi";
 import { DocumentSelector } from "./DocumentSelector";
@@ -32,25 +30,32 @@ export function ChatDocumentDisplay({
   tokenLimitReached,
 }: DocumentDisplayProps) {
   const isInternet = document.is_internet;
+  // Consider reintroducing null scored docs in the future
+
+  if (document.score === null) {
+    return null;
+  }
 
   return (
-    <div key={document.semantic_identifier} className="text-sm px-3">
-      <div className="flex relative w-full overflow-y-visible">
+    <div
+      key={document.semantic_identifier}
+      className={`p-2 w-[350px] justify-start rounded-md ${
+        isSelected ? "bg-background-200" : "bg-background-125"
+      } text-sm mx-3`}
+    >
+      <div className="flex relative justify-start overflow-y-visible">
         <a
           className={
-            "rounded-lg flex font-bold flex-shrink truncate items-center " +
+            "rounded-lg flex font-bold flex-shrink truncate" +
             (document.link ? "" : "pointer-events-none")
           }
-          href={document.link}
-          target="_blank"
-          rel="noopener noreferrer"
         >
           {isInternet ? (
             <InternetSearchIcon url={document.link} />
           ) : (
             <SourceIcon sourceType={document.source_type} iconSize={18} />
           )}
-          <p className="overflow-hidden text-ellipsis mx-2 my-auto text-sm ">
+          <p className="overflow-hidden text-left text-ellipsis mx-2 my-auto text-sm ">
             {document.semantic_identifier || document.document_id}
           </p>
         </a>
@@ -75,6 +80,21 @@ export function ChatDocumentDisplay({
                 />
               </div>
             )}
+            <div
+              className={`
+                text-xs
+                text-emphasis
+                bg-hover
+                rounded
+                p-0.5
+                w-fit
+                my-auto
+                select-none
+                my-auto
+                mr-2`}
+            >
+              {Math.abs(document.score).toFixed(2)}
+            </div>
           </div>
         )}
 
@@ -91,8 +111,9 @@ export function ChatDocumentDisplay({
           <DocumentMetadataBlock document={document} />
         </div>
       </div>
-      <p className="pl-1 pt-2 pb-1 break-words">
+      <p className="line-clamp-3 pl-1 pt-2 mb-1 text-start break-words">
         {buildDocumentSummaryDisplay(document.match_highlights, document.blurb)}
+        test
       </p>
       <div className="mb-2">
         {/* 

web/src/app/chat/documentSidebar/DocumentSelector.tsx

@@ -32,9 +32,8 @@ export function DocumentSelector({
         }
         onClick={onClick}
       >
-        <p className="mr-2 my-auto">Select</p>
         <input
-          className="my-auto"
+          className="cursor-pointer my-auto"
           type="checkbox"
           checked={isSelected}
           // dummy function to prevent warning

web/src/app/chat/documentSidebar/DocumentSidebar.tsx

@@ -1,5 +1,5 @@
 import { DanswerDocument } from "@/lib/search/interfaces";
-import { Text } from "@tremor/react";
+import { Divider, Text } from "@tremor/react";
 import { ChatDocumentDisplay } from "./ChatDocumentDisplay";
 import { usePopup } from "@/components/admin/connectors/Popup";
 import { FiAlertTriangle, FiFileText } from "react-icons/fi";
@@ -50,6 +50,7 @@ interface DocumentSidebarProps {
   maxTokens: number;
   isLoading: boolean;
   initialWidth: number;
+  isOpen: boolean;
 }
 
 export const DocumentSidebar = forwardRef<HTMLDivElement, DocumentSidebarProps>(
@@ -64,6 +65,7 @@ export const DocumentSidebar = forwardRef<HTMLDivElement, DocumentSidebarProps>(
       maxTokens,
       isLoading,
       initialWidth,
+      isOpen,
     },
     ref: ForwardedRef<HTMLDivElement>
   ) => {
@@ -86,43 +88,53 @@ export const DocumentSidebar = forwardRef<HTMLDivElement, DocumentSidebarProps>(
 
     return (
       <div
-        style={{ width: initialWidth }}
-        ref={ref}
-        className={`sidebar absolute right-0 h-screen border-l border-l-border`}
+        className={`fixed inset-0 transition-opacity duration-300 z-50 bg-black/80 ${
+          isOpen ? "opacity-100" : "opacity-0 pointer-events-none"
+        }`}
+        onClick={(e) => {
+          if (e.target === e.currentTarget) {
+            closeSidebar();
+          }
+        }}
       >
         <div
-          className="w-full flex-initial 
-          overflow-y-hidden
-          flex
-          flex-col h-screen"
+          className={`ml-auto rounded-l-lg relative border-l bg-text-100 sidebar z-50 absolute right-0 h-screen transition-all duration-300 ${
+            isOpen ? "opacity-100 translate-x-0" : "opacity-0 translate-x-[10%]"
+          }`}
+          ref={ref}
+          style={{
+            width: initialWidth,
+          }}
         >
-          {popup}
-
-          <div className="h-4/6 flex flex-col">
-            <div className="pl-3 pr-6 mb-3 flex border-b border-border">
-              <SectionHeader
-                name={
-                  selectedMessageRetrievalType === RetrievalType.SelectedDocs
-                    ? "Referenced Documents"
-                    : "Retrieved Documents"
-                }
-                icon={FiFileText}
-                closeHeader={closeSidebar}
-              />
+          <div className="pb-6 flex-initial overflow-y-hidden flex flex-col h-screen ">
+            {popup}
+            <div className="pl-3 mx-2 pr-6 mt-3 flex text-text-800 flex-col text-2xl text-emphasis flex font-semibold">
+              {dedupedDocuments.length} Documents
+              <p className="text-sm font-semibold flex flex-wrap gap-x-2 text-text-600 mt-1">
+                Select to add to continuous context
+                <a
+                  href="https://docs.danswer.dev/introduction"
+                  className="underline cursor-pointer hover:text-strong"
+                >
+                  Learn more
+                </a>
+              </p>
             </div>
 
+            <Divider className="mb-0 mt-4 pb-2" />
+
             {currentDocuments ? (
-              <div className="overflow-y-auto dark-scrollbar flex flex-col">
+              <div className="overflow-y-auto flex-grow dark-scrollbar flex relative flex-col">
                 <div>
                   {dedupedDocuments.length > 0 ? (
                     dedupedDocuments.map((document, ind) => (
                       <div
                         key={document.document_id}
-                        className={
+                        className={`${
                           ind === dedupedDocuments.length - 1
                             ? "mb-5"
-                            : "border-b border-border-light mb-3"
-                        }
+                            : "border-b  border-border-light mb-3"
+                        }`}
                       >
                         <ChatDocumentDisplay
                           document={document}
@@ -163,84 +175,25 @@ export const DocumentSidebar = forwardRef<HTMLDivElement, DocumentSidebarProps>(
             )}
           </div>
 
-          <div className="text-sm mb-4 border-t border-border pt-4 overflow-y-hidden flex flex-col">
-            <div className="flex border-b border-border px-3">
-              <div className="flex">
-                <SectionHeader name="Selected Documents" icon={FiFileText} />
-                {tokenLimitReached && (
-                  <div className="ml-2 my-auto">
-                    <div className="mb-2">
-                      <HoverPopup
-                        mainContent={
-                          <FiAlertTriangle
-                            className="text-alert my-auto"
-                            size="16"
-                          />
-                        }
-                        popupContent={
-                          <Text className="w-40">
-                            Over LLM context length by:{" "}
-                            <i>{selectedDocumentTokens - maxTokens} tokens</i>
-                            <br />
-                            <br />
-                            {selectedDocuments &&
-                              selectedDocuments.length > 0 && (
-                                <>
-                                  Truncating: &quot;
-                                  <i>
-                                    {
-                                      selectedDocuments[
-                                        selectedDocuments.length - 1
-                                      ].semantic_identifier
-                                    }
-                                  </i>
-                                  &quot;
-                                </>
-                              )}
-                          </Text>
-                        }
-                        direction="left"
-                      />
-                    </div>
-                  </div>
-                )}
-              </div>
-              {selectedDocuments && selectedDocuments.length > 0 && (
-                <div
-                  className="ml-auto my-auto"
-                  onClick={clearSelectedDocuments}
-                >
-                  <BasicSelectable selected={false}>
-                    De-Select All
-                  </BasicSelectable>
-                </div>
-              )}
-            </div>
+          <div className="absolute left-0 bottom-0 w-full bg-gradient-to-b from-neutral-100/0 via-neutral-100/40 backdrop-blur-xs to-neutral-100 h-[100px]" />
+          <div className="sticky bottom-4 w-full left-0 justify-center flex gap-x-4">
+            <button
+              className="bg-[#84e49e] text-xs p-2 rounded text-text-800"
+              onClick={() => closeSidebar()}
+            >
+              Save Changes
+            </button>
 
-            {selectedDocuments && selectedDocuments.length > 0 ? (
-              <div className="flex flex-col gap-y-2 py-3 px-3 overflow-y-auto dark-scrollbar max-h-full">
-                {selectedDocuments.map((document) => (
-                  <SelectedDocumentDisplay
-                    key={document.document_id}
-                    document={document}
-                    handleDeselect={(documentId) => {
-                      toggleDocumentSelection(
-                        dedupedDocuments.find(
-                          (document) => document.document_id === documentId
-                        )!
-                      );
-                    }}
-                  />
-                ))}
-              </div>
-            ) : (
-              !isLoading && (
-                <Text className="mx-3 py-3">
-                  Select documents from the retrieved documents section to chat
-                  specifically with them!
-                </Text>
-              )
-            )}
+            <button
+              className="bg-error text-xs p-2 rounded text-text-200"
+              onClick={() => {
+                clearSelectedDocuments();
+
+                closeSidebar();
+              }}
+            >
+              Delete Context
+            </button>
           </div>
         </div>
       </div>

web/src/app/chat/files/InputBarPreview.tsx

@@ -1,8 +1,9 @@
-import { useState } from "react";
+import { useEffect, useRef, useState } from "react";
 import { ChatFileType, FileDescriptor } from "../interfaces";
-import { DocumentPreview } from "./documents/DocumentPreview";
+
+import { FiX, FiLoader, FiFileText } from "react-icons/fi";
 import { InputBarPreviewImage } from "./images/InputBarPreviewImage";
-import { FiX, FiLoader } from "react-icons/fi";
+import { Tooltip } from "@/components/tooltip/Tooltip";
 
 function DeleteButton({ onDelete }: { onDelete: () => void }) {
   return (
@@ -15,7 +16,7 @@ function DeleteButton({ onDelete }: { onDelete: () => void }) {
         cursor-pointer
         border-none
         bg-hover
-        p-1
+        p-.5
         rounded-full
         z-10
       "
@@ -25,6 +26,45 @@ function DeleteButton({ onDelete }: { onDelete: () => void }) {
   );
 }
 
+export function InputBarPreviewImageProvider({
+  file,
+  onDelete,
+  isUploading,
+}: {
+  file: FileDescriptor;
+  onDelete: () => void;
+  isUploading: boolean;
+}) {
+  const [isHovered, setIsHovered] = useState(false);
+
+  return (
+    <div
+      className="h-10 relative"
+      onMouseEnter={() => setIsHovered(true)}
+      onMouseLeave={() => setIsHovered(false)}
+    >
+      {isHovered && <DeleteButton onDelete={onDelete} />}
+      {isUploading && (
+        <div
+          className="
+            absolute
+            inset-0
+            flex
+            items-center
+            justify-center
+            bg-opacity-50
+            rounded-lg
+            z-0
+          "
+        >
+          <FiLoader className="animate-spin text-white" />
+        </div>
+      )}
+      <InputBarPreviewImage fileId={file.id} />
+    </div>
+  );
+}
+
 export function InputBarPreview({
   file,
   onDelete,
@@ -36,12 +76,16 @@ export function InputBarPreview({
 }) {
   const [isHovered, setIsHovered] = useState(false);
 
-  const renderContent = () => {
-    if (file.type === ChatFileType.IMAGE) {
-      return <InputBarPreviewImage fileId={file.id} />;
+  const fileNameRef = useRef<HTMLDivElement>(null);
+  const [isOverflowing, setIsOverflowing] = useState(false);
+
+  useEffect(() => {
+    if (fileNameRef.current) {
+      setIsOverflowing(
+        fileNameRef.current.scrollWidth > fileNameRef.current.clientWidth
+      );
     }
-    return <DocumentPreview fileName={file.name || file.id} />;
-  };
+  }, [file.name]);
 
   return (
     <div
@@ -49,7 +93,6 @@ export function InputBarPreview({
       onMouseEnter={() => setIsHovered(true)}
       onMouseLeave={() => setIsHovered(false)}
     >
-      {isHovered && <DeleteButton onDelete={onDelete} />}
       {isUploading && (
         <div
           className="
@@ -58,7 +101,6 @@ export function InputBarPreview({
             flex
             items-center
             justify-center
-            bg-black
             bg-opacity-50
             rounded-lg
             z-0
@@ -67,7 +109,58 @@ export function InputBarPreview({
           <FiLoader className="animate-spin text-white" />
         </div>
       )}
-      {renderContent()}
+      <div
+        className={`
+        flex
+        items-center
+        p-2
+        bg-hover
+        border
+        border-border
+        rounded-md
+        box-border
+        h-10
+      `}
+      >
+        <div className="flex-shrink-0">
+          <div
+            className="
+            w-6
+            h-6
+            bg-document
+            flex
+            items-center
+            justify-center
+            rounded-md
+          "
+          >
+            <FiFileText className="w-4 h-4 text-white" />
+          </div>
+        </div>
+        <div className="ml-2 relative">
+          <Tooltip content={file.name} side="top" align="start">
+            <div
+              ref={fileNameRef}
+              className={`font-medium text-sm line-clamp-1 break-all ellipses max-w-48`}
+            >
+              {file.name}
+            </div>
+          </Tooltip>
+        </div>
+        <button
+          onClick={onDelete}
+          className="
+            cursor-pointer
+            border-none
+            bg-hover
+            p-1
+            rounded-full
+            z-10
+          "
+        >
+          <FiX />
+        </button>
+      </div>
     </div>
   );
 }

web/src/app/chat/files/documents/DocumentPreview.tsx

@@ -5,9 +5,11 @@ import { Tooltip } from "@/components/tooltip/Tooltip";
 export function DocumentPreview({
   fileName,
   maxWidth,
+  alignBubble,
 }: {
   fileName: string;
   maxWidth?: string;
+  alignBubble?: boolean;
 }) {
   const [isOverflowing, setIsOverflowing] = useState(false);
   const fileNameRef = useRef<HTMLDivElement>(null);
@@ -22,7 +24,8 @@ export function DocumentPreview({
 
   return (
     <div
-      className="
+      className={`
+        ${alignBubble && "w-64"}
         flex
         items-center
         p-2
@@ -32,7 +35,7 @@ export function DocumentPreview({
         rounded-md
         box-border
         h-16
-      "
+      `}
     >
       <div className="flex-shrink-0">
         <div
@@ -53,7 +56,7 @@ export function DocumentPreview({
         <Tooltip content={fileName} side="top" align="start">
           <div
             ref={fileNameRef}
-            className={`font-medium text-sm truncate ${
+            className={`font-medium text-sm line-clamp-1 break-all ellipses ${
               maxWidth ? maxWidth : "max-w-48"
             }`}
           >
@@ -65,3 +68,69 @@ export function DocumentPreview({
     </div>
   );
 }
+
+export function InputDocumentPreview({
+  fileName,
+  maxWidth,
+  alignBubble,
+}: {
+  fileName: string;
+  maxWidth?: string;
+  alignBubble?: boolean;
+}) {
+  const [isOverflowing, setIsOverflowing] = useState(false);
+  const fileNameRef = useRef<HTMLDivElement>(null);
+
+  useEffect(() => {
+    if (fileNameRef.current) {
+      setIsOverflowing(
+        fileNameRef.current.scrollWidth > fileNameRef.current.clientWidth
+      );
+    }
+  }, [fileName]);
+
+  return (
+    <div
+      className={`
+        ${alignBubble && "w-64"}
+        flex
+        items-center
+        p-2
+        bg-hover
+        border
+        border-border
+        rounded-md
+        box-border
+        h-10
+      `}
+    >
+      <div className="flex-shrink-0">
+        <div
+          className="
+            w-6
+            h-6
+            bg-document
+            flex
+            items-center
+            justify-center
+            rounded-md
+          "
+        >
+          <FiFileText className="w-4 h-4 text-white" />
+        </div>
+      </div>
+      <div className="ml-2 relative">
+        <Tooltip content={fileName} side="top" align="start">
+          <div
+            ref={fileNameRef}
+            className={`font-medium text-sm line-clamp-1 break-all ellipses ${
+              maxWidth ? maxWidth : "max-w-48"
+            }`}
+          >
+            {fileName}
+          </div>
+        </Tooltip>
+      </div>
+    </div>
+  );
+}

web/src/app/chat/files/images/InMessageImage.tsx

@@ -16,14 +16,7 @@ export function InMessageImage({ fileId }: { fileId: string }) {
       />
 
       <img
-        className={`
-          max-w-lg 
-          rounded-lg 
-          bg-transparent 
-          cursor-pointer 
-          transition-opacity 
-          duration-300 
-          opacity-100`}
+        className="object-cover object-center overflow-hidden rounded-lg w-full h-full max-w-64 max-h-64 transition-opacity duration-300 opacity-100"
         onClick={() => setFullImageShowing(true)}
         src={buildImgUrl(fileId)}
         loading="lazy"

web/src/app/chat/files/images/InputBarPreviewImage.tsx

@@ -14,10 +14,24 @@ export function InputBarPreviewImage({ fileId }: { fileId: string }) {
         open={fullImageShowing}
         onOpenChange={(open) => setFullImageShowing(open)}
       />
-      <div>
+      <div
+        className={`
+          bg-transparent
+          border-none
+          flex
+          items-center
+          p-2
+          bg-hover
+          border
+          border-border
+          rounded-md
+          box-border
+          h-10
+      `}
+      >
         <img
           onClick={() => setFullImageShowing(true)}
-          className="h-16 w-16 object-cover rounded-lg bg-background cursor-pointer"
+          className="h-8 w-8 object-cover rounded-lg bg-background cursor-pointer"
           src={buildImgUrl(fileId)}
         />
       </div>

web/src/app/chat/folders/FolderList.tsx

@@ -1,6 +1,6 @@
 "use client";
 
-import React, { useState, useEffect } from "react";
+import React, { useState, useEffect, useRef } from "react";
 import { Folder } from "./interfaces";
 import { ChatSessionDisplay } from "../sessionSidebar/ChatSessionDisplay"; // Ensure this is correctly imported
 import {
@@ -22,6 +22,9 @@ import { usePopup } from "@/components/admin/connectors/Popup";
 import { useRouter } from "next/navigation";
 import { CHAT_SESSION_ID_KEY } from "@/lib/drag/constants";
 import Cookies from "js-cookie";
+import { CustomTooltip } from "@/components/tooltip/CustomTooltip";
+import { Tooltip } from "@/components/tooltip/Tooltip";
+import { Popover } from "@/components/popover/Popover";
 
 const FolderItem = ({
   folder,
@@ -54,6 +57,8 @@ const FolderItem = ({
       if (newIsExpanded) {
         openedFolders[folder.folder_id] = true;
       } else {
+        setShowDeleteConfirm(false);
+
         delete openedFolders[folder.folder_id];
       }
       Cookies.set("openedFolders", JSON.stringify(openedFolders));
@@ -87,18 +92,47 @@ const FolderItem = ({
     }
   };
 
-  const deleteFolderHandler = async (
-    event: React.MouseEvent<HTMLDivElement>
-  ) => {
-    event.stopPropagation(); // Prevent the event from bubbling up to the toggle expansion
+  const [showDeleteConfirm, setShowDeleteConfirm] = useState<boolean>(false);
+  const deleteConfirmRef = useRef<HTMLDivElement>(null);
+
+  const handleDeleteClick = (event: React.MouseEvent<HTMLDivElement>) => {
+    event.stopPropagation();
+    setShowDeleteConfirm(true);
+  };
+
+  const confirmDelete = async (event: React.MouseEvent<HTMLButtonElement>) => {
+    event.stopPropagation();
     try {
       await deleteFolder(folder.folder_id);
-      router.refresh(); // Refresh values to update the sidebar
+      router.refresh();
     } catch (error) {
       setPopup({ message: "Failed to delete folder", type: "error" });
+    } finally {
+      setShowDeleteConfirm(false);
     }
   };
 
+  const cancelDelete = (event: React.MouseEvent<HTMLButtonElement>) => {
+    event.stopPropagation();
+    setShowDeleteConfirm(false);
+  };
+
+  useEffect(() => {
+    const handleClickOutside = (event: MouseEvent) => {
+      if (
+        deleteConfirmRef.current &&
+        !deleteConfirmRef.current.contains(event.target as Node)
+      ) {
+        setShowDeleteConfirm(false);
+      }
+    };
+
+    document.addEventListener("mousedown", handleClickOutside);
+    return () => {
+      document.removeEventListener("mousedown", handleClickOutside);
+    };
+  }, []);
+
   const handleDrop = async (event: React.DragEvent<HTMLDivElement>) => {
     event.preventDefault();
     setIsDragOver(false);
@@ -134,13 +168,38 @@ const FolderItem = ({
         isDragOver ? "bg-hover" : ""
       }`}
     >
+      {showDeleteConfirm && (
+        <div
+          ref={deleteConfirmRef}
+          className="absolute max-w-xs border z-[100] border-neutral-300 top-0 right-0 w-[250px] -bo-0 top-2 mt-4 p-2 bg-background-100 rounded shadow-lg z-10"
+        >
+          <p className="text-sm mb-2">
+            Are you sure you want to delete <i>{folder.folder_name}</i>? All the
+            content inside this folder will also be deleted
+          </p>
+          <div className="flex justify-end">
+            <button
+              onClick={confirmDelete}
+              className="bg-red-500 hover:bg-red-600 text-white px-2 py-1 rounded text-xs mr-2"
+            >
+              Yes
+            </button>
+            <button
+              onClick={cancelDelete}
+              className="bg-gray-300 hover:bg-gray-200 px-2 py-1 rounded text-xs"
+            >
+              No
+            </button>
+          </div>
+        </div>
+      )}
       <BasicSelectable fullWidth selected={false}>
         <div
           onMouseEnter={() => setIsHovering(true)}
           onMouseLeave={() => setIsHovering(false)}
         >
           <div onClick={toggleFolderExpansion} className="cursor-pointer">
-            <div className="text-sm text-medium flex items-center justify-start w-full">
+            <div className="text-sm text-text-600 flex items-center justify-start w-full">
               <div className="mr-2">
                 {isExpanded ? (
                   <FiChevronDown size={16} />
@@ -172,14 +231,24 @@ const FolderItem = ({
                   >
                     <FiEdit2 size={16} />
                   </div>
-                  <div
+                  <div className="relative">
+                    <div
+                      onClick={handleDeleteClick}
+                      className="hover:bg-black/10 p-1 -m-1 rounded ml-2"
+                    >
+                      <FiTrash size={16} />
+                    </div>
+                  </div>
+
+                  {/* <div
                     onClick={deleteFolderHandler}
                     className="hover:bg-black/10 p-1 -m-1 rounded ml-2"
                   >
                     <FiTrash size={16} />
-                  </div>
+                  </div> */}
                 </div>
               )}
+
               {isEditing && (
                 <div className="flex ml-auto my-auto">
                   <div
@@ -223,7 +292,7 @@ export const FolderList = ({
 }: {
   folders: Folder[];
   currentChatId?: number;
-  openedFolders: { [key: number]: boolean };
+  openedFolders?: { [key: number]: boolean };
 }) => {
   if (folders.length === 0) {
     return null;
@@ -236,7 +305,9 @@ export const FolderList = ({
           key={folder.folder_id}
           folder={folder}
           currentChatId={currentChatId}
-          isInitiallyExpanded={openedFolders[folder.folder_id] || false}
+          isInitiallyExpanded={
+            openedFolders ? openedFolders[folder.folder_id] || false : false
+          }
         />
       ))}
     </div>

web/src/app/chat/input/ChatInputAssistant.tsx

@@ -0,0 +1,52 @@
+"use client";
+
+import { Persona } from "@/app/admin/assistants/interfaces";
+import { AssistantIcon } from "@/components/assistants/AssistantIcon";
+import { Tooltip } from "@/components/tooltip/Tooltip";
+import { ForwardedRef, forwardRef, useState } from "react";
+import { FiX } from "react-icons/fi";
+
+interface DocumentSidebarProps {
+  alternativeAssistant: Persona;
+  unToggle: () => void;
+}
+
+export const ChatInputAssistant = forwardRef<
+  HTMLDivElement,
+  DocumentSidebarProps
+>(({ alternativeAssistant, unToggle }, ref: ForwardedRef<HTMLDivElement>) => {
+  const [isHovered, setIsHovered] = useState(false);
+
+  return (
+    <div
+      onMouseEnter={() => setIsHovered(true)}
+      onMouseLeave={() => setIsHovered(false)}
+      className="flex-none h-10 duration-300 h-10 items-center rounded-lg bg-background-150"
+    >
+      <Tooltip
+        content={
+          <p className="max-w-xs flex">{alternativeAssistant.description}</p>
+        }
+      >
+        <div
+          ref={ref}
+          className="p-2 gap-x-1 relative rounded-t-lg items-center flex"
+        >
+          <AssistantIcon assistant={alternativeAssistant} border />
+          <p className="ml-1 line-clamp-1 ellipsis break-all my-auto">
+            {alternativeAssistant.name}
+          </p>
+          <div
+            className="rounded-lg rounded h-fit cursor-pointer"
+            onClick={unToggle}
+          >
+            <FiX />
+          </div>
+        </div>
+      </Tooltip>
+    </div>
+  );
+});
+
+ChatInputAssistant.displayName = "TempAssistant";
+export default ChatInputAssistant;

web/src/app/chat/input/ChatInputBar.tsx

@@ -1,36 +1,37 @@
-import React, {
-  Dispatch,
-  SetStateAction,
-  useEffect,
-  useRef,
-  useState,
-} from "react";
-import {
-  FiSend,
-  FiFilter,
-  FiPlusCircle,
-  FiCpu,
-  FiX,
-  FiPlus,
-  FiInfo,
-} from "react-icons/fi";
-import ChatInputOption from "./ChatInputOption";
-import { FaBrain } from "react-icons/fa";
+import React, { useEffect, useRef, useState } from "react";
+import { FiPlusCircle, FiPlus, FiInfo, FiX } from "react-icons/fi";
+import { ChatInputOption } from "./ChatInputOption";
 import { Persona } from "@/app/admin/assistants/interfaces";
 import { FilterManager, LlmOverrideManager } from "@/lib/hooks";
 import { SelectedFilterDisplay } from "./SelectedFilterDisplay";
 import { useChatContext } from "@/components/context/ChatContext";
 import { getFinalLLM } from "@/lib/llm/utils";
-import { FileDescriptor } from "../interfaces";
-import { InputBarPreview } from "../files/InputBarPreview";
-import { RobotIcon } from "@/components/icons/icons";
-import { Hoverable } from "@/components/Hoverable";
+import { ChatFileType, FileDescriptor } from "../interfaces";
+import {
+  InputBarPreview,
+  InputBarPreviewImageProvider,
+} from "../files/InputBarPreview";
+import {
+  AssistantsIconSkeleton,
+  CpuIconSkeleton,
+  FileIcon,
+  SendIcon,
+} from "@/components/icons/icons";
+import { IconType } from "react-icons";
+import Popup from "../../../components/popup/Popup";
+import { LlmTab } from "../modal/configuration/LlmTab";
+import { AssistantsTab } from "../modal/configuration/AssistantsTab";
+import ChatInputAssistant from "./ChatInputAssistant";
+import { DanswerDocument } from "@/lib/search/interfaces";
 import { AssistantIcon } from "@/components/assistants/AssistantIcon";
 import { Tooltip } from "@/components/tooltip/Tooltip";
+import { Hoverable } from "@/components/Hoverable";
 const MAX_INPUT_HEIGHT = 200;
 
 export function ChatInputBar({
   personas,
+  showDocs,
+  selectedDocuments,
   message,
   setMessage,
   onSubmit,
@@ -42,13 +43,21 @@ export function ChatInputBar({
   onSetSelectedAssistant,
   selectedAssistant,
   files,
+
+  setSelectedAssistant,
   setFiles,
   handleFileUpload,
   setConfigModalActiveTab,
   textAreaRef,
   alternativeAssistant,
+  chatSessionId,
+  availableAssistants,
 }: {
+  showDocs: () => void;
+  selectedDocuments: DanswerDocument[];
+  availableAssistants: Persona[];
   onSetSelectedAssistant: (alternativeAssistant: Persona | null) => void;
+  setSelectedAssistant: (assistant: Persona) => void;
   personas: Persona[];
   message: string;
   setMessage: (message: string) => void;
@@ -65,6 +74,7 @@ export function ChatInputBar({
   handleFileUpload: (files: File[]) => void;
   setConfigModalActiveTab: (tab: string) => void;
   textAreaRef: React.RefObject<HTMLTextAreaElement>;
+  chatSessionId?: number;
 }) {
   // handle re-sizing of the text area
   useEffect(() => {
@@ -102,19 +112,6 @@ export function ChatInputBar({
   const [showSuggestions, setShowSuggestions] = useState(false);
 
   const interactionsRef = useRef<HTMLDivElement | null>(null);
-
-  const hideSuggestions = () => {
-    setShowSuggestions(false);
-    setAssistantIconIndex(0);
-  };
-
-  // Update selected persona
-  const updateCurrentPersona = (persona: Persona) => {
-    onSetSelectedAssistant(persona.id == selectedAssistant.id ? null : persona);
-    hideSuggestions();
-    setMessage("");
-  };
-
   // Click out of assistant suggestions
   useEffect(() => {
     const handleClickOutside = (event: MouseEvent) => {
@@ -133,6 +130,18 @@ export function ChatInputBar({
     };
   }, []);
 
+  const hideSuggestions = () => {
+    setShowSuggestions(false);
+    setAssistantIconIndex(0);
+  };
+
+  // Update selected persona
+  const updateCurrentPersona = (persona: Persona) => {
+    onSetSelectedAssistant(persona.id == selectedAssistant.id ? null : persona);
+    hideSuggestions();
+    setMessage("");
+  };
+
   // Complete user input handling
   const handleInputChange = (event: React.ChangeEvent<HTMLTextAreaElement>) => {
     const text = event.target.value;
@@ -143,7 +152,6 @@ export function ChatInputBar({
       return;
     }
 
-    // If looking for an assistant...fup
     const match = text.match(/(?:\s|^)@(\w*)$/);
     if (match) {
       setShowSuggestions(true);
@@ -179,7 +187,12 @@ export function ChatInputBar({
           filteredPersonas[assistantIconIndex >= 0 ? assistantIconIndex : 0];
         updateCurrentPersona(option);
       }
-    } else if (e.key === "ArrowDown") {
+    }
+    if (!showSuggestions) {
+      return;
+    }
+
+    if (e.key === "ArrowDown") {
       e.preventDefault();
       setAssistantIconIndex((assistantIconIndex) =>
         Math.min(assistantIconIndex + 1, filteredPersonas.length)
@@ -197,13 +210,12 @@ export function ChatInputBar({
       <div className="flex justify-center pb-2 max-w-screen-lg mx-auto mb-2">
         <div
           className="
-            w-full
+            w-[90%]
             shrink
+            bg-background
             relative
             px-4
-            w-searchbar-xs
-            2xl:w-searchbar-sm
-            3xl:w-searchbar
+            max-w-searchbar-max
             mx-auto
           "
         >
@@ -212,16 +224,18 @@ export function ChatInputBar({
               ref={suggestionsRef}
               className="text-sm absolute inset-x-0 top-0 w-full transform -translate-y-full"
             >
-              <div className="rounded-lg py-1.5 bg-white border border-border-medium overflow-hidden shadow-lg mx-2 px-1.5 mt-2 rounded z-10">
+              <div className="rounded-lg py-1.5 bg-background border border-border-medium shadow-lg mx-2 px-1.5 mt-2 rounded z-10">
                 {filteredPersonas.map((currentPersona, index) => (
                   <button
                     key={index}
-                    className={`px-2 ${assistantIconIndex == index && "bg-hover"} rounded content-start flex gap-x-1 py-1.5 w-full  hover:bg-hover cursor-pointer`}
+                    className={`px-2 ${
+                      assistantIconIndex == index && "bg-hover-lightish"
+                    } rounded  rounded-lg content-start flex gap-x-1 py-2 w-full  hover:bg-hover-lightish cursor-pointer`}
                     onClick={() => {
                       updateCurrentPersona(currentPersona);
                     }}
                   >
-                    <p className="font-bold ">{currentPersona.name}</p>
+                    <p className="font-bold">{currentPersona.name}</p>
                     <p className="line-clamp-1">
                       {currentPersona.id == selectedAssistant.id &&
                         "(default) "}
@@ -232,7 +246,9 @@ export function ChatInputBar({
                 <a
                   key={filteredPersonas.length}
                   target="_blank"
-                  className={`${assistantIconIndex == filteredPersonas.length && "bg-hover"} px-3 flex gap-x-1 py-2 w-full  items-center  hover:bg-hover-light cursor-pointer"`}
+                  className={`${
+                    assistantIconIndex == filteredPersonas.length && "bg-hover"
+                  } rounded rounded-lg px-3 flex gap-x-1 py-2 w-full  items-center  hover:bg-hover-lightish cursor-pointer"`}
                   href="/assistants/new"
                 >
                   <FiPlus size={17} />
@@ -241,23 +257,21 @@ export function ChatInputBar({
               </div>
             </div>
           )}
-
           <div>
             <SelectedFilterDisplay filterManager={filterManager} />
           </div>
-
           <div
             className="
               opacity-100
               w-full
               h-fit
+              bg-bl
               flex
               flex-col
               border
-              border-border-medium
+              border-[#E5E7EB]
               rounded-lg
-              overflow-hidden
-              bg-background-weak
+              bg-background-100
               [&:has(textarea:focus)]::ring-1
               [&:has(textarea:focus)]::ring-black
             "
@@ -266,13 +280,13 @@ export function ChatInputBar({
               <div className="flex flex-wrap gap-y-1 gap-x-2 px-2 pt-1.5 w-full">
                 <div
                   ref={interactionsRef}
-                  className="bg-background-subtle p-2 rounded-t-lg  items-center flex w-full"
+                  className="bg-background-200 p-2 rounded-t-lg items-center flex w-full"
                 >
                   <AssistantIcon assistant={alternativeAssistant} border />
                   <p className="ml-3 text-strong my-auto">
                     {alternativeAssistant.name}
                   </p>
-                  <div className="flex gap-x-1 ml-auto ">
+                  <div className="flex gap-x-1 ml-auto">
                     <Tooltip
                       content={
                         <p className="max-w-xs flex flex-wrap">
@@ -293,24 +307,50 @@ export function ChatInputBar({
                 </div>
               </div>
             )}
-
-            {files.length > 0 && (
-              <div className="flex flex-wrap gap-y-1 gap-x-2 px-2 pt-2">
-                {files.map((file) => (
-                  <div key={file.id}>
-                    <InputBarPreview
-                      file={file}
-                      onDelete={() => {
-                        setFiles(
-                          files.filter(
-                            (fileInFilter) => fileInFilter.id !== file.id
-                          )
-                        );
-                      }}
-                      isUploading={file.isUploading || false}
-                    />
-                  </div>
-                ))}
+            {(selectedDocuments.length > 0 || files.length > 0) && (
+              <div className="flex gap-x-2 px-2 pt-2">
+                <div className="flex gap-x-1 px-2 overflow-y-auto overflow-x-scroll items-end miniscroll">
+                  {selectedDocuments.length > 0 && (
+                    <button
+                      onClick={showDocs}
+                      className="flex-none flex cursor-pointer hover:bg-background-200 transition-colors duration-300 h-10 p-1 items-center gap-x-1 rounded-lg bg-background-150 max-w-[100px]"
+                    >
+                      <FileIcon size={24} />
+                      <p className="text-xs">
+                        {selectedDocuments.length} selected
+                      </p>
+                    </button>
+                  )}
+                  {files.map((file) => (
+                    <div className="flex-none" key={file.id}>
+                      {file.type === ChatFileType.IMAGE ? (
+                        <InputBarPreviewImageProvider
+                          file={file}
+                          onDelete={() => {
+                            setFiles(
+                              files.filter(
+                                (fileInFilter) => fileInFilter.id !== file.id
+                              )
+                            );
+                          }}
+                          isUploading={file.isUploading || false}
+                        />
+                      ) : (
+                        <InputBarPreview
+                          file={file}
+                          onDelete={() => {
+                            setFiles(
+                              files.filter(
+                                (fileInFilter) => fileInFilter.id !== file.id
+                              )
+                            );
+                          }}
+                          isUploading={file.isUploading || false}
+                        />
+                      )}
+                    </div>
+                  ))}
+                </div>
               </div>
             )}
 
@@ -324,21 +364,20 @@ export function ChatInputBar({
                 w-full
                 shrink
                 resize-none
+                rounded-lg
                 border-0
-                bg-background-weak
+                bg-background-100
                 ${
                   textAreaRef.current &&
                   textAreaRef.current.scrollHeight > MAX_INPUT_HEIGHT
                     ? "overflow-y-auto mt-2"
                     : ""
                 }
-                overflow-hidden
                 whitespace-normal
                 break-word
                 overscroll-contain
                 outline-none
                 placeholder-subtle
-                overflow-hidden
                 resize-none
                 pl-4
                 pr-12
@@ -349,7 +388,7 @@ export function ChatInputBar({
               style={{ scrollbarWidth: "thin" }}
               role="textarea"
               aria-multiline
-              placeholder="Send a message..."
+              placeholder="Send a message or @ to tag an assistant..."
               value={message}
               onKeyDown={(event) => {
                 if (
@@ -364,39 +403,60 @@ export function ChatInputBar({
               }}
               suppressContentEditableWarning={true}
             />
-            <div className="flex items-center space-x-3 mr-12 px-4 pb-2 overflow-hidden">
-              <ChatInputOption
-                flexPriority="shrink"
-                name={selectedAssistant ? selectedAssistant.name : "Assistants"}
-                icon={FaBrain}
-                onClick={() => setConfigModalActiveTab("assistants")}
-              />
 
-              <ChatInputOption
-                flexPriority="second"
-                name={
-                  llmOverrideManager.llmOverride.modelName ||
-                  (selectedAssistant
-                    ? selectedAssistant.llm_model_version_override || llmName
-                    : llmName)
-                }
-                icon={FiCpu}
-                onClick={() => setConfigModalActiveTab("llms")}
-              />
-
-              {!retrievalDisabled && (
+            <div className="flex items-center space-x-3 mr-12 px-4 pb-2 ">
+              <Popup
+                removePadding
+                content={(close) => (
+                  <AssistantsTab
+                    availableAssistants={availableAssistants}
+                    llmProviders={llmProviders}
+                    selectedAssistant={selectedAssistant}
+                    onSelect={(assistant) => {
+                      setSelectedAssistant(assistant);
+                      close();
+                    }}
+                  />
+                )}
+                position="top"
+              >
                 <ChatInputOption
-                  flexPriority="stiff"
-                  name="Filters"
-                  icon={FiFilter}
-                  onClick={() => setConfigModalActiveTab("filters")}
+                  flexPriority="shrink"
+                  name={
+                    selectedAssistant ? selectedAssistant.name : "Assistants"
+                  }
+                  Icon={AssistantsIconSkeleton as IconType}
                 />
-              )}
+              </Popup>
+
+              <Popup
+                content={(close, ref) => (
+                  <LlmTab
+                    close={close}
+                    ref={ref}
+                    llmOverrideManager={llmOverrideManager}
+                    chatSessionId={chatSessionId}
+                    currentAssistant={selectedAssistant}
+                  />
+                )}
+                position="top"
+              >
+                <ChatInputOption
+                  flexPriority="second"
+                  name={
+                    llmOverrideManager.llmOverride.modelName ||
+                    (selectedAssistant
+                      ? selectedAssistant.llm_model_version_override || llmName
+                      : llmName)
+                  }
+                  Icon={CpuIconSkeleton}
+                />
+              </Popup>
 
               <ChatInputOption
                 flexPriority="stiff"
                 name="File"
-                icon={FiPlusCircle}
+                Icon={FiPlusCircle}
                 onClick={() => {
                   const input = document.createElement("input");
                   input.type = "file";
@@ -426,10 +486,10 @@ export function ChatInputBar({
                   }
                 }}
               >
-                <FiSend
-                  size={18}
-                  className={`text-emphasis w-9 h-9 p-2 rounded-lg ${
-                    message ? "bg-blue-200" : ""
+                <SendIcon
+                  size={28}
+                  className={`text-emphasis text-white p-1 rounded-full ${
+                    message ? "bg-background-800" : "bg-[#D7D7D7]"
                   }`}
                 />
               </div>