Merge pull request #1467 from AnishSarkar22/feat/ui-fixes

feat: improve chat navigation, automation UI, and summary-free indexing

surfsense_backend/alembic/versions/149_add_gateway_tables.py

@@ -54,6 +54,17 @@ USER_COLS = [
     "premium_credit_micros_used",
 ]
 
+AUTOMATION_RUN_COLS = [
+    "id",
+    "automation_id",
+    "trigger_id",
+    "status",
+    "step_results",
+    "started_at",
+    "finished_at",
+    "created_at",
+]
+
 def _has_zero_version(conn, table: str) -> bool:
     return (
         conn.execute(
@@ -150,7 +161,8 @@ def _build_set_table_ddl(
         f"new_chat_messages, "
         f"chat_comments, "
         f"chat_session_state, "
-        f'"user" ({_cols(user_cols)})'
+        f'"user" ({_cols(user_cols)}), '
+        f"automation_runs ({_cols(AUTOMATION_RUN_COLS)})"
     )
 
 
@@ -523,7 +535,7 @@ def downgrade() -> None:
     if exists:
         documents_has_zero_ver = _has_zero_version(conn, "documents")
         user_has_zero_ver = _has_zero_version(conn, "user")
-        # Restore the publication shape from migration 143.
+        # Restore the publication shape from migration 148.
         doc_cols = DOCUMENT_COLS + (['"_0_version"'] if documents_has_zero_ver else [])
         user_cols = USER_COLS + (['"_0_version"'] if user_has_zero_ver else [])
         ddl = (
@@ -535,7 +547,8 @@ def downgrade() -> None:
             f"new_chat_messages, "
             f"chat_comments, "
             f"chat_session_state, "
-            f'"user" ({_cols(user_cols)})'
+            f'"user" ({_cols(user_cols)}), '
+            f"automation_runs ({_cols(AUTOMATION_RUN_COLS)})"
         )
         tx = conn.begin_nested() if conn.in_transaction() else conn.begin()
         with tx:

surfsense_backend/alembic/versions/154_remove_document_summary_llm.py

@@ -0,0 +1,134 @@
+"""remove document summary llm settings
+
+Revision ID: 154
+Revises: 153
+"""
+
+from collections.abc import Sequence
+
+import sqlalchemy as sa
+
+from alembic import op
+
+revision: str = "154"
+down_revision: str | None = "153"
+branch_labels: str | Sequence[str] | None = None
+depends_on: str | Sequence[str] | None = None
+
+PUBLICATION_NAME = "zero_publication"
+
+DOCUMENT_COLS = [
+    "id",
+    "title",
+    "document_type",
+    "search_space_id",
+    "folder_id",
+    "created_by_id",
+    "status",
+    "created_at",
+    "updated_at",
+]
+
+USER_COLS = [
+    "id",
+    "pages_limit",
+    "pages_used",
+    "premium_credit_micros_limit",
+    "premium_credit_micros_used",
+]
+
+AUTOMATION_RUN_COLS = [
+    "id",
+    "automation_id",
+    "trigger_id",
+    "status",
+    "step_results",
+    "started_at",
+    "finished_at",
+    "created_at",
+]
+
+
+def _column_exists(conn, table: str, column: str) -> bool:
+    return (
+        conn.execute(
+            sa.text(
+                "SELECT 1 FROM information_schema.columns "
+                "WHERE table_name = :table AND column_name = :column"
+            ),
+            {"table": table, "column": column},
+        ).fetchone()
+        is not None
+    )
+
+
+def _has_zero_version(conn, table: str) -> bool:
+    return _column_exists(conn, table, "_0_version")
+
+
+def _set_table_ddl(conn) -> str:
+    doc_cols = DOCUMENT_COLS + (['"_0_version"'] if _has_zero_version(conn, "documents") else [])
+    user_cols = USER_COLS + (['"_0_version"'] if _has_zero_version(conn, "user") else [])
+    tables = [
+        "notifications",
+        f"documents ({', '.join(doc_cols)})",
+        "folders",
+        "search_source_connectors",
+        "new_chat_messages",
+        "chat_comments",
+        "chat_session_state",
+        f'"user" ({", ".join(user_cols)})',
+        f"automation_runs ({', '.join(AUTOMATION_RUN_COLS)})",
+    ]
+    return f"ALTER PUBLICATION {PUBLICATION_NAME} SET TABLE " + ", ".join(tables)
+
+
+def _resync_zero_publication(tag: str) -> None:
+    conn = op.get_bind()
+    exists = conn.execute(
+        sa.text("SELECT 1 FROM pg_publication WHERE pubname = :name"),
+        {"name": PUBLICATION_NAME},
+    ).fetchone()
+    if not exists:
+        return
+
+    tx = conn.begin_nested() if conn.in_transaction() else conn.begin()
+    with tx:
+        conn.execute(sa.text(f"COMMENT ON PUBLICATION {PUBLICATION_NAME} IS 'pre-{tag}'"))
+        conn.execute(sa.text(_set_table_ddl(conn)))
+        conn.execute(sa.text(f"COMMENT ON PUBLICATION {PUBLICATION_NAME} IS 'post-{tag}'"))
+
+
+def upgrade() -> None:
+    conn = op.get_bind()
+
+    if _column_exists(conn, "searchspaces", "document_summary_llm_id"):
+        op.drop_column("searchspaces", "document_summary_llm_id")
+
+    if _column_exists(conn, "search_source_connectors", "enable_summary"):
+        op.drop_column("search_source_connectors", "enable_summary")
+
+    _resync_zero_publication("154-summary-removal")
+
+
+def downgrade() -> None:
+    conn = op.get_bind()
+
+    if not _column_exists(conn, "searchspaces", "document_summary_llm_id"):
+        op.add_column(
+            "searchspaces",
+            sa.Column("document_summary_llm_id", sa.Integer(), nullable=True, server_default="0"),
+        )
+
+    if not _column_exists(conn, "search_source_connectors", "enable_summary"):
+        op.add_column(
+            "search_source_connectors",
+            sa.Column(
+                "enable_summary",
+                sa.Boolean(),
+                nullable=False,
+                server_default=sa.text("false"),
+            ),
+        )
+
+    _resync_zero_publication("154-summary-removal-downgrade")

surfsense_backend/app/agents/multi_agent_chat/subagents/builtins/deliverables/tools/report.py

@@ -16,7 +16,7 @@ from app.agents.shared.receipt import make_receipt
 from app.agents.shared.receipt_command import with_receipt
 from app.db import Report, shielded_async_session
 from app.services.connector_service import ConnectorService
-from app.services.llm_service import get_document_summary_llm
+from app.services.llm_service import get_agent_llm
 
 logger = logging.getLogger(__name__)
 
@@ -546,7 +546,7 @@ def create_generate_report_tool(
     Factory function to create the generate_report tool with injected dependencies.
 
     The tool generates a Markdown report inline using the search space's
-    document summary LLM, saves it to the database, and returns immediately.
+    agent LLM, saves it to the database, and returns immediately.
 
     Uses short-lived database sessions for each DB operation so no connection
     is held during the long LLM API call.
@@ -767,7 +767,7 @@ def create_generate_report_tool(
                             "creating standalone report"
                         )
 
-                llm = await get_document_summary_llm(read_session, search_space_id)
+                llm = await get_agent_llm(read_session, search_space_id)
             # read_session closed — connection returned to pool
 
             if not llm:

surfsense_backend/app/agents/multi_agent_chat/subagents/builtins/deliverables/tools/resume.py

@@ -17,7 +17,7 @@ from langgraph.types import Command
 from app.agents.shared.receipt import make_receipt
 from app.agents.shared.receipt_command import with_receipt
 from app.db import Report, shielded_async_session
-from app.services.llm_service import get_document_summary_llm
+from app.services.llm_service import get_agent_llm
 
 logger = logging.getLogger(__name__)
 
@@ -578,7 +578,7 @@ def create_generate_resume_tool(
                             f"(group {report_group_id})"
                         )
 
-                llm = await get_document_summary_llm(read_session, search_space_id)
+                llm = await get_agent_llm(read_session, search_space_id)
 
             if not llm:
                 error_msg = (

surfsense_backend/app/agents/new_chat/tools/report.py

@@ -35,7 +35,7 @@ from langchain_core.tools import tool
 
 from app.db import Report, shielded_async_session
 from app.services.connector_service import ConnectorService
-from app.services.llm_service import get_document_summary_llm
+from app.services.llm_service import get_agent_llm
 
 logger = logging.getLogger(__name__)
 
@@ -565,7 +565,7 @@ def create_generate_report_tool(
     Factory function to create the generate_report tool with injected dependencies.
 
     The tool generates a Markdown report inline using the search space's
-    document summary LLM, saves it to the database, and returns immediately.
+    agent LLM, saves it to the database, and returns immediately.
 
     Uses short-lived database sessions for each DB operation so no connection
     is held during the long LLM API call.
@@ -768,7 +768,7 @@ def create_generate_report_tool(
                             "creating standalone report"
                         )
 
-                llm = await get_document_summary_llm(read_session, search_space_id)
+                llm = await get_agent_llm(read_session, search_space_id)
             # read_session closed — connection returned to pool
 
             if not llm:

surfsense_backend/app/agents/new_chat/tools/resume.py

@@ -26,7 +26,7 @@ from langchain_core.messages import HumanMessage
 from langchain_core.tools import tool
 
 from app.db import Report, shielded_async_session
-from app.services.llm_service import get_document_summary_llm
+from app.services.llm_service import get_agent_llm
 
 logger = logging.getLogger(__name__)
 
@@ -547,7 +547,7 @@ def create_generate_resume_tool(
                             f"(group {report_group_id})"
                         )
 
-                llm = await get_document_summary_llm(read_session, search_space_id)
+                llm = await get_agent_llm(read_session, search_space_id)
 
             if not llm:
                 error_msg = (

surfsense_backend/app/agents/podcaster/nodes.py

@@ -31,12 +31,10 @@ async def create_podcast_transcript(
     search_space_id = configuration.search_space_id
     user_prompt = configuration.user_prompt
 
-    # Get search space's document summary LLM
+    # Use the search space's agent LLM for podcast transcript generation.
     llm = await get_agent_llm(state.db_session, search_space_id)
     if not llm:
-        error_message = (
+        error_message = f"No agent LLM configured for search space {search_space_id}"
-            f"No document summary LLM configured for search space {search_space_id}"
-        )
         print(error_message)
         raise RuntimeError(error_message)
 

surfsense_backend/app/celery_app.py

@@ -103,7 +103,7 @@ def init_worker(**kwargs):
     """Initialize the LLM Router and Image Gen Router when a Celery worker process starts.
 
     This ensures the Auto mode (LiteLLM Router) is available for background tasks
-    like document summarization and image generation.
+    like agent workflows and image generation.
     """
     from app.observability.bootstrap import init_otel
 

surfsense_backend/app/connectors/google_drive/content_extractor.py

@@ -141,7 +141,6 @@ async def download_and_process_file(
     task_logger: TaskLoggingService,
     log_entry: Log,
     connector_id: int | None = None,
-    enable_summary: bool = True,
 ) -> tuple[Any, str | None, dict[str, Any] | None]:
     """
     Download Google Drive file and process using Surfsense file processors.
@@ -215,8 +214,6 @@ async def download_and_process_file(
                 "source_connector": "google_drive",
             },
         }
-        # Include connector_id for de-indexing support
-        connector_info["enable_summary"] = enable_summary
         if connector_id is not None:
             connector_info["connector_id"] = connector_id
 

surfsense_backend/app/db.py

@@ -1781,9 +1781,6 @@ class SearchSpace(BaseModel, TimestampMixin):
     agent_llm_id = Column(
         Integer, nullable=True, default=0
     )  # For agent/chat operations, defaults to Auto mode
-    document_summary_llm_id = Column(
-        Integer, nullable=True, default=0
-    )  # For document summarization, defaults to Auto mode
     image_generation_config_id = Column(
         Integer, nullable=True, default=0
     )  # For image generation, defaults to Auto mode
@@ -1951,12 +1948,6 @@ class SearchSourceConnector(BaseModel, TimestampMixin):
     last_indexed_at = Column(TIMESTAMP(timezone=True), nullable=True)
     config = Column(JSON, nullable=False)
 
-    # Summary generation (LLM-based) - disabled by default to save resources.
-    # When enabled, improves hybrid search quality at the cost of LLM calls.
-    enable_summary = Column(
-        Boolean, nullable=False, default=False, server_default="false"
-    )
-
     # Vision LLM for image files - disabled by default to save cost/time.
     # When enabled, images are described via a vision language model instead
     # of falling back to the document parser.
@@ -2972,7 +2963,7 @@ async def shielded_async_session():
 async def setup_indexes():
     async with engine.begin() as conn:
         # Create indexes
-        # Document Summary Indexes
+        # Document embedding indexes
         await conn.execute(
             text(
                 "CREATE INDEX IF NOT EXISTS document_vector_index ON documents USING hnsw (embedding public.vector_cosine_ops)"

surfsense_backend/app/indexing_pipeline/adapters/file_upload_adapter.py

@@ -18,8 +18,6 @@ class UploadDocumentAdapter:
         etl_service: str,
         search_space_id: int,
         user_id: str,
-        llm,
-        should_summarize: bool = False,
     ) -> None:
         connector_doc = ConnectorDocument(
             title=filename,
@@ -29,9 +27,7 @@ class UploadDocumentAdapter:
             search_space_id=search_space_id,
             created_by_id=user_id,
             connector_id=None,
-            should_summarize=should_summarize,
             should_use_code_chunker=False,
-            fallback_summary=markdown_content[:4000],
             metadata={
                 "FILE_NAME": filename,
                 "ETL_SERVICE": etl_service,
@@ -43,7 +39,7 @@ class UploadDocumentAdapter:
         if not documents:
             raise RuntimeError("prepare_for_indexing returned no documents")
 
-        indexed = await self._service.index(documents[0], connector_doc, llm)
+        indexed = await self._service.index(documents[0], connector_doc)
 
         if not DocumentStatus.is_state(indexed.status, DocumentStatus.READY):
             raise RuntimeError(indexed.status.get("reason", "Indexing failed"))
@@ -51,7 +47,7 @@ class UploadDocumentAdapter:
         indexed.content_needs_reindexing = False
         await self._session.commit()
 
-    async def reindex(self, document: Document, llm) -> None:
+    async def reindex(self, document: Document) -> None:
         """Re-index an existing document after its source_markdown has been updated."""
         if not document.source_markdown:
             raise RuntimeError("Document has no source_markdown to reindex")
@@ -66,15 +62,13 @@ class UploadDocumentAdapter:
             search_space_id=document.search_space_id,
             created_by_id=str(document.created_by_id),
             connector_id=document.connector_id,
-            should_summarize=True,
             should_use_code_chunker=False,
-            fallback_summary=document.source_markdown[:4000],
             metadata=metadata,
         )
 
         document.content_hash = compute_content_hash(connector_doc)
 
-        indexed = await self._service.index(document, connector_doc, llm)
+        indexed = await self._service.index(document, connector_doc)
 
         if not DocumentStatus.is_state(indexed.status, DocumentStatus.READY):
             raise RuntimeError(indexed.status.get("reason", "Reindexing failed"))

surfsense_backend/app/indexing_pipeline/connector_document.py

@@ -11,9 +11,7 @@ class ConnectorDocument(BaseModel):
     unique_id: str
     document_type: DocumentType
     search_space_id: int = Field(gt=0)
-    should_summarize: bool = True
     should_use_code_chunker: bool = False
-    fallback_summary: str | None = None
     metadata: dict = {}
     connector_id: int | None = None
     created_by_id: str

surfsense_backend/app/indexing_pipeline/document_summarizer.py

@@ -1,30 +0,0 @@
-from app.prompts import SUMMARY_PROMPT_TEMPLATE
-from app.utils.document_converters import optimize_content_for_context_window
-
-
-async def summarize_document(
-    source_markdown: str, llm, metadata: dict | None = None
-) -> str:
-    """Generate a text summary of a document using an LLM, prefixed with metadata when provided."""
-    model_name = getattr(llm, "model", "gpt-3.5-turbo")
-    optimized_content = optimize_content_for_context_window(
-        source_markdown, metadata, model_name
-    )
-
-    summary_chain = SUMMARY_PROMPT_TEMPLATE | llm
-    content_with_metadata = (
-        f"<DOCUMENT><DOCUMENT_METADATA>\n\n{metadata}\n\n</DOCUMENT_METADATA>"
-        f"\n\n<DOCUMENT_CONTENT>\n\n{optimized_content}\n\n</DOCUMENT_CONTENT></DOCUMENT>"
-    )
-    summary_result = await summary_chain.ainvoke({"document": content_with_metadata})
-    summary_content = summary_result.content
-
-    if metadata:
-        metadata_parts = ["# DOCUMENT METADATA"]
-        for key, value in metadata.items():
-            if value:
-                metadata_parts.append(f"**{key.replace('_', ' ').title()}:** {value}")
-        metadata_section = "\n".join(metadata_parts)
-        return f"{metadata_section}\n\n# DOCUMENT SUMMARY\n\n{summary_content}"
-
-    return summary_content

surfsense_backend/app/indexing_pipeline/indexing_pipeline_service.py

@@ -31,7 +31,6 @@ from app.indexing_pipeline.document_persistence import (
     attach_chunks_to_document,
     rollback_and_persist_failure,
 )
-from app.indexing_pipeline.document_summarizer import summarize_document
 from app.indexing_pipeline.exceptions import (
     EMBEDDING_ERRORS,
     PERMANENT_LLM_ERRORS,
@@ -203,9 +202,7 @@ class IndexingPipelineService:
 
         await self.session.commit()
 
-    async def index_batch(
+    async def index_batch(self, connector_docs: list[ConnectorDocument]) -> list[Document]:
-        self, connector_docs: list[ConnectorDocument], llm
-    ) -> list[Document]:
         """Convenience method: prepare_for_indexing then index each document.
 
         Indexers that need heartbeat callbacks or custom per-document logic
@@ -218,7 +215,7 @@ class IndexingPipelineService:
             connector_doc = doc_map.get(document.unique_identifier_hash)
             if connector_doc is None:
                 continue
-            result = await self.index(document, connector_doc, llm)
+            result = await self.index(document, connector_doc)
             results.append(result)
         return results
 
@@ -350,11 +347,9 @@ class IndexingPipelineService:
             await self.session.rollback()
             return []
 
-    async def index(
+    async def index(self, document: Document, connector_doc: ConnectorDocument) -> Document:
-        self, document: Document, connector_doc: ConnectorDocument, llm
-    ) -> Document:
         """
-        Run summarization, embedding, and chunking for a document and persist the results.
+        Run deterministic content storage, embedding, and chunking for a document.
         """
         ctx = PipelineLogContext(
             connector_id=connector_doc.connector_id,
@@ -379,20 +374,7 @@ class IndexingPipelineService:
             document.status = DocumentStatus.processing()
             await self.session.commit()
 
-            t_step = time.perf_counter()
+            content = connector_doc.source_markdown
-            if connector_doc.should_summarize and llm is not None:
-                content = await summarize_document(
-                    connector_doc.source_markdown, llm, connector_doc.metadata
-                )
-                perf.info(
-                    "[indexing] summarize_document doc=%d in %.3fs",
-                    document.id,
-                    time.perf_counter() - t_step,
-                )
-            elif connector_doc.should_summarize and connector_doc.fallback_summary:
-                content = connector_doc.fallback_summary
-            else:
-                content = connector_doc.source_markdown
 
             await self.session.execute(
                 delete(Chunk).where(Chunk.document_id == document.id)
@@ -523,7 +505,6 @@ class IndexingPipelineService:
     async def index_batch_parallel(
         self,
         connector_docs: list[ConnectorDocument],
-        get_llm: Callable[[AsyncSession], Awaitable],
         *,
         max_concurrency: int = 4,
         on_heartbeat: Callable[[int], Awaitable[None]] | None = None,
@@ -532,8 +513,8 @@ class IndexingPipelineService:
         """Index documents in parallel with bounded concurrency.
 
         Phase 1 (serial): prepare_for_indexing using self.session.
-        Phase 2 (parallel): index each document in an isolated session,
+        Phase 2 (parallel): index each document in an isolated session, bounded
-        bounded by a semaphore to avoid overwhelming APIs/DB.
+        by a semaphore to avoid overwhelming embedding APIs/DB.
         """
         logger = logging.getLogger(__name__)
         perf = get_perf_logger()
@@ -577,9 +558,8 @@ class IndexingPipelineService:
                                 failed_count += 1
                             return document
 
-                        llm = await get_llm(isolated_session)
                         iso_pipeline = IndexingPipelineService(isolated_session)
-                        result = await iso_pipeline.index(refetched, connector_doc, llm)
+                        result = await iso_pipeline.index(refetched, connector_doc)
 
                         async with lock:
                             if DocumentStatus.is_state(

surfsense_backend/app/routes/documents_routes.py

@@ -125,7 +125,6 @@ async def create_documents(
 async def create_documents_file_upload(
     files: list[UploadFile],
     search_space_id: int = Form(...),
-    should_summarize: bool = Form(False),
     use_vision_llm: bool = Form(False),
     processing_mode: str = Form("basic"),
     session: AsyncSession = Depends(get_async_session),
@@ -309,7 +308,6 @@ async def create_documents_file_upload(
                 filename=filename,
                 search_space_id=search_space_id,
                 user_id=str(user.id),
-                should_summarize=should_summarize,
                 use_vision_llm=use_vision_llm,
                 processing_mode=validated_mode.value,
             )
@@ -1586,7 +1584,6 @@ async def folder_upload(
     search_space_id: int = Form(...),
     relative_paths: str = Form(...),
     root_folder_id: int | None = Form(None),
-    enable_summary: bool = Form(False),
     use_vision_llm: bool = Form(False),
     processing_mode: str = Form("basic"),
     session: AsyncSession = Depends(get_async_session),
@@ -1719,7 +1716,6 @@ async def folder_upload(
         user_id=str(user.id),
         folder_name=folder_name,
         root_folder_id=root_folder_id,
-        enable_summary=enable_summary,
         use_vision_llm=use_vision_llm,
         file_mappings=list(file_mappings),
         processing_mode=validated_mode.value,

surfsense_backend/app/routes/search_spaces_routes.py

@@ -617,9 +617,6 @@ async def get_llm_preferences(
 
         # Get full config objects for each role
         agent_llm = await _get_llm_config_by_id(session, search_space.agent_llm_id)
-        document_summary_llm = await _get_llm_config_by_id(
-            session, search_space.document_summary_llm_id
-        )
         image_generation_config = await _get_image_gen_config_by_id(
             session, search_space.image_generation_config_id
         )
@@ -629,11 +626,9 @@ async def get_llm_preferences(
 
         return LLMPreferencesRead(
             agent_llm_id=search_space.agent_llm_id,
-            document_summary_llm_id=search_space.document_summary_llm_id,
             image_generation_config_id=search_space.image_generation_config_id,
             vision_llm_config_id=search_space.vision_llm_config_id,
             agent_llm=agent_llm,
-            document_summary_llm=document_summary_llm,
             image_generation_config=image_generation_config,
             vision_llm_config=vision_llm_config,
         )
@@ -707,9 +702,6 @@ async def update_llm_preferences(
 
         # Get full config objects for response
         agent_llm = await _get_llm_config_by_id(session, search_space.agent_llm_id)
-        document_summary_llm = await _get_llm_config_by_id(
-            session, search_space.document_summary_llm_id
-        )
         image_generation_config = await _get_image_gen_config_by_id(
             session, search_space.image_generation_config_id
         )
@@ -719,11 +711,9 @@ async def update_llm_preferences(
 
         return LLMPreferencesRead(
             agent_llm_id=search_space.agent_llm_id,
-            document_summary_llm_id=search_space.document_summary_llm_id,
             image_generation_config_id=search_space.image_generation_config_id,
             vision_llm_config_id=search_space.vision_llm_config_id,
             agent_llm=agent_llm,
-            document_summary_llm=document_summary_llm,
             image_generation_config=image_generation_config,
             vision_llm_config=vision_llm_config,
         )

surfsense_backend/app/schemas/new_llm_config.py

@@ -221,9 +221,6 @@ class LLMPreferencesRead(BaseModel):
     agent_llm_id: int | None = Field(
         None, description="ID of the LLM config to use for agent/chat tasks"
     )
-    document_summary_llm_id: int | None = Field(
-        None, description="ID of the LLM config to use for document summarization"
-    )
     image_generation_config_id: int | None = Field(
         None, description="ID of the image generation config to use"
     )
@@ -234,9 +231,6 @@ class LLMPreferencesRead(BaseModel):
     agent_llm: dict[str, Any] | None = Field(
         None, description="Full config for agent LLM"
     )
-    document_summary_llm: dict[str, Any] | None = Field(
-        None, description="Full config for document summary LLM"
-    )
     image_generation_config: dict[str, Any] | None = Field(
         None, description="Full config for image generation"
     )
@@ -253,9 +247,6 @@ class LLMPreferencesUpdate(BaseModel):
     agent_llm_id: int | None = Field(
         None, description="ID of the LLM config to use for agent/chat tasks"
     )
-    document_summary_llm_id: int | None = Field(
-        None, description="ID of the LLM config to use for document summarization"
-    )
     image_generation_config_id: int | None = Field(
         None, description="ID of the image generation config to use"
     )

surfsense_backend/app/schemas/search_source_connector.py

@@ -16,7 +16,6 @@ class SearchSourceConnectorBase(BaseModel):
     is_indexable: bool
     last_indexed_at: datetime | None = None
     config: dict[str, Any]
-    enable_summary: bool = False
     enable_vision_llm: bool = False
     periodic_indexing_enabled: bool = False
     indexing_frequency_minutes: int | None = None
@@ -67,7 +66,6 @@ class SearchSourceConnectorUpdate(BaseModel):
     is_indexable: bool | None = None
     last_indexed_at: datetime | None = None
     config: dict[str, Any] | None = None
-    enable_summary: bool | None = None
     enable_vision_llm: bool | None = None
     periodic_indexing_enabled: bool | None = None
     indexing_frequency_minutes: int | None = None

surfsense_backend/app/services/confluence/kb_sync_service.py

@@ -9,7 +9,6 @@ from app.utils.document_converters import (
     create_document_chunks,
     embed_text,
     generate_content_hash,
-    generate_document_summary,
     generate_unique_identifier_hash,
 )
 
@@ -65,29 +64,11 @@ class ConfluenceKBSyncService:
             if dup:
                 content_hash = unique_hash
 
-            from app.services.llm_service import get_user_long_context_llm
 
-            user_llm = await get_user_long_context_llm(
-                self.db_session,
-                user_id,
-                search_space_id,
-                disable_streaming=True,
-            )
 
-            doc_metadata_for_summary = {
-                "page_title": page_title,
-                "space_id": space_id,
-                "document_type": "Confluence Page",
-                "connector_type": "Confluence",
-            }
 
-            if user_llm:
+            summary_content = f"Confluence Page: {page_title}\n\n{page_content}"
-                summary_content, summary_embedding = await generate_document_summary(
+            summary_embedding = embed_text(summary_content)
-                    page_content, user_llm, doc_metadata_for_summary
-                )
-            else:
-                summary_content = f"Confluence Page: {page_title}\n\n{page_content}"
-                summary_embedding = embed_text(summary_content)
 
             chunks = await create_document_chunks(page_content)
             now_str = datetime.now().strftime("%Y-%m-%d %H:%M:%S")
@@ -185,25 +166,10 @@ class ConfluenceKBSyncService:
 
             space_id = (document.document_metadata or {}).get("space_id", "")
 
-            from app.services.llm_service import get_user_long_context_llm
 
-            user_llm = await get_user_long_context_llm(
-                self.db_session, user_id, search_space_id, disable_streaming=True
-            )
 
-            if user_llm:
+            summary_content = f"Confluence Page: {page_title}\n\n{page_content}"
-                doc_meta = {
+            summary_embedding = embed_text(summary_content)
-                    "page_title": page_title,
-                    "space_id": space_id,
-                    "document_type": "Confluence Page",
-                    "connector_type": "Confluence",
-                }
-                summary_content, summary_embedding = await generate_document_summary(
-                    page_content, user_llm, doc_meta
-                )
-            else:
-                summary_content = f"Confluence Page: {page_title}\n\n{page_content}"
-                summary_embedding = embed_text(summary_content)
 
             chunks = await create_document_chunks(page_content)
 

surfsense_backend/app/services/docling_service.py

@@ -191,149 +191,6 @@ class DoclingService:
             logger.error(f"Full traceback: {traceback.format_exc()}")
             raise RuntimeError(f"Docling processing failed: {e}") from e
 
-    async def process_large_document_summary(
-        self, content: str, llm, document_title: str = "Document"
-    ) -> str:
-        """
-        Process large documents using chunked LLM summarization.
-
-        Args:
-            content: The full document content
-            llm: The language model to use for summarization
-            document_title: Title of the document for context
-
-        Returns:
-            Final summary of the document
-        """
-        # Large document threshold (100K characters ≈ 25K tokens)
-        large_document_threshold = 100_000
-
-        if len(content) <= large_document_threshold:
-            # For smaller documents, use direct processing
-            logger.info(
-                f"📄 Document size: {len(content)} chars - using direct processing"
-            )
-            from app.prompts import SUMMARY_PROMPT_TEMPLATE
-
-            summary_chain = SUMMARY_PROMPT_TEMPLATE | llm
-            result = await summary_chain.ainvoke({"document": content})
-            return result.content
-
-        logger.info(
-            f"📚 Large document detected: {len(content)} chars - using chunked processing"
-        )
-
-        # Import chunker from config
-        # Create LLM-optimized chunks (8K tokens max for safety)
-        from chonkie import OverlapRefinery, RecursiveChunker
-        from langchain_core.prompts import PromptTemplate
-
-        llm_chunker = RecursiveChunker(
-            chunk_size=8000  # Conservative for most LLMs
-        )
-
-        # Apply overlap refinery for context preservation (10% overlap = 800 tokens)
-        overlap_refinery = OverlapRefinery(
-            context_size=0.1,  # 10% overlap for context preservation
-            method="suffix",  # Add next chunk context to current chunk
-        )
-
-        # First chunk the content, then apply overlap refinery
-        initial_chunks = llm_chunker.chunk(content)
-        chunks = overlap_refinery.refine(initial_chunks)
-        total_chunks = len(chunks)
-
-        logger.info(f"📄 Split into {total_chunks} chunks for LLM processing")
-
-        # Template for chunk processing
-        chunk_template = PromptTemplate(
-            input_variables=["chunk", "chunk_number", "total_chunks"],
-            template="""<INSTRUCTIONS>
-You are summarizing chunk {chunk_number} of {total_chunks} from a large document.
-
-Create a comprehensive summary of this document chunk. Focus on:
-- Key concepts, facts, and information
-- Important details and context
-- Main topics and themes
-
-Provide a clear, structured summary that captures the essential content.
-
-Chunk {chunk_number}/{total_chunks}:
-<document_chunk>
-{chunk}
-</document_chunk>
-</INSTRUCTIONS>""",
-        )
-
-        # Process each chunk individually
-        chunk_summaries = []
-        for i, chunk in enumerate(chunks, 1):
-            try:
-                logger.info(
-                    f"🔄 Processing chunk {i}/{total_chunks} ({len(chunk.text)} chars)"
-                )
-
-                chunk_chain = chunk_template | llm
-                chunk_result = await chunk_chain.ainvoke(
-                    {
-                        "chunk": chunk.text,
-                        "chunk_number": i,
-                        "total_chunks": total_chunks,
-                    }
-                )
-
-                chunk_summary = chunk_result.content
-                chunk_summaries.append(f"=== Section {i} ===\n{chunk_summary}")
-
-                logger.info(f"✅ Completed chunk {i}/{total_chunks}")
-
-            except Exception as e:
-                logger.error(f"❌ Failed to process chunk {i}/{total_chunks}: {e}")
-                chunk_summaries.append(f"=== Section {i} ===\n[Processing failed]")
-
-        # Combine summaries into final document summary
-        logger.info(f"🔄 Combining {len(chunk_summaries)} chunk summaries")
-
-        try:
-            combine_template = PromptTemplate(
-                input_variables=["summaries", "document_title"],
-                template="""<INSTRUCTIONS>
-You are combining multiple section summaries into a final comprehensive document summary.
-
-Create a unified, coherent summary from the following section summaries of "{document_title}".
-Ensure:
-- Logical flow and organization
-- No redundancy or repetition  
-- Comprehensive coverage of all key points
-- Professional, objective tone
-
-<section_summaries>
-{summaries}
-</section_summaries>
-</INSTRUCTIONS>""",
-            )
-
-            combined_summaries = "\n\n".join(chunk_summaries)
-            combine_chain = combine_template | llm
-
-            final_result = await combine_chain.ainvoke(
-                {"summaries": combined_summaries, "document_title": document_title}
-            )
-
-            final_summary = final_result.content
-            logger.info(
-                f"✅ Large document processing complete: {len(final_summary)} chars summary"
-            )
-
-            return final_summary
-
-        except Exception as e:
-            logger.error(f"❌ Failed to combine summaries: {e}")
-            # Fallback: return concatenated chunk summaries
-            fallback_summary = "\n\n".join(chunk_summaries)
-            logger.warning("⚠️ Using fallback combined summary")
-            return fallback_summary
-
 
 def create_docling_service() -> DoclingService:
     """Create a Docling service instance."""

surfsense_backend/app/services/dropbox/kb_sync_service.py

@@ -9,7 +9,6 @@ from app.utils.document_converters import (
     create_document_chunks,
     embed_text,
     generate_content_hash,
-    generate_document_summary,
 )
 
 logger = logging.getLogger(__name__)
@@ -72,29 +71,11 @@ class DropboxKBSyncService:
                 )
                 content_hash = unique_hash
 
-            from app.services.llm_service import get_user_long_context_llm
 
-            user_llm = await get_user_long_context_llm(
-                self.db_session,
-                user_id,
-                search_space_id,
-                disable_streaming=True,
-            )
 
-            doc_metadata_for_summary = {
-                "file_name": file_name,
-                "document_type": "Dropbox File",
-                "connector_type": "Dropbox",
-            }
 
-            if user_llm:
+            summary_content = f"Dropbox File: {file_name}\n\n{indexable_content}"
-                summary_content, summary_embedding = await generate_document_summary(
+            summary_embedding = embed_text(summary_content)
-                    indexable_content, user_llm, doc_metadata_for_summary
-                )
-            else:
-                logger.warning("No LLM configured — using fallback summary")
-                summary_content = f"Dropbox File: {file_name}\n\n{indexable_content}"
-                summary_embedding = embed_text(summary_content)
 
             chunks = await create_document_chunks(indexable_content)
             now_str = datetime.now().strftime("%Y-%m-%d %H:%M:%S")

surfsense_backend/app/services/gmail/kb_sync_service.py

@@ -9,7 +9,6 @@ from app.utils.document_converters import (
     create_document_chunks,
     embed_text,
     generate_content_hash,
-    generate_document_summary,
     generate_unique_identifier_hash,
 )
 
@@ -78,30 +77,11 @@ class GmailKBSyncService:
                 )
                 content_hash = unique_hash
 
-            from app.services.llm_service import get_user_long_context_llm
 
-            user_llm = await get_user_long_context_llm(
-                self.db_session,
-                user_id,
-                search_space_id,
-                disable_streaming=True,
-            )
 
-            doc_metadata_for_summary = {
-                "subject": subject,
-                "sender": sender,
-                "document_type": "Gmail Message",
-                "connector_type": "Gmail",
-            }
 
-            if user_llm:
+            summary_content = f"Gmail Message: {subject}\n\n{indexable_content}"
-                summary_content, summary_embedding = await generate_document_summary(
+            summary_embedding = await asyncio.to_thread(embed_text, summary_content)
-                    indexable_content, user_llm, doc_metadata_for_summary
-                )
-            else:
-                logger.warning("No LLM configured -- using fallback summary")
-                summary_content = f"Gmail Message: {subject}\n\n{indexable_content}"
-                summary_embedding = await asyncio.to_thread(embed_text, summary_content)
 
             chunks = await create_document_chunks(indexable_content)
             now_str = datetime.now().strftime("%Y-%m-%d %H:%M:%S")

surfsense_backend/app/services/google_calendar/kb_sync_service.py

@@ -19,7 +19,6 @@ from app.utils.document_converters import (
     create_document_chunks,
     embed_text,
     generate_content_hash,
-    generate_document_summary,
     generate_unique_identifier_hash,
 )
 
@@ -90,33 +89,13 @@ class GoogleCalendarKBSyncService:
                 )
                 content_hash = unique_hash
 
-            from app.services.llm_service import get_user_long_context_llm
 
-            user_llm = await get_user_long_context_llm(
+
-                self.db_session,
+
-                user_id,
+            summary_content = (
-                search_space_id,
+                f"Google Calendar Event: {event_summary}\n\n{indexable_content}"
-                disable_streaming=True,
             )
-
+            summary_embedding = await asyncio.to_thread(embed_text, summary_content)
-            doc_metadata_for_summary = {
-                "event_summary": event_summary,
-                "start_time": start_time,
-                "end_time": end_time,
-                "document_type": "Google Calendar Event",
-                "connector_type": "Google Calendar",
-            }
-
-            if user_llm:
-                summary_content, summary_embedding = await generate_document_summary(
-                    indexable_content, user_llm, doc_metadata_for_summary
-                )
-            else:
-                logger.warning("No LLM configured -- using fallback summary")
-                summary_content = (
-                    f"Google Calendar Event: {event_summary}\n\n{indexable_content}"
-                )
-                summary_embedding = await asyncio.to_thread(embed_text, summary_content)
 
             chunks = await create_document_chunks(indexable_content)
             now_str = datetime.now().strftime("%Y-%m-%d %H:%M:%S")
@@ -273,29 +252,13 @@ class GoogleCalendarKBSyncService:
             if not indexable_content:
                 return {"status": "error", "message": "Event produced empty content"}
 
-            from app.services.llm_service import get_user_long_context_llm
 
-            user_llm = await get_user_long_context_llm(
+
-                self.db_session, user_id, search_space_id, disable_streaming=True
+
+            summary_content = (
+                f"Google Calendar Event: {event_summary}\n\n{indexable_content}"
             )
-
+            summary_embedding = await asyncio.to_thread(embed_text, summary_content)
-            doc_metadata_for_summary = {
-                "event_summary": event_summary,
-                "start_time": start_time,
-                "end_time": end_time,
-                "document_type": "Google Calendar Event",
-                "connector_type": "Google Calendar",
-            }
-
-            if user_llm:
-                summary_content, summary_embedding = await generate_document_summary(
-                    indexable_content, user_llm, doc_metadata_for_summary
-                )
-            else:
-                summary_content = (
-                    f"Google Calendar Event: {event_summary}\n\n{indexable_content}"
-                )
-                summary_embedding = await asyncio.to_thread(embed_text, summary_content)
 
             chunks = await create_document_chunks(indexable_content)
             now_str = datetime.now().strftime("%Y-%m-%d %H:%M:%S")

surfsense_backend/app/services/google_drive/kb_sync_service.py

@@ -8,7 +8,6 @@ from app.utils.document_converters import (
     create_document_chunks,
     embed_text,
     generate_content_hash,
-    generate_document_summary,
     generate_unique_identifier_hash,
 )
 
@@ -74,32 +73,13 @@ class GoogleDriveKBSyncService:
                 )
                 content_hash = unique_hash
 
-            from app.services.llm_service import get_user_long_context_llm
 
-            user_llm = await get_user_long_context_llm(
+
-                self.db_session,
+
-                user_id,
+            summary_content = (
-                search_space_id,
+                f"Google Drive File: {file_name}\n\n{indexable_content}"
-                disable_streaming=True,
             )
-
+            summary_embedding = embed_text(summary_content)
-            doc_metadata_for_summary = {
-                "file_name": file_name,
-                "mime_type": mime_type,
-                "document_type": "Google Drive File",
-                "connector_type": "Google Drive",
-            }
-
-            if user_llm:
-                summary_content, summary_embedding = await generate_document_summary(
-                    indexable_content, user_llm, doc_metadata_for_summary
-                )
-            else:
-                logger.warning("No LLM configured — using fallback summary")
-                summary_content = (
-                    f"Google Drive File: {file_name}\n\n{indexable_content}"
-                )
-                summary_embedding = embed_text(summary_content)
 
             chunks = await create_document_chunks(indexable_content)
             now_str = datetime.now().strftime("%Y-%m-%d %H:%M:%S")

surfsense_backend/app/services/linear/kb_sync_service.py

@@ -9,7 +9,6 @@ from app.utils.document_converters import (
     create_document_chunks,
     embed_text,
     generate_content_hash,
-    generate_document_summary,
     generate_unique_identifier_hash,
 )
 
@@ -84,32 +83,13 @@ class LinearKBSyncService:
                 )
                 content_hash = unique_hash
 
-            from app.services.llm_service import get_user_long_context_llm
 
-            user_llm = await get_user_long_context_llm(
+
-                self.db_session,
+
-                user_id,
+            summary_content = (
-                search_space_id,
+                f"Linear Issue {issue_identifier}: {issue_title}\n\n{issue_content}"
-                disable_streaming=True,
             )
-
+            summary_embedding = embed_text(summary_content)
-            doc_metadata_for_summary = {
-                "issue_id": issue_identifier,
-                "issue_title": issue_title,
-                "document_type": "Linear Issue",
-                "connector_type": "Linear",
-            }
-
-            if user_llm:
-                summary_content, summary_embedding = await generate_document_summary(
-                    issue_content, user_llm, doc_metadata_for_summary
-                )
-            else:
-                logger.warning("No LLM configured — using fallback summary")
-                summary_content = (
-                    f"Linear Issue {issue_identifier}: {issue_title}\n\n{issue_content}"
-                )
-                summary_embedding = embed_text(summary_content)
 
             chunks = await create_document_chunks(issue_content)
             now_str = datetime.now().strftime("%Y-%m-%d %H:%M:%S")
@@ -227,30 +207,12 @@ class LinearKBSyncService:
             comment_count = len(formatted_issue.get("comments", []))
             formatted_issue.get("description", "")
 
-            from app.services.llm_service import get_user_long_context_llm
 
-            user_llm = await get_user_long_context_llm(
+
-                self.db_session, user_id, search_space_id, disable_streaming=True
+            summary_content = (
+                f"Linear Issue {issue_identifier}: {issue_title}\n\n{issue_content}"
             )
-
+            summary_embedding = embed_text(summary_content)
-            if user_llm:
-                document_metadata_for_summary = {
-                    "issue_id": issue_identifier,
-                    "issue_title": issue_title,
-                    "state": state,
-                    "priority": priority,
-                    "comment_count": comment_count,
-                    "document_type": "Linear Issue",
-                    "connector_type": "Linear",
-                }
-                summary_content, summary_embedding = await generate_document_summary(
-                    issue_content, user_llm, document_metadata_for_summary
-                )
-            else:
-                summary_content = (
-                    f"Linear Issue {issue_identifier}: {issue_title}\n\n{issue_content}"
-                )
-                summary_embedding = embed_text(summary_content)
 
             chunks = await create_document_chunks(issue_content)
 

surfsense_backend/app/services/llm_service.py

@@ -68,7 +68,6 @@ def _is_interactive_auth_provider(
 
 class LLMRole:
     AGENT = "agent"  # For agent/chat operations
-    DOCUMENT_SUMMARY = "document_summary"  # For document summarization
 
 
 def get_global_llm_config(llm_config_id: int) -> dict | None:
@@ -266,7 +265,7 @@ async def get_search_space_llm_instance(
     Args:
         session: Database session
         search_space_id: Search Space ID
-        role: LLM role ('agent' or 'document_summary')
+        role: LLM role ('agent')
 
     Returns:
         ChatLiteLLM or ChatLiteLLMRouter instance, or None if not found
@@ -283,11 +282,8 @@ async def get_search_space_llm_instance(
             return None
 
         # Get the appropriate LLM config ID based on role
-        llm_config_id = None
         if role == LLMRole.AGENT:
             llm_config_id = search_space.agent_llm_id
-        elif role == LLMRole.DOCUMENT_SUMMARY:
-            llm_config_id = search_space.document_summary_llm_id
         else:
             logger.error(f"Invalid LLM role: {role}")
             return None
@@ -470,20 +466,13 @@ async def get_search_space_llm_instance(
 
 
 async def get_agent_llm(
-    session: AsyncSession, search_space_id: int
-) -> ChatLiteLLM | ChatLiteLLMRouter | None:
-    """Get the search space's agent LLM instance for chat operations."""
-    return await get_search_space_llm_instance(session, search_space_id, LLMRole.AGENT)
-
-
-async def get_document_summary_llm(
     session: AsyncSession, search_space_id: int, disable_streaming: bool = False
 ) -> ChatLiteLLM | ChatLiteLLMRouter | None:
-    """Get the search space's document summary LLM instance."""
+    """Get the search space's agent LLM instance for chat operations."""
     return await get_search_space_llm_instance(
         session,
         search_space_id,
-        LLMRole.DOCUMENT_SUMMARY,
+        LLMRole.AGENT,
         disable_streaming=disable_streaming,
     )
 
@@ -645,22 +634,6 @@ async def get_vision_llm(
         return None
 
 
-# Backward-compatible alias (LLM preferences are now per-search-space, not per-user)
-async def get_user_long_context_llm(
-    session: AsyncSession,
-    user_id: str,
-    search_space_id: int,
-    disable_streaming: bool = False,
-) -> ChatLiteLLM | ChatLiteLLMRouter | None:
-    """
-    Deprecated: Use get_document_summary_llm instead.
-    The user_id parameter is ignored as LLM preferences are now per-search-space.
-    """
-    return await get_document_summary_llm(
-        session, search_space_id, disable_streaming=disable_streaming
-    )
-
-
 def get_planner_llm() -> ChatLiteLLM | None:
     """Return a planner LLM instance from the first global config marked
     ``is_planner: true``, or ``None`` if no planner config is defined.

surfsense_backend/app/services/notion/kb_sync_service.py

@@ -8,7 +8,6 @@ from app.utils.document_converters import (
     create_document_chunks,
     embed_text,
     generate_content_hash,
-    generate_document_summary,
     generate_unique_identifier_hash,
 )
 
@@ -73,30 +72,11 @@ class NotionKBSyncService:
                 )
                 content_hash = unique_hash
 
-            from app.services.llm_service import get_user_long_context_llm
 
-            user_llm = await get_user_long_context_llm(
-                self.db_session,
-                user_id,
-                search_space_id,
-                disable_streaming=True,
-            )
 
-            doc_metadata_for_summary = {
-                "page_title": page_title,
-                "page_id": page_id,
-                "document_type": "Notion Page",
-                "connector_type": "Notion",
-            }
 
-            if user_llm:
+            summary_content = f"Notion Page: {page_title}\n\n{markdown_content}"
-                summary_content, summary_embedding = await generate_document_summary(
+            summary_embedding = embed_text(summary_content)
-                    markdown_content, user_llm, doc_metadata_for_summary
-                )
-            else:
-                logger.warning("No LLM configured — using fallback summary")
-                summary_content = f"Notion Page: {page_title}\n\n{markdown_content}"
-                summary_embedding = embed_text(summary_content)
 
             chunks = await create_document_chunks(markdown_content)
             now_str = datetime.now().strftime("%Y-%m-%d %H:%M:%S")
@@ -245,31 +225,11 @@ class NotionKBSyncService:
                 f"Final content length: {len(full_content)} chars, verified={content_verified}"
             )
 
-            from app.services.llm_service import get_user_long_context_llm
 
             logger.debug("Generating summary and embeddings")
-            user_llm = await get_user_long_context_llm(
-                self.db_session,
-                user_id,
-                search_space_id,
-                disable_streaming=True,  # disable streaming to avoid leaking into the chat
-            )
 
-            if user_llm:
+            summary_content = f"Notion Page: {document.document_metadata.get('page_title')}\n\n{full_content}"
-                document_metadata_for_summary = {
+            summary_embedding = embed_text(summary_content)
-                    "page_title": document.document_metadata.get("page_title"),
-                    "page_id": document.document_metadata.get("page_id"),
-                    "document_type": "Notion Page",
-                    "connector_type": "Notion",
-                }
-                summary_content, summary_embedding = await generate_document_summary(
-                    full_content, user_llm, document_metadata_for_summary
-                )
-                logger.debug(f"Generated summary length: {len(summary_content)} chars")
-            else:
-                logger.warning("No LLM configured - using fallback summary")
-                summary_content = f"Notion Page: {document.document_metadata.get('page_title')}\n\n{full_content}"
-                summary_embedding = embed_text(summary_content)
 
             logger.debug("Creating new chunks")
             chunks = await create_document_chunks(full_content)

surfsense_backend/app/services/obsidian_plugin_indexer.py

@@ -233,18 +233,6 @@ async def _resolve_attachment_vision_llm(
     return await get_vision_llm(session, search_space_id)
 
 
-async def _resolve_summary_llm(
-    session: AsyncSession, *, user_id: str, search_space_id: int, should_summarize: bool
-):
-    """Fetch summary LLM only when indexing summary is enabled."""
-    if not should_summarize:
-        return None
-
-    from app.services.llm_service import get_user_long_context_llm
-
-    return await get_user_long_context_llm(session, user_id, search_space_id)
-
-
 def _require_extracted_attachment_content(
     *, content: str, etl_meta: dict[str, Any], path: str
 ) -> str:
@@ -349,13 +337,6 @@ async def upsert_note(
             path=payload.path,
         )
 
-    llm = await _resolve_summary_llm(
-        session,
-        user_id=str(user_id),
-        search_space_id=search_space_id,
-        should_summarize=connector.enable_summary,
-    )
-
     document_string = _build_document_string(
         payload, vault_name, content_override=content_for_index
     )
@@ -374,8 +355,6 @@ async def upsert_note(
         search_space_id=search_space_id,
         connector_id=connector.id,
         created_by_id=str(user_id),
-        should_summarize=connector.enable_summary,
-        fallback_summary=f"Obsidian Note: {payload.name}\n\n{content_for_index}",
         metadata=metadata,
     )
 
@@ -388,7 +367,7 @@ async def upsert_note(
 
     document = prepared[0]
 
-    return await pipeline.index(document, connector_doc, llm)
+    return await pipeline.index(document, connector_doc)
 
 
 async def rename_note(

surfsense_backend/app/services/onedrive/kb_sync_service.py

@@ -10,7 +10,6 @@ from app.utils.document_converters import (
     create_document_chunks,
     embed_text,
     generate_content_hash,
-    generate_document_summary,
 )
 
 logger = logging.getLogger(__name__)
@@ -73,30 +72,11 @@ class OneDriveKBSyncService:
                 )
                 content_hash = unique_hash
 
-            from app.services.llm_service import get_user_long_context_llm
 
-            user_llm = await get_user_long_context_llm(
-                self.db_session,
-                user_id,
-                search_space_id,
-                disable_streaming=True,
-            )
 
-            doc_metadata_for_summary = {
-                "file_name": file_name,
-                "mime_type": mime_type,
-                "document_type": "OneDrive File",
-                "connector_type": "OneDrive",
-            }
 
-            if user_llm:
+            summary_content = f"OneDrive File: {file_name}\n\n{indexable_content}"
-                summary_content, summary_embedding = await generate_document_summary(
+            summary_embedding = await asyncio.to_thread(embed_text, summary_content)
-                    indexable_content, user_llm, doc_metadata_for_summary
-                )
-            else:
-                logger.warning("No LLM configured — using fallback summary")
-                summary_content = f"OneDrive File: {file_name}\n\n{indexable_content}"
-                summary_embedding = await asyncio.to_thread(embed_text, summary_content)
 
             chunks = await create_document_chunks(indexable_content)
             now_str = datetime.now().strftime("%Y-%m-%d %H:%M:%S")

surfsense_backend/app/services/task_dispatcher.py

@@ -18,7 +18,6 @@ class TaskDispatcher(Protocol):
         filename: str,
         search_space_id: int,
         user_id: str,
-        should_summarize: bool = False,
         use_vision_llm: bool = False,
         processing_mode: str = "basic",
     ) -> None: ...
@@ -35,7 +34,6 @@ class CeleryTaskDispatcher:
         filename: str,
         search_space_id: int,
         user_id: str,
-        should_summarize: bool = False,
         use_vision_llm: bool = False,
         processing_mode: str = "basic",
     ) -> None:
@@ -49,7 +47,6 @@ class CeleryTaskDispatcher:
             filename=filename,
             search_space_id=search_space_id,
             user_id=user_id,
-            should_summarize=should_summarize,
             use_vision_llm=use_vision_llm,
             processing_mode=processing_mode,
         )

surfsense_backend/app/tasks/celery_tasks/document_reindex_tasks.py

@@ -9,7 +9,6 @@ from sqlalchemy.orm import selectinload
 from app.celery_app import celery_app
 from app.db import Document
 from app.indexing_pipeline.adapters.file_upload_adapter import UploadDocumentAdapter
-from app.services.llm_service import get_user_long_context_llm
 from app.services.task_logging_service import TaskLoggingService
 from app.tasks.celery_tasks import get_celery_session_maker, run_async_celery_task
 
@@ -68,12 +67,8 @@ async def _reindex_document(document_id: int, user_id: str):
 
             logger.info(f"Reindexing document {document_id} ({document.title})")
 
-            user_llm = await get_user_long_context_llm(
-                session, user_id, document.search_space_id
-            )
-
             adapter = UploadDocumentAdapter(session)
-            await adapter.reindex(document=document, llm=user_llm)
+            await adapter.reindex(document=document)
 
             await task_logger.log_task_success(
                 log_entry,

surfsense_backend/app/tasks/celery_tasks/document_tasks.py

@@ -765,7 +765,6 @@ def process_file_upload_with_document_task(
     filename: str,
     search_space_id: int,
     user_id: str,
-    should_summarize: bool = False,
     use_vision_llm: bool = False,
     processing_mode: str = "basic",
 ):
@@ -782,7 +781,6 @@ def process_file_upload_with_document_task(
         filename: Original filename
         search_space_id: ID of the search space
         user_id: ID of the user
-        should_summarize: Whether to generate an LLM summary
     """
     import traceback
 
@@ -814,7 +812,6 @@ def process_file_upload_with_document_task(
                 filename,
                 search_space_id,
                 user_id,
-                should_summarize=should_summarize,
                 use_vision_llm=use_vision_llm,
                 processing_mode=processing_mode,
             )
@@ -850,7 +847,6 @@ async def _process_file_with_document(
     filename: str,
     search_space_id: int,
     user_id: str,
-    should_summarize: bool = False,
     use_vision_llm: bool = False,
     processing_mode: str = "basic",
 ):
@@ -954,7 +950,6 @@ async def _process_file_with_document(
                 task_logger=task_logger,
                 log_entry=log_entry,
                 notification=notification,
-                should_summarize=should_summarize,
                 use_vision_llm=use_vision_llm,
                 processing_mode=processing_mode,
             )
@@ -1258,7 +1253,6 @@ def index_local_folder_task(
     exclude_patterns: list[str] | None = None,
     file_extensions: list[str] | None = None,
     root_folder_id: int | None = None,
-    enable_summary: bool = False,
     target_file_paths: list[str] | None = None,
 ):
     """Celery task to index a local folder. Config is passed directly — no connector row."""
@@ -1271,7 +1265,6 @@ def index_local_folder_task(
             exclude_patterns=exclude_patterns,
             file_extensions=file_extensions,
             root_folder_id=root_folder_id,
-            enable_summary=enable_summary,
             target_file_paths=target_file_paths,
         )
     )
@@ -1285,7 +1278,6 @@ async def _index_local_folder_async(
     exclude_patterns: list[str] | None = None,
     file_extensions: list[str] | None = None,
     root_folder_id: int | None = None,
-    enable_summary: bool = False,
     target_file_paths: list[str] | None = None,
 ):
     """Run local folder indexing with notification + heartbeat."""
@@ -1343,8 +1335,7 @@ async def _index_local_folder_async(
                 exclude_patterns=exclude_patterns,
                 file_extensions=file_extensions,
                 root_folder_id=root_folder_id,
-                enable_summary=enable_summary,
+                    target_file_paths=target_file_paths,
-                target_file_paths=target_file_paths,
                 on_heartbeat_callback=_heartbeat_progress
                 if (is_batch or is_full_scan)
                 else None,
@@ -1400,7 +1391,6 @@ def index_uploaded_folder_files_task(
     user_id: str,
     folder_name: str,
     root_folder_id: int,
-    enable_summary: bool,
     file_mappings: list[dict],
     use_vision_llm: bool = False,
     processing_mode: str = "basic",
@@ -1412,7 +1402,6 @@ def index_uploaded_folder_files_task(
             user_id=user_id,
             folder_name=folder_name,
             root_folder_id=root_folder_id,
-            enable_summary=enable_summary,
             file_mappings=file_mappings,
             use_vision_llm=use_vision_llm,
             processing_mode=processing_mode,
@@ -1425,7 +1414,6 @@ async def _index_uploaded_folder_files_async(
     user_id: str,
     folder_name: str,
     root_folder_id: int,
-    enable_summary: bool,
     file_mappings: list[dict],
     use_vision_llm: bool = False,
     processing_mode: str = "basic",
@@ -1475,8 +1463,7 @@ async def _index_uploaded_folder_files_async(
                 user_id=user_id,
                 folder_name=folder_name,
                 root_folder_id=root_folder_id,
-                enable_summary=enable_summary,
+                    file_mappings=file_mappings,
-                file_mappings=file_mappings,
                 on_heartbeat_callback=_heartbeat_progress,
                 use_vision_llm=use_vision_llm,
                 processing_mode=processing_mode,
@@ -1563,12 +1550,10 @@ async def _ai_sort_search_space_async(search_space_id: int, user_id: str):
     t_start = time.perf_counter()
     try:
         from app.services.ai_file_sort_service import ai_sort_all_documents
-        from app.services.llm_service import get_document_summary_llm
+        from app.services.llm_service import get_agent_llm
 
         async with get_celery_session_maker()() as session:
-            llm = await get_document_summary_llm(
+            llm = await get_agent_llm(session, search_space_id, disable_streaming=True)
-                session, search_space_id, disable_streaming=True
-            )
             if llm is None:
                 logger.warning(
                     "No LLM configured for search_space=%d, skipping AI sort",
@@ -1604,7 +1589,7 @@ def ai_sort_document_task(self, search_space_id: int, user_id: str, document_id:
 async def _ai_sort_document_async(search_space_id: int, user_id: str, document_id: int):
     from app.db import Document
     from app.services.ai_file_sort_service import ai_sort_document
-    from app.services.llm_service import get_document_summary_llm
+    from app.services.llm_service import get_agent_llm
 
     async with get_celery_session_maker()() as session:
         document = await session.get(Document, document_id)
@@ -1612,9 +1597,7 @@ async def _ai_sort_document_async(search_space_id: int, user_id: str, document_i
             logger.warning("Document %d not found, skipping AI sort", document_id)
             return
 
-        llm = await get_document_summary_llm(
+        llm = await get_agent_llm(session, search_space_id, disable_streaming=True)
-            session, search_space_id, disable_streaming=True
-        )
         if llm is None:
             logger.warning(
                 "No LLM for search_space=%d, skipping AI sort of doc=%d",

surfsense_backend/app/tasks/chat/streaming/flows/new_chat/input_state.py

@@ -62,6 +62,7 @@ async def build_new_chat_input_state(
     user_image_data_urls: list[str] | None,
     mentioned_document_ids: list[int] | None,
     mentioned_folder_ids: list[int] | None,
+    mentioned_connectors: list[dict[str, Any]] | None,
     mentioned_documents: list[dict[str, Any]] | None,
     needs_history_bootstrap: bool,
     thread_visibility: ChatVisibility,
@@ -110,6 +111,7 @@ async def build_new_chat_input_state(
 
     final_query = _render_query_with_context(
         agent_user_query=agent_user_query,
+        mentioned_connectors=mentioned_connectors,
         recent_reports=recent_reports,
     )
 
@@ -196,11 +198,16 @@ async def _resolve_mentions_for_query(
 def _render_query_with_context(
     *,
     agent_user_query: str,
+    mentioned_connectors: list[dict[str, Any]] | None,
     recent_reports: list[Report],
 ) -> str:
-    """Prepend recent-reports XML block to the user query."""
+    """Prepend connector/report XML context blocks to the user query."""
     context_parts: list[str] = []
 
+    connector_context = _render_mentioned_connectors(mentioned_connectors)
+    if connector_context:
+        context_parts.append(connector_context)
+
     if recent_reports:
         report_lines: list[str] = []
         for r in recent_reports:
@@ -225,3 +232,40 @@ def _render_query_with_context(
         return f"{context}\n\n<user_query>{agent_user_query}</user_query>"
 
     return agent_user_query
+
+
+def _render_mentioned_connectors(
+    mentioned_connectors: list[dict[str, Any]] | None,
+) -> str | None:
+    """Render selected connector account metadata for connector-backed tools."""
+    if not mentioned_connectors:
+        return None
+
+    connector_lines: list[str] = []
+    for connector in mentioned_connectors:
+        if not isinstance(connector, dict):
+            continue
+        connector_id = connector.get("id")
+        connector_type = connector.get("connector_type") or connector.get(
+            "document_type"
+        )
+        account_name = connector.get("account_name") or connector.get("title")
+        if connector_id is None or connector_type is None:
+            continue
+        connector_lines.append(
+            f'  - connector_id={connector_id}, connector_type="{connector_type}", '
+            f'account_name="{account_name or ""}"'
+        )
+
+    if not connector_lines:
+        return None
+
+    return (
+        "<mentioned_connectors>\n"
+        "The user selected these exact connector accounts with @. "
+        "These entries are selection metadata, not retrieved connector content. "
+        "When a connector-backed tool needs an account, use the matching "
+        "connector_id from this list if the tool supports connector_id:\n"
+        + "\n".join(connector_lines)
+        + "\n</mentioned_connectors>"
+    )

surfsense_backend/app/tasks/chat/streaming/flows/new_chat/orchestrator.py

@@ -124,6 +124,8 @@ async def stream_new_chat(
     llm_config_id: int = -1,
     mentioned_document_ids: list[int] | None = None,
     mentioned_folder_ids: list[int] | None = None,
+    mentioned_connector_ids: list[int] | None = None,
+    mentioned_connectors: list[dict[str, Any]] | None = None,
     mentioned_documents: list[dict[str, Any]] | None = None,
     checkpoint_id: str | None = None,
     needs_history_bootstrap: bool = False,
@@ -435,6 +437,7 @@ async def stream_new_chat(
             user_image_data_urls=user_image_data_urls,
             mentioned_document_ids=mentioned_document_ids,
             mentioned_folder_ids=mentioned_folder_ids,
+            mentioned_connectors=mentioned_connectors,
             mentioned_documents=mentioned_documents,
             needs_history_bootstrap=needs_history_bootstrap,
             thread_visibility=visibility,
@@ -588,6 +591,8 @@ async def stream_new_chat(
             mentioned_document_ids=mentioned_document_ids,
             accepted_folder_ids=accepted_folder_ids,
             mentioned_folder_ids=mentioned_folder_ids,
+            mentioned_connector_ids=mentioned_connector_ids,
+            mentioned_connectors=mentioned_connectors,
             request_id=request_id,
             turn_id=stream_result.turn_id,
         )

surfsense_backend/app/tasks/chat/streaming/flows/new_chat/runtime_context.py

@@ -8,6 +8,8 @@ mention lists / request ids / turn ids without rebuilding the graph.
 
 from __future__ import annotations
 
+from typing import Any
+
 from app.agents.new_chat.context import SurfSenseContextSchema
 
 
@@ -17,6 +19,8 @@ def build_new_chat_runtime_context(
     mentioned_document_ids: list[int] | None,
     accepted_folder_ids: list[int],
     mentioned_folder_ids: list[int] | None,
+    mentioned_connector_ids: list[int] | None,
+    mentioned_connectors: list[dict[str, Any]] | None,
     request_id: str | None,
     turn_id: str,
 ) -> SurfSenseContextSchema:
@@ -31,6 +35,8 @@ def build_new_chat_runtime_context(
         search_space_id=search_space_id,
         mentioned_document_ids=list(mentioned_document_ids or []),
         mentioned_folder_ids=list(accepted_folder_ids or mentioned_folder_ids or []),
+        mentioned_connector_ids=list(mentioned_connector_ids or []),
+        mentioned_connectors=list(mentioned_connectors or []),
         request_id=request_id,
         turn_id=turn_id,
     )

surfsense_backend/app/tasks/connector_indexers/airtable_indexer.py

@@ -14,13 +14,11 @@ from sqlalchemy.ext.asyncio import AsyncSession
 
 from app.connectors.airtable_history import AirtableHistoryConnector
 from app.db import Document, DocumentStatus, DocumentType, SearchSourceConnectorType
-from app.services.llm_service import get_user_long_context_llm
 from app.services.task_logging_service import TaskLoggingService
 from app.utils.document_converters import (
     create_document_chunks,
     embed_text,
     generate_content_hash,
-    generate_document_summary,
     generate_unique_identifier_hash,
 )
 
@@ -394,29 +392,10 @@ async def index_airtable_records(
                     document.status = DocumentStatus.processing()
                     await session.commit()
 
-                    # Heavy processing (LLM, embeddings, chunks)
+                    # Heavy processing (embeddings, chunks)
-                    user_llm = await get_user_long_context_llm(
-                        session, user_id, search_space_id
-                    )
 
-                    if user_llm and connector.enable_summary:
+                    summary_content = f"Airtable Record: {item['record_id']}\n\n{item['markdown_content']}"
-                        document_metadata_for_summary = {
+                    summary_embedding = embed_text(summary_content)
-                            "record_id": item["record_id"],
-                            "created_time": item["record"].get("CREATED_TIME()", ""),
-                            "document_type": "Airtable Record",
-                            "connector_type": "Airtable",
-                        }
-                        (
-                            summary_content,
-                            summary_embedding,
-                        ) = await generate_document_summary(
-                            item["markdown_content"],
-                            user_llm,
-                            document_metadata_for_summary,
-                        )
-                    else:
-                        summary_content = f"Airtable Record: {item['record_id']}\n\n{item['markdown_content']}"
-                        summary_embedding = embed_text(summary_content)
 
                     chunks = await create_document_chunks(item["markdown_content"])
 

surfsense_backend/app/tasks/connector_indexers/bookstack_indexer.py

@@ -15,13 +15,11 @@ from sqlalchemy.ext.asyncio import AsyncSession
 
 from app.connectors.bookstack_connector import BookStackConnector
 from app.db import Document, DocumentStatus, DocumentType, SearchSourceConnectorType
-from app.services.llm_service import get_user_long_context_llm
 from app.services.task_logging_service import TaskLoggingService
 from app.utils.document_converters import (
     create_document_chunks,
     embed_text,
     generate_content_hash,
-    generate_document_summary,
     generate_unique_identifier_hash,
 )
 
@@ -384,10 +382,7 @@ async def index_bookstack_pages(
                 document.status = DocumentStatus.processing()
                 await session.commit()
 
-                # Heavy processing (LLM, embeddings, chunks)
+                # Heavy processing (embeddings, chunks)
-                user_llm = await get_user_long_context_llm(
-                    session, user_id, search_space_id
-                )
 
                 # Build document metadata
                 doc_metadata = {
@@ -403,23 +398,8 @@ async def index_bookstack_pages(
                     "connector_id": connector_id,
                 }
 
-                if user_llm and connector.enable_summary:
+                summary_content = f"BookStack Page: {item['page_name']}\n\nBook ID: {item['book_id']}\n\n{item['full_content']}"
-                    summary_metadata = {
+                summary_embedding = embed_text(summary_content)
-                        "page_name": item["page_name"],
-                        "page_id": item["page_id"],
-                        "book_id": item["book_id"],
-                        "document_type": "BookStack Page",
-                        "connector_type": "BookStack",
-                    }
-                    (
-                        summary_content,
-                        summary_embedding,
-                    ) = await generate_document_summary(
-                        item["full_content"], user_llm, summary_metadata
-                    )
-                else:
-                    summary_content = f"BookStack Page: {item['page_name']}\n\nBook ID: {item['book_id']}\n\n{item['full_content']}"
-                    summary_embedding = embed_text(summary_content)
 
                 # Process chunks - using the full page content
                 chunks = await create_document_chunks(item["full_content"])

surfsense_backend/app/tasks/connector_indexers/clickup_indexer.py

@@ -16,13 +16,11 @@ from sqlalchemy.ext.asyncio import AsyncSession
 
 from app.connectors.clickup_history import ClickUpHistoryConnector
 from app.db import Document, DocumentStatus, DocumentType, SearchSourceConnectorType
-from app.services.llm_service import get_user_long_context_llm
 from app.services.task_logging_service import TaskLoggingService
 from app.utils.document_converters import (
     create_document_chunks,
     embed_text,
     generate_content_hash,
-    generate_document_summary,
     generate_unique_identifier_hash,
 )
 
@@ -393,32 +391,10 @@ async def index_clickup_tasks(
                 document.status = DocumentStatus.processing()
                 await session.commit()
 
-                # Heavy processing (LLM, embeddings, chunks)
+                # Heavy processing (embeddings, chunks)
-                user_llm = await get_user_long_context_llm(
-                    session, user_id, search_space_id
-                )
 
-                if user_llm and connector.enable_summary:
+                summary_content = item["task_content"]
-                    document_metadata_for_summary = {
+                summary_embedding = embed_text(item["task_content"])
-                        "task_id": item["task_id"],
-                        "task_name": item["task_name"],
-                        "task_status": item["task_status"],
-                        "task_priority": item["task_priority"],
-                        "task_list": item["task_list_name"],
-                        "task_space": item["task_space_name"],
-                        "assignees": len(item["task_assignees"]),
-                        "document_type": "ClickUp Task",
-                        "connector_type": "ClickUp",
-                    }
-                    (
-                        summary_content,
-                        summary_embedding,
-                    ) = await generate_document_summary(
-                        item["task_content"], user_llm, document_metadata_for_summary
-                    )
-                else:
-                    summary_content = item["task_content"]
-                    summary_embedding = embed_text(item["task_content"])
 
                 chunks = await create_document_chunks(item["task_content"])
 

surfsense_backend/app/tasks/connector_indexers/confluence_indexer.py

@@ -14,7 +14,6 @@ from app.indexing_pipeline.indexing_pipeline_service import (
     IndexingPipelineService,
     PlaceholderInfo,
 )
-from app.services.llm_service import get_user_long_context_llm
 from app.services.task_logging_service import TaskLoggingService
 
 from .base import (
@@ -36,7 +35,6 @@ def _build_connector_doc(
     connector_id: int,
     search_space_id: int,
     user_id: str,
-    enable_summary: bool,
 ) -> ConnectorDocument:
     """Map a raw Confluence page dict to a ConnectorDocument."""
     page_id = page.get("id", "")
@@ -54,10 +52,6 @@ def _build_connector_doc(
         "connector_type": "Confluence",
     }
 
-    fallback_summary = (
-        f"Confluence Page: {page_title}\n\nSpace ID: {space_id}\n\n{full_content}"
-    )
-
     return ConnectorDocument(
         title=page_title,
         source_markdown=full_content,
@@ -66,8 +60,6 @@ def _build_connector_doc(
         search_space_id=search_space_id,
         connector_id=connector_id,
         created_by_id=user_id,
-        should_summarize=enable_summary,
-        fallback_summary=fallback_summary,
         metadata=metadata,
     )
 
@@ -268,8 +260,7 @@ async def index_confluence_pages(
                     connector_id=connector_id,
                     search_space_id=search_space_id,
                     user_id=user_id,
-                    enable_summary=connector.enable_summary,
+                    )
-                )
 
                 with session.no_autoflush:
                     duplicate_by_content = await check_duplicate_document_by_hash(
@@ -297,12 +288,8 @@ async def index_confluence_pages(
 
         await pipeline.migrate_legacy_docs(connector_docs)
 
-        async def _get_llm(s: AsyncSession):
-            return await get_user_long_context_llm(s, user_id, search_space_id)
-
         _, documents_indexed, documents_failed = await pipeline.index_batch_parallel(
             connector_docs,
-            _get_llm,
             max_concurrency=3,
             on_heartbeat=on_heartbeat_callback,
             heartbeat_interval=HEARTBEAT_INTERVAL_SECONDS,

surfsense_backend/app/tasks/connector_indexers/dropbox_indexer.py

@@ -27,7 +27,6 @@ from app.db import Document, DocumentStatus, DocumentType, SearchSourceConnector
 from app.indexing_pipeline.connector_document import ConnectorDocument
 from app.indexing_pipeline.document_hashing import compute_identifier_hash
 from app.indexing_pipeline.indexing_pipeline_service import IndexingPipelineService
-from app.services.llm_service import get_user_long_context_llm
 from app.services.page_limit_service import PageLimitService
 from app.services.task_logging_service import TaskLoggingService
 from app.tasks.connector_indexers.base import (
@@ -126,7 +125,6 @@ def _build_connector_doc(
     connector_id: int,
     search_space_id: int,
     user_id: str,
-    enable_summary: bool,
 ) -> ConnectorDocument:
     file_id = file.get("id", "")
     file_name = file.get("name", "Unknown")
@@ -138,8 +136,6 @@ def _build_connector_doc(
         "connector_type": "Dropbox",
     }
 
-    fallback_summary = f"File: {file_name}\n\n{markdown[:4000]}"
-
     return ConnectorDocument(
         title=file_name,
         source_markdown=markdown,
@@ -148,8 +144,6 @@ def _build_connector_doc(
         search_space_id=search_space_id,
         connector_id=connector_id,
         created_by_id=user_id,
-        should_summarize=enable_summary,
-        fallback_summary=fallback_summary,
         metadata=metadata,
     )
 
@@ -161,7 +155,6 @@ async def _download_files_parallel(
     connector_id: int,
     search_space_id: int,
     user_id: str,
-    enable_summary: bool,
     max_concurrency: int = 3,
     on_heartbeat: HeartbeatCallbackType | None = None,
     vision_llm=None,
@@ -191,7 +184,6 @@ async def _download_files_parallel(
                 connector_id=connector_id,
                 search_space_id=search_space_id,
                 user_id=user_id,
-                enable_summary=enable_summary,
             )
             async with hb_lock:
                 completed_count += 1
@@ -223,7 +215,6 @@ async def _download_and_index(
     connector_id: int,
     search_space_id: int,
     user_id: str,
-    enable_summary: bool,
     on_heartbeat: HeartbeatCallbackType | None = None,
     vision_llm=None,
 ) -> tuple[int, int]:
@@ -234,7 +225,6 @@ async def _download_and_index(
         connector_id=connector_id,
         search_space_id=search_space_id,
         user_id=user_id,
-        enable_summary=enable_summary,
         on_heartbeat=on_heartbeat,
         vision_llm=vision_llm,
     )
@@ -243,13 +233,8 @@ async def _download_and_index(
     batch_failed = 0
     if connector_docs:
         pipeline = IndexingPipelineService(session)
-
-        async def _get_llm(s):
-            return await get_user_long_context_llm(s, user_id, search_space_id)
-
         _, batch_indexed, batch_failed = await pipeline.index_batch_parallel(
             connector_docs,
-            _get_llm,
             max_concurrency=3,
             on_heartbeat=on_heartbeat,
         )
@@ -289,7 +274,6 @@ async def _index_with_delta_sync(
     log_entry: object,
     max_files: int,
     on_heartbeat_callback: HeartbeatCallbackType | None = None,
-    enable_summary: bool = True,
     vision_llm=None,
 ) -> tuple[int, int, int, str]:
     """Delta sync using Dropbox cursor-based change tracking.
@@ -361,7 +345,6 @@ async def _index_with_delta_sync(
         connector_id=connector_id,
         search_space_id=search_space_id,
         user_id=user_id,
-        enable_summary=enable_summary,
         on_heartbeat=on_heartbeat_callback,
         vision_llm=vision_llm,
     )
@@ -388,7 +371,6 @@ async def _index_full_scan(
     include_subfolders: bool = True,
     incremental_sync: bool = True,
     on_heartbeat_callback: HeartbeatCallbackType | None = None,
-    enable_summary: bool = True,
     vision_llm=None,
 ) -> tuple[int, int, int]:
     """Full scan indexing of a folder.
@@ -473,7 +455,6 @@ async def _index_full_scan(
         connector_id=connector_id,
         search_space_id=search_space_id,
         user_id=user_id,
-        enable_summary=enable_summary,
         on_heartbeat=on_heartbeat_callback,
         vision_llm=vision_llm,
     )
@@ -502,7 +483,6 @@ async def _index_selected_files(
     connector_id: int,
     search_space_id: int,
     user_id: str,
-    enable_summary: bool,
     incremental_sync: bool = True,
     on_heartbeat: HeartbeatCallbackType | None = None,
     vision_llm=None,
@@ -563,7 +543,6 @@ async def _index_selected_files(
         connector_id=connector_id,
         search_space_id=search_space_id,
         user_id=user_id,
-        enable_summary=enable_summary,
         on_heartbeat=on_heartbeat,
         vision_llm=vision_llm,
     )
@@ -629,7 +608,6 @@ async def index_dropbox_files(
             )
             return 0, 0, error_msg, 0
 
-        connector_enable_summary = getattr(connector, "enable_summary", True)
         connector_enable_vision_llm = getattr(connector, "enable_vision_llm", False)
         vision_llm = None
         if connector_enable_vision_llm:
@@ -664,7 +642,6 @@ async def index_dropbox_files(
                 connector_id=connector_id,
                 search_space_id=search_space_id,
                 user_id=user_id,
-                enable_summary=connector_enable_summary,
                 incremental_sync=incremental_sync,
                 vision_llm=vision_llm,
             )
@@ -700,7 +677,6 @@ async def index_dropbox_files(
                     task_logger,
                     log_entry,
                     max_files,
-                    enable_summary=connector_enable_summary,
                     vision_llm=vision_llm,
                 )
                 folder_cursors[folder_path] = new_cursor
@@ -720,7 +696,6 @@ async def index_dropbox_files(
                     max_files,
                     include_subfolders,
                     incremental_sync=incremental_sync,
-                    enable_summary=connector_enable_summary,
                     vision_llm=vision_llm,
                 )
                 total_unsupported += unsup

surfsense_backend/app/tasks/connector_indexers/github_indexer.py

@@ -18,13 +18,11 @@ from sqlalchemy.ext.asyncio import AsyncSession
 
 from app.connectors.github_connector import GitHubConnector
 from app.db import Document, DocumentStatus, DocumentType, SearchSourceConnectorType
-from app.services.llm_service import get_user_long_context_llm
 from app.services.task_logging_service import TaskLoggingService
 from app.utils.document_converters import (
     create_document_chunks,
     embed_text,
     generate_content_hash,
-    generate_document_summary,
     generate_unique_identifier_hash,
 )
 
@@ -351,42 +349,14 @@ async def index_github_repos(
                 document.status = DocumentStatus.processing()
                 await session.commit()
 
-                # Heavy processing (LLM, embeddings, chunks)
+                # Heavy processing (embeddings, chunks)
-                user_llm = await get_user_long_context_llm(
+
-                    session, user_id, search_space_id
+                summary_text = (
+                    f"# GitHub Repository: {repo_full_name}\n\n"
+                    f"## Summary\n{digest.summary}\n\n"
+                    f"## File Structure\n{digest.tree}"
                 )
-
+                summary_embedding = embed_text(summary_text)
-                document_metadata_for_summary = {
-                    "repository": repo_full_name,
-                    "document_type": "GitHub Repository",
-                    "connector_type": "GitHub",
-                    "ingestion_method": "gitingest",
-                    "file_tree": digest.tree[:2000]
-                    if len(digest.tree) > 2000
-                    else digest.tree,
-                    "estimated_tokens": digest.estimated_tokens,
-                }
-
-                if user_llm and connector.enable_summary:
-                    # Prepare content for summarization
-                    summary_content = digest.full_digest
-                    if len(summary_content) > MAX_DIGEST_CHARS:
-                        summary_content = (
-                            f"# Repository: {repo_full_name}\n\n"
-                            f"## File Structure\n\n{digest.tree}\n\n"
-                            f"## File Contents (truncated)\n\n{digest.content[: MAX_DIGEST_CHARS - len(digest.tree) - 200]}..."
-                        )
-
-                    summary_text, summary_embedding = await generate_document_summary(
-                        summary_content, user_llm, document_metadata_for_summary
-                    )
-                else:
-                    summary_text = (
-                        f"# GitHub Repository: {repo_full_name}\n\n"
-                        f"## Summary\n{digest.summary}\n\n"
-                        f"## File Structure\n{digest.tree}"
-                    )
-                    summary_embedding = embed_text(summary_text)
 
                 # Chunk the full digest content for granular search
                 try:

surfsense_backend/app/tasks/connector_indexers/google_calendar_indexer.py

@@ -2,7 +2,7 @@
 Google Calendar connector indexer.
 
 Uses the shared IndexingPipelineService for document deduplication,
-summarization, chunking, and embedding.
+chunking, and embedding.
 """
 
 from collections.abc import Awaitable, Callable
@@ -21,7 +21,6 @@ from app.indexing_pipeline.indexing_pipeline_service import (
     PlaceholderInfo,
 )
 from app.services.composio_service import ComposioService
-from app.services.llm_service import get_user_long_context_llm
 from app.services.task_logging_service import TaskLoggingService
 from app.utils.google_credentials import COMPOSIO_GOOGLE_CONNECTOR_TYPES
 
@@ -53,7 +52,6 @@ def _build_connector_doc(
     connector_id: int,
     search_space_id: int,
     user_id: str,
-    enable_summary: bool,
 ) -> ConnectorDocument:
     """Map a raw Google Calendar API event dict to a ConnectorDocument."""
     event_id = event.get("id", "")
@@ -78,8 +76,6 @@ def _build_connector_doc(
         "connector_type": "Google Calendar",
     }
 
-    fallback_summary = f"Google Calendar Event: {event_summary}\n\n{event_markdown}"
-
     return ConnectorDocument(
         title=event_summary,
         source_markdown=event_markdown,
@@ -88,8 +84,6 @@ def _build_connector_doc(
         search_space_id=search_space_id,
         connector_id=connector_id,
         created_by_id=user_id,
-        should_summarize=enable_summary,
-        fallback_summary=fallback_summary,
         metadata=metadata,
     )
 
@@ -420,8 +414,7 @@ async def index_google_calendar_events(
                     connector_id=connector_id,
                     search_space_id=search_space_id,
                     user_id=user_id,
-                    enable_summary=connector.enable_summary,
+                    )
-                )
 
                 with session.no_autoflush:
                     duplicate = await check_duplicate_document_by_hash(
@@ -448,13 +441,8 @@ async def index_google_calendar_events(
 
         # ── Pipeline: migrate legacy docs + parallel index ─────────────
         await pipeline.migrate_legacy_docs(connector_docs)
-
-        async def _get_llm(s):
-            return await get_user_long_context_llm(s, user_id, search_space_id)
-
         _, documents_indexed, documents_failed = await pipeline.index_batch_parallel(
             connector_docs,
-            _get_llm,
             max_concurrency=3,
             on_heartbeat=on_heartbeat_callback,
             heartbeat_interval=HEARTBEAT_INTERVAL_SECONDS,

surfsense_backend/app/tasks/connector_indexers/google_drive_indexer.py

@@ -40,7 +40,6 @@ from app.indexing_pipeline.indexing_pipeline_service import (
     PlaceholderInfo,
 )
 from app.services.composio_service import ComposioService
-from app.services.llm_service import get_user_long_context_llm
 from app.services.page_limit_service import PageLimitService
 from app.services.task_logging_service import TaskLoggingService
 from app.tasks.connector_indexers.base import (
@@ -381,7 +380,6 @@ def _build_connector_doc(
     connector_id: int,
     search_space_id: int,
     user_id: str,
-    enable_summary: bool,
 ) -> ConnectorDocument:
     """Build a ConnectorDocument from Drive file metadata + extracted markdown."""
     file_id = file.get("id", "")
@@ -394,8 +392,6 @@ def _build_connector_doc(
         "connector_type": "Google Drive",
     }
 
-    fallback_summary = f"File: {file_name}\n\n{markdown[:4000]}"
-
     return ConnectorDocument(
         title=file_name,
         source_markdown=markdown,
@@ -404,8 +400,6 @@ def _build_connector_doc(
         search_space_id=search_space_id,
         connector_id=connector_id,
         created_by_id=user_id,
-        should_summarize=enable_summary,
-        fallback_summary=fallback_summary,
         metadata=metadata,
     )
 
@@ -461,7 +455,6 @@ async def _download_files_parallel(
     connector_id: int,
     search_space_id: int,
     user_id: str,
-    enable_summary: bool,
     max_concurrency: int = 3,
     on_heartbeat: HeartbeatCallbackType | None = None,
     vision_llm=None,
@@ -494,7 +487,6 @@ async def _download_files_parallel(
                 connector_id=connector_id,
                 search_space_id=search_space_id,
                 user_id=user_id,
-                enable_summary=enable_summary,
             )
             async with hb_lock:
                 completed_count += 1
@@ -525,7 +517,6 @@ async def _process_single_file(
     connector_id: int,
     search_space_id: int,
     user_id: str,
-    enable_summary: bool = True,
     vision_llm=None,
 ) -> tuple[int, int, int]:
     """Download, extract, and index a single Drive file via the pipeline.
@@ -561,8 +552,7 @@ async def _process_single_file(
             connector_id=connector_id,
             search_space_id=search_space_id,
             user_id=user_id,
-            enable_summary=enable_summary,
+            )
-        )
 
         pipeline = IndexingPipelineService(session)
         documents = await pipeline.prepare_for_indexing([doc])
@@ -578,10 +568,7 @@ async def _process_single_file(
             connector_doc = doc_map.get(document.unique_identifier_hash)
             if not connector_doc:
                 continue
-            user_llm = await get_user_long_context_llm(
+            await pipeline.index(document, connector_doc)
-                session, user_id, search_space_id
-            )
-            await pipeline.index(document, connector_doc, user_llm)
 
         await page_limit_service.update_page_usage(
             user_id, estimated_pages, allow_exceed=True
@@ -636,7 +623,6 @@ async def _download_and_index(
     connector_id: int,
     search_space_id: int,
     user_id: str,
-    enable_summary: bool,
     on_heartbeat: HeartbeatCallbackType | None = None,
     vision_llm=None,
 ) -> tuple[int, int]:
@@ -650,7 +636,6 @@ async def _download_and_index(
         connector_id=connector_id,
         search_space_id=search_space_id,
         user_id=user_id,
-        enable_summary=enable_summary,
         on_heartbeat=on_heartbeat,
         vision_llm=vision_llm,
     )
@@ -659,13 +644,8 @@ async def _download_and_index(
     batch_failed = 0
     if connector_docs:
         pipeline = IndexingPipelineService(session)
-
-        async def _get_llm(s):
-            return await get_user_long_context_llm(s, user_id, search_space_id)
-
         _, batch_indexed, batch_failed = await pipeline.index_batch_parallel(
             connector_docs,
-            _get_llm,
             max_concurrency=3,
             on_heartbeat=on_heartbeat,
         )
@@ -681,7 +661,6 @@ async def _index_selected_files(
     connector_id: int,
     search_space_id: int,
     user_id: str,
-    enable_summary: bool,
     on_heartbeat: HeartbeatCallbackType | None = None,
     vision_llm=None,
 ) -> tuple[int, int, int, list[str]]:
@@ -746,7 +725,6 @@ async def _index_selected_files(
         connector_id=connector_id,
         search_space_id=search_space_id,
         user_id=user_id,
-        enable_summary=enable_summary,
         on_heartbeat=on_heartbeat,
         vision_llm=vision_llm,
     )
@@ -781,7 +759,6 @@ async def _index_full_scan(
     max_files: int,
     include_subfolders: bool = False,
     on_heartbeat_callback: HeartbeatCallbackType | None = None,
-    enable_summary: bool = True,
     vision_llm=None,
 ) -> tuple[int, int, int]:
     """Full scan indexing of a folder.
@@ -911,7 +888,6 @@ async def _index_full_scan(
         connector_id=connector_id,
         search_space_id=search_space_id,
         user_id=user_id,
-        enable_summary=enable_summary,
         on_heartbeat=on_heartbeat_callback,
         vision_llm=vision_llm,
     )
@@ -946,7 +922,6 @@ async def _index_with_delta_sync(
     max_files: int,
     include_subfolders: bool = False,
     on_heartbeat_callback: HeartbeatCallbackType | None = None,
-    enable_summary: bool = True,
     vision_llm=None,
 ) -> tuple[int, int, int]:
     """Delta sync using change tracking.
@@ -1054,7 +1029,6 @@ async def _index_with_delta_sync(
         connector_id=connector_id,
         search_space_id=search_space_id,
         user_id=user_id,
-        enable_summary=enable_summary,
         on_heartbeat=on_heartbeat_callback,
         vision_llm=vision_llm,
     )
@@ -1142,7 +1116,6 @@ async def index_google_drive_files(
             )
             return 0, 0, client_error, 0
 
-        connector_enable_summary = getattr(connector, "enable_summary", True)
         connector_enable_vision_llm = getattr(connector, "enable_vision_llm", False)
         vision_llm = None
         if connector_enable_vision_llm:
@@ -1189,7 +1162,6 @@ async def index_google_drive_files(
                 max_files,
                 include_subfolders,
                 on_heartbeat_callback,
-                connector_enable_summary,
                 vision_llm=vision_llm,
             )
             documents_unsupported += du
@@ -1208,7 +1180,6 @@ async def index_google_drive_files(
                 max_files,
                 include_subfolders,
                 on_heartbeat_callback,
-                connector_enable_summary,
                 vision_llm=vision_llm,
             )
             documents_indexed += ri
@@ -1234,7 +1205,6 @@ async def index_google_drive_files(
                 max_files,
                 include_subfolders,
                 on_heartbeat_callback,
-                connector_enable_summary,
                 vision_llm=vision_llm,
             )
 
@@ -1346,7 +1316,6 @@ async def index_google_drive_single_file(
             )
             return 0, client_error
 
-        connector_enable_summary = getattr(connector, "enable_summary", True)
         connector_enable_vision_llm = getattr(connector, "enable_vision_llm", False)
         vision_llm = None
         if connector_enable_vision_llm:
@@ -1370,7 +1339,6 @@ async def index_google_drive_single_file(
             connector_id,
             search_space_id,
             user_id,
-            connector_enable_summary,
             vision_llm=vision_llm,
         )
         await session.commit()
@@ -1467,7 +1435,6 @@ async def index_google_drive_selected_files(
             )
             return 0, 0, [error_msg]
 
-        connector_enable_summary = getattr(connector, "enable_summary", True)
         connector_enable_vision_llm = getattr(connector, "enable_vision_llm", False)
         vision_llm = None
         if connector_enable_vision_llm:
@@ -1481,7 +1448,6 @@ async def index_google_drive_selected_files(
             connector_id=connector_id,
             search_space_id=search_space_id,
             user_id=user_id,
-            enable_summary=connector_enable_summary,
             on_heartbeat=on_heartbeat_callback,
             vision_llm=vision_llm,
         )

surfsense_backend/app/tasks/connector_indexers/google_gmail_indexer.py

@@ -2,7 +2,7 @@
 Google Gmail connector indexer.
 
 Uses the shared IndexingPipelineService for document deduplication,
-summarization, chunking, and embedding.
+chunking, and embedding.
 """
 
 from collections.abc import Awaitable, Callable
@@ -21,7 +21,6 @@ from app.indexing_pipeline.indexing_pipeline_service import (
     PlaceholderInfo,
 )
 from app.services.composio_service import ComposioService
-from app.services.llm_service import get_user_long_context_llm
 from app.services.task_logging_service import TaskLoggingService
 from app.utils.google_credentials import COMPOSIO_GOOGLE_CONNECTOR_TYPES
 
@@ -105,7 +104,6 @@ def _build_connector_doc(
     connector_id: int,
     search_space_id: int,
     user_id: str,
-    enable_summary: bool,
 ) -> ConnectorDocument:
     """Map a raw Gmail API message dict to a ConnectorDocument."""
     message_id = message.get("id", "")
@@ -138,12 +136,6 @@ def _build_connector_doc(
         "connector_type": "Google Gmail",
     }
 
-    fallback_summary = (
-        f"Google Gmail Message: {subject}\n\n"
-        f"From: {sender}\nDate: {date_str}\n\n"
-        f"{markdown_content}"
-    )
-
     return ConnectorDocument(
         title=subject,
         source_markdown=markdown_content,
@@ -152,8 +144,6 @@ def _build_connector_doc(
         search_space_id=search_space_id,
         connector_id=connector_id,
         created_by_id=user_id,
-        should_summarize=enable_summary,
-        fallback_summary=fallback_summary,
         metadata=metadata,
     )
 
@@ -454,8 +444,7 @@ async def index_google_gmail_messages(
                     connector_id=connector_id,
                     search_space_id=search_space_id,
                     user_id=user_id,
-                    enable_summary=connector.enable_summary,
+                    )
-                )
 
                 with session.no_autoflush:
                     duplicate = await check_duplicate_document_by_hash(
@@ -483,13 +472,8 @@ async def index_google_gmail_messages(
 
         # ── Pipeline: migrate legacy docs + parallel index ─────────────
         await pipeline.migrate_legacy_docs(connector_docs)
-
-        async def _get_llm(s):
-            return await get_user_long_context_llm(s, user_id, search_space_id)
-
         _, documents_indexed, documents_failed = await pipeline.index_batch_parallel(
             connector_docs,
-            _get_llm,
             max_concurrency=3,
             on_heartbeat=on_heartbeat_callback,
             heartbeat_interval=HEARTBEAT_INTERVAL_SECONDS,

surfsense_backend/app/tasks/connector_indexers/linear_indexer.py

@@ -2,7 +2,7 @@
 Linear connector indexer.
 
 Uses the shared IndexingPipelineService for document deduplication,
-summarization, chunking, and embedding with bounded parallel indexing.
+chunking, and embedding with bounded parallel indexing.
 """
 
 from collections.abc import Awaitable, Callable
@@ -18,7 +18,6 @@ from app.indexing_pipeline.indexing_pipeline_service import (
     IndexingPipelineService,
     PlaceholderInfo,
 )
-from app.services.llm_service import get_user_long_context_llm
 from app.services.task_logging_service import TaskLoggingService
 
 from .base import (
@@ -41,7 +40,6 @@ def _build_connector_doc(
     connector_id: int,
     search_space_id: int,
     user_id: str,
-    enable_summary: bool,
 ) -> ConnectorDocument:
     """Map a raw Linear issue dict to a ConnectorDocument."""
     issue_id = issue.get("id", "")
@@ -63,11 +61,6 @@ def _build_connector_doc(
         "connector_type": "Linear",
     }
 
-    fallback_summary = (
-        f"Linear Issue {issue_identifier}: {issue_title}\n\n"
-        f"Status: {state}\n\n{issue_content}"
-    )
-
     return ConnectorDocument(
         title=f"{issue_identifier}: {issue_title}",
         source_markdown=issue_content,
@@ -76,8 +69,6 @@ def _build_connector_doc(
         search_space_id=search_space_id,
         connector_id=connector_id,
         created_by_id=user_id,
-        should_summarize=enable_summary,
-        fallback_summary=fallback_summary,
         metadata=metadata,
     )
 
@@ -277,8 +268,7 @@ async def index_linear_issues(
                     connector_id=connector_id,
                     search_space_id=search_space_id,
                     user_id=user_id,
-                    enable_summary=connector.enable_summary,
+                    )
-                )
 
                 with session.no_autoflush:
                     duplicate = await check_duplicate_document_by_hash(
@@ -306,13 +296,8 @@ async def index_linear_issues(
 
         # ── Pipeline: migrate legacy docs + parallel index ────────────
         await pipeline.migrate_legacy_docs(connector_docs)
-
-        async def _get_llm(s):
-            return await get_user_long_context_llm(s, user_id, search_space_id)
-
         _, documents_indexed, documents_failed = await pipeline.index_batch_parallel(
             connector_docs,
-            _get_llm,
             max_concurrency=3,
             on_heartbeat=on_heartbeat_callback,
             heartbeat_interval=HEARTBEAT_INTERVAL_SECONDS,

surfsense_backend/app/tasks/connector_indexers/local_folder_indexer.py

@@ -33,7 +33,6 @@ from app.db import (
 from app.indexing_pipeline.connector_document import ConnectorDocument
 from app.indexing_pipeline.document_hashing import compute_identifier_hash
 from app.indexing_pipeline.indexing_pipeline_service import IndexingPipelineService
-from app.services.llm_service import get_user_long_context_llm
 from app.services.page_limit_service import PageLimitExceededError, PageLimitService
 from app.services.task_logging_service import TaskLoggingService
 from app.tasks.celery_tasks import get_celery_session_maker
@@ -478,7 +477,6 @@ def _build_connector_doc(
     *,
     search_space_id: int,
     user_id: str,
-    enable_summary: bool,
 ) -> ConnectorDocument:
     """Build a ConnectorDocument from a local file's extracted content."""
     unique_id = f"{folder_name}:{relative_path}"
@@ -488,7 +486,6 @@ def _build_connector_doc(
         "document_type": "Local Folder File",
         "connector_type": "Local Folder",
     }
-    fallback_summary = f"File: {title}\n\n{content[:4000]}"
 
     return ConnectorDocument(
         title=title,
@@ -498,8 +495,6 @@ def _build_connector_doc(
         search_space_id=search_space_id,
         connector_id=None,
         created_by_id=user_id,
-        should_summarize=enable_summary,
-        fallback_summary=fallback_summary,
         metadata=metadata,
     )
 
@@ -513,7 +508,6 @@ async def index_local_folder(
     exclude_patterns: list[str] | None = None,
     file_extensions: list[str] | None = None,
     root_folder_id: int | None = None,
-    enable_summary: bool = False,
     target_file_paths: list[str] | None = None,
     on_heartbeat_callback: HeartbeatCallbackType | None = None,
 ) -> tuple[int, int, int | None, str | None]:
@@ -574,8 +568,7 @@ async def index_local_folder(
                         folder_path=folder_path,
                         folder_name=folder_name,
                         target_file_path=target_file_paths[0],
-                        enable_summary=enable_summary,
+                                root_folder_id=root_folder_id,
-                        root_folder_id=root_folder_id,
                         task_logger=task_logger,
                         log_entry=log_entry,
                     )
@@ -587,8 +580,7 @@ async def index_local_folder(
                     folder_path=folder_path,
                     folder_name=folder_name,
                     target_file_paths=target_file_paths,
-                    enable_summary=enable_summary,
+                        root_folder_id=root_folder_id,
-                    root_folder_id=root_folder_id,
                     on_progress_callback=on_heartbeat_callback,
                 )
                 if err:
@@ -774,8 +766,7 @@ async def index_local_folder(
                     folder_name=folder_name,
                     search_space_id=search_space_id,
                     user_id=user_id,
-                    enable_summary=enable_summary,
+                    )
-                )
                 connector_docs.append(doc)
                 file_meta_map[unique_identifier] = {
                     "relative_path": relative_path,
@@ -845,15 +836,13 @@ async def index_local_folder(
             doc_map = {compute_unique_identifier_hash(cd): cd for cd in connector_docs}
             documents = await pipeline.prepare_for_indexing(connector_docs)
 
-            llm = await get_user_long_context_llm(session, user_id, search_space_id)
-
             for document in documents:
                 connector_doc = doc_map.get(document.unique_identifier_hash)
                 if connector_doc is None:
                     failed_count += 1
                     continue
 
-                result = await pipeline.index(document, connector_doc, llm)
+                result = await pipeline.index(document, connector_doc)
 
                 if DocumentStatus.is_state(result.status, DocumentStatus.READY):
                     indexed_count += 1
@@ -960,7 +949,6 @@ async def _index_batch_files(
     folder_path: str,
     folder_name: str,
     target_file_paths: list[str],
-    enable_summary: bool,
     root_folder_id: int | None,
     on_progress_callback: HeartbeatCallbackType | None = None,
 ) -> tuple[int, int, str | None]:
@@ -995,8 +983,7 @@ async def _index_batch_files(
                         folder_path=folder_path,
                         folder_name=folder_name,
                         target_file_path=file_path,
-                        enable_summary=enable_summary,
+                                root_folder_id=root_folder_id,
-                        root_folder_id=root_folder_id,
                         task_logger=task_logger,
                         log_entry=log_entry,
                     )
@@ -1036,7 +1023,6 @@ async def _index_single_file(
     folder_path: str,
     folder_name: str,
     target_file_path: str,
-    enable_summary: bool,
     root_folder_id: int | None,
     task_logger,
     log_entry,
@@ -1125,8 +1111,7 @@ async def _index_single_file(
             folder_name=folder_name,
             search_space_id=search_space_id,
             user_id=user_id,
-            enable_summary=enable_summary,
+            )
-        )
 
         if root_folder_id:
             connector_doc.folder_id = await _resolve_folder_for_file(
@@ -1134,7 +1119,6 @@ async def _index_single_file(
             )
 
         pipeline = IndexingPipelineService(session)
-        llm = await get_user_long_context_llm(session, user_id, search_space_id)
         documents = await pipeline.prepare_for_indexing([connector_doc])
 
         if not documents:
@@ -1142,7 +1126,7 @@ async def _index_single_file(
 
         db_doc = documents[0]
 
-        await pipeline.index(db_doc, connector_doc, llm)
+        await pipeline.index(db_doc, connector_doc)
 
         await session.refresh(db_doc)
         doc_meta = dict(db_doc.document_metadata or {})
@@ -1275,7 +1259,6 @@ async def index_uploaded_files(
     user_id: str,
     folder_name: str,
     root_folder_id: int,
-    enable_summary: bool,
     file_mappings: list[dict],
     on_heartbeat_callback: HeartbeatCallbackType | None = None,
     use_vision_llm: bool = False,
@@ -1318,7 +1301,6 @@ async def index_uploaded_files(
 
         page_limit_service = PageLimitService(session)
         pipeline = IndexingPipelineService(session)
-        llm = await get_user_long_context_llm(session, user_id, search_space_id)
 
         vision_llm_instance = None
         if use_vision_llm:
@@ -1414,8 +1396,7 @@ async def index_uploaded_files(
                     folder_name=folder_name,
                     search_space_id=search_space_id,
                     user_id=user_id,
-                    enable_summary=enable_summary,
+                    )
-                )
 
                 connector_doc.folder_id = await _resolve_folder_for_file(
                     session,
@@ -1432,7 +1413,7 @@ async def index_uploaded_files(
 
                 db_doc = documents[0]
 
-                await pipeline.index(db_doc, connector_doc, llm)
+                await pipeline.index(db_doc, connector_doc)
 
                 await session.refresh(db_doc)
                 doc_meta = dict(db_doc.document_metadata or {})

surfsense_backend/app/tasks/connector_indexers/luma_indexer.py

@@ -16,13 +16,11 @@ from sqlalchemy.ext.asyncio import AsyncSession
 
 from app.connectors.luma_connector import LumaConnector
 from app.db import Document, DocumentStatus, DocumentType, SearchSourceConnectorType
-from app.services.llm_service import get_user_long_context_llm
 from app.services.task_logging_service import TaskLoggingService
 from app.utils.document_converters import (
     create_document_chunks,
     embed_text,
     generate_content_hash,
-    generate_document_summary,
     generate_unique_identifier_hash,
 )
 
@@ -437,38 +435,14 @@ async def index_luma_events(
                 document.status = DocumentStatus.processing()
                 await session.commit()
 
-                # Heavy processing (LLM, embeddings, chunks)
+                # Heavy processing (embeddings, chunks)
-                user_llm = await get_user_long_context_llm(
-                    session, user_id, search_space_id
-                )
 
-                if user_llm and connector.enable_summary:
+                summary_content = (
-                    document_metadata_for_summary = {
+                    f"Luma Event: {item['event_name']}\n\n{item['event_markdown']}"
-                        "event_id": item["event_id"],
+                )
-                        "event_name": item["event_name"],
+                summary_embedding = await asyncio.to_thread(
-                        "event_url": item["event_url"],
+                    embed_text, summary_content
-                        "start_at": item["start_at"],
+                )
-                        "end_at": item["end_at"],
-                        "timezone": item["timezone"],
-                        "location": item["location"] or "No location",
-                        "city": item["city"],
-                        "hosts": item["host_names"],
-                        "document_type": "Luma Event",
-                        "connector_type": "Luma",
-                    }
-                    (
-                        summary_content,
-                        summary_embedding,
-                    ) = await generate_document_summary(
-                        item["event_markdown"], user_llm, document_metadata_for_summary
-                    )
-                else:
-                    summary_content = (
-                        f"Luma Event: {item['event_name']}\n\n{item['event_markdown']}"
-                    )
-                    summary_embedding = await asyncio.to_thread(
-                        embed_text, summary_content
-                    )
 
                 chunks = await create_document_chunks(item["event_markdown"])
 

surfsense_backend/app/tasks/connector_indexers/notion_indexer.py

@@ -2,7 +2,7 @@
 Notion connector indexer.
 
 Uses the shared IndexingPipelineService for document deduplication,
-summarization, chunking, and embedding with bounded parallel indexing.
+chunking, and embedding with bounded parallel indexing.
 """
 
 from collections.abc import Awaitable, Callable
@@ -19,7 +19,6 @@ from app.indexing_pipeline.indexing_pipeline_service import (
     IndexingPipelineService,
     PlaceholderInfo,
 )
-from app.services.llm_service import get_user_long_context_llm
 from app.services.task_logging_service import TaskLoggingService
 from app.utils.notion_utils import process_blocks
 
@@ -43,7 +42,6 @@ def _build_connector_doc(
     connector_id: int,
     search_space_id: int,
     user_id: str,
-    enable_summary: bool,
 ) -> ConnectorDocument:
     """Map a raw Notion page dict to a ConnectorDocument."""
     page_id = page.get("page_id", "")
@@ -57,8 +55,6 @@ def _build_connector_doc(
         "connector_type": "Notion",
     }
 
-    fallback_summary = f"Notion Page: {page_title}\n\n{markdown_content}"
-
     return ConnectorDocument(
         title=page_title,
         source_markdown=markdown_content,
@@ -67,8 +63,6 @@ def _build_connector_doc(
         search_space_id=search_space_id,
         connector_id=connector_id,
         created_by_id=user_id,
-        should_summarize=enable_summary,
-        fallback_summary=fallback_summary,
         metadata=metadata,
     )
 
@@ -314,8 +308,7 @@ async def index_notion_pages(
                     connector_id=connector_id,
                     search_space_id=search_space_id,
                     user_id=user_id,
-                    enable_summary=connector.enable_summary,
+                    )
-                )
 
                 with session.no_autoflush:
                     duplicate = await check_duplicate_document_by_hash(
@@ -343,13 +336,8 @@ async def index_notion_pages(
 
         # ── Pipeline: migrate legacy docs + parallel index ────────────
         await pipeline.migrate_legacy_docs(connector_docs)
-
-        async def _get_llm(s):
-            return await get_user_long_context_llm(s, user_id, search_space_id)
-
         _, documents_indexed, documents_failed = await pipeline.index_batch_parallel(
             connector_docs,
-            _get_llm,
             max_concurrency=3,
             on_heartbeat=on_heartbeat_callback,
             heartbeat_interval=HEARTBEAT_INTERVAL_SECONDS,

surfsense_backend/app/tasks/connector_indexers/onedrive_indexer.py

@@ -27,7 +27,6 @@ from app.db import Document, DocumentStatus, DocumentType, SearchSourceConnector
 from app.indexing_pipeline.connector_document import ConnectorDocument
 from app.indexing_pipeline.document_hashing import compute_identifier_hash
 from app.indexing_pipeline.indexing_pipeline_service import IndexingPipelineService
-from app.services.llm_service import get_user_long_context_llm
 from app.services.page_limit_service import PageLimitService
 from app.services.task_logging_service import TaskLoggingService
 from app.tasks.connector_indexers.base import (
@@ -133,7 +132,6 @@ def _build_connector_doc(
     connector_id: int,
     search_space_id: int,
     user_id: str,
-    enable_summary: bool,
 ) -> ConnectorDocument:
     file_id = file.get("id", "")
     file_name = file.get("name", "Unknown")
@@ -145,8 +143,6 @@ def _build_connector_doc(
         "connector_type": "OneDrive",
     }
 
-    fallback_summary = f"File: {file_name}\n\n{markdown[:4000]}"
-
     return ConnectorDocument(
         title=file_name,
         source_markdown=markdown,
@@ -155,8 +151,6 @@ def _build_connector_doc(
         search_space_id=search_space_id,
         connector_id=connector_id,
         created_by_id=user_id,
-        should_summarize=enable_summary,
-        fallback_summary=fallback_summary,
         metadata=metadata,
     )
 
@@ -168,7 +162,6 @@ async def _download_files_parallel(
     connector_id: int,
     search_space_id: int,
     user_id: str,
-    enable_summary: bool,
     max_concurrency: int = 3,
     on_heartbeat: HeartbeatCallbackType | None = None,
     vision_llm=None,
@@ -198,7 +191,6 @@ async def _download_files_parallel(
                 connector_id=connector_id,
                 search_space_id=search_space_id,
                 user_id=user_id,
-                enable_summary=enable_summary,
             )
             async with hb_lock:
                 completed_count += 1
@@ -230,7 +222,6 @@ async def _download_and_index(
     connector_id: int,
     search_space_id: int,
     user_id: str,
-    enable_summary: bool,
     on_heartbeat: HeartbeatCallbackType | None = None,
     vision_llm=None,
 ) -> tuple[int, int]:
@@ -241,7 +232,6 @@ async def _download_and_index(
         connector_id=connector_id,
         search_space_id=search_space_id,
         user_id=user_id,
-        enable_summary=enable_summary,
         on_heartbeat=on_heartbeat,
         vision_llm=vision_llm,
     )
@@ -250,13 +240,8 @@ async def _download_and_index(
     batch_failed = 0
     if connector_docs:
         pipeline = IndexingPipelineService(session)
-
-        async def _get_llm(s):
-            return await get_user_long_context_llm(s, user_id, search_space_id)
-
         _, batch_indexed, batch_failed = await pipeline.index_batch_parallel(
             connector_docs,
-            _get_llm,
             max_concurrency=3,
             on_heartbeat=on_heartbeat,
         )
@@ -294,7 +279,6 @@ async def _index_selected_files(
     connector_id: int,
     search_space_id: int,
     user_id: str,
-    enable_summary: bool,
     on_heartbeat: HeartbeatCallbackType | None = None,
     vision_llm=None,
 ) -> tuple[int, int, int, list[str]]:
@@ -345,7 +329,6 @@ async def _index_selected_files(
         connector_id=connector_id,
         search_space_id=search_space_id,
         user_id=user_id,
-        enable_summary=enable_summary,
         on_heartbeat=on_heartbeat,
         vision_llm=vision_llm,
     )
@@ -379,7 +362,6 @@ async def _index_full_scan(
     max_files: int,
     include_subfolders: bool = True,
     on_heartbeat_callback: HeartbeatCallbackType | None = None,
-    enable_summary: bool = True,
     vision_llm=None,
 ) -> tuple[int, int, int]:
     """Full scan indexing of a folder.
@@ -454,7 +436,6 @@ async def _index_full_scan(
         connector_id=connector_id,
         search_space_id=search_space_id,
         user_id=user_id,
-        enable_summary=enable_summary,
         on_heartbeat=on_heartbeat_callback,
         vision_llm=vision_llm,
     )
@@ -487,7 +468,6 @@ async def _index_with_delta_sync(
     log_entry: object,
     max_files: int,
     on_heartbeat_callback: HeartbeatCallbackType | None = None,
-    enable_summary: bool = True,
     vision_llm=None,
 ) -> tuple[int, int, int, str | None]:
     """Delta sync using OneDrive change tracking.
@@ -579,7 +559,6 @@ async def _index_with_delta_sync(
         connector_id=connector_id,
         search_space_id=search_space_id,
         user_id=user_id,
-        enable_summary=enable_summary,
         on_heartbeat=on_heartbeat_callback,
         vision_llm=vision_llm,
     )
@@ -651,7 +630,6 @@ async def index_onedrive_files(
             )
             return 0, 0, error_msg, 0
 
-        connector_enable_summary = getattr(connector, "enable_summary", True)
         connector_enable_vision_llm = getattr(connector, "enable_vision_llm", False)
         vision_llm = None
         if connector_enable_vision_llm:
@@ -681,7 +659,6 @@ async def index_onedrive_files(
                 connector_id=connector_id,
                 search_space_id=search_space_id,
                 user_id=user_id,
-                enable_summary=connector_enable_summary,
                 vision_llm=vision_llm,
             )
             total_indexed += indexed
@@ -711,7 +688,6 @@ async def index_onedrive_files(
                     task_logger,
                     log_entry,
                     max_files,
-                    enable_summary=connector_enable_summary,
                     vision_llm=vision_llm,
                 )
                 total_indexed += indexed
@@ -738,7 +714,6 @@ async def index_onedrive_files(
                     log_entry,
                     max_files,
                     include_subfolders,
-                    enable_summary=connector_enable_summary,
                     vision_llm=vision_llm,
                 )
                 total_indexed += ri
@@ -758,7 +733,6 @@ async def index_onedrive_files(
                     log_entry,
                     max_files,
                     include_subfolders,
-                    enable_summary=connector_enable_summary,
                     vision_llm=vision_llm,
                 )
                 total_indexed += indexed

surfsense_backend/app/tasks/connector_indexers/webcrawler_indexer.py

@@ -15,13 +15,11 @@ from sqlalchemy.ext.asyncio import AsyncSession
 
 from app.connectors.webcrawler_connector import WebCrawlerConnector
 from app.db import Document, DocumentStatus, DocumentType, SearchSourceConnectorType
-from app.services.llm_service import get_user_long_context_llm
 from app.services.task_logging_service import TaskLoggingService
 from app.utils.document_converters import (
     create_document_chunks,
     embed_text,
     generate_content_hash,
-    generate_document_summary,
     generate_unique_identifier_hash,
 )
 from app.utils.webcrawler_utils import parse_webcrawler_urls
@@ -372,29 +370,10 @@ async def index_crawled_urls(
                         documents_skipped += 1
                         continue
 
-                # Generate summary with LLM
+                # Select deterministic document content
-                user_llm = await get_user_long_context_llm(
-                    session, user_id, search_space_id
-                )
 
-                if user_llm and connector.enable_summary:
+                summary_content = f"Crawled URL: {title}\n\nURL: {url}\n\n{content}"
-                    document_metadata_for_summary = {
+                summary_embedding = embed_text(summary_content)
-                        "url": url,
-                        "title": title,
-                        "description": description,
-                        "language": language,
-                        "document_type": "Crawled URL",
-                        "crawler_type": crawler_type,
-                    }
-                    (
-                        summary_content,
-                        summary_embedding,
-                    ) = await generate_document_summary(
-                        structured_document, user_llm, document_metadata_for_summary
-                    )
-                else:
-                    summary_content = f"Crawled URL: {title}\n\nURL: {url}\n\n{content}"
-                    summary_embedding = embed_text(summary_content)
 
                 # Process chunks
                 chunks = await create_document_chunks(content)

surfsense_backend/app/tasks/document_processors/_save.py

@@ -1,20 +1,15 @@
-"""
+"""Unified document save/update logic for file processors."""
-Unified document save/update logic for file processors.
-"""
 
-import asyncio
 import logging
 
 from sqlalchemy.exc import SQLAlchemyError
 from sqlalchemy.ext.asyncio import AsyncSession
 
 from app.db import Document, DocumentStatus, DocumentType
-from app.services.llm_service import get_user_long_context_llm
 from app.utils.document_converters import (
     create_document_chunks,
     embed_text,
     generate_content_hash,
-    generate_document_summary,
 )
 
 from ._helpers import (
@@ -24,59 +19,6 @@ from ._helpers import (
 )
 from .base import get_current_timestamp, safe_set_chunks
 
-# ---------------------------------------------------------------------------
-# Summary generation
-# ---------------------------------------------------------------------------
-
-
-async def _generate_summary(
-    markdown_content: str,
-    file_name: str,
-    etl_service: str,
-    user_llm,
-    enable_summary: bool,
-) -> tuple[str, list[float]]:
-    """
-    Generate a document summary and embedding.
-
-    Docling uses its own large-document summary strategy; other ETL services
-    use the standard ``generate_document_summary`` helper.
-    """
-    if not enable_summary:
-        summary = f"File: {file_name}\n\n{markdown_content[:4000]}"
-        return summary, await asyncio.to_thread(embed_text, summary)
-
-    if etl_service == "DOCLING":
-        from app.services.docling_service import create_docling_service
-
-        docling_service = create_docling_service()
-        summary_text = await docling_service.process_large_document_summary(
-            content=markdown_content, llm=user_llm, document_title=file_name
-        )
-
-        meta = {
-            "file_name": file_name,
-            "etl_service": etl_service,
-            "document_type": "File Document",
-        }
-        parts = ["# DOCUMENT METADATA"]
-        for key, value in meta.items():
-            if value:
-                formatted_key = key.replace("_", " ").title()
-                parts.append(f"**{formatted_key}:** {value}")
-
-        enhanced = "\n".join(parts) + "\n\n# DOCUMENT SUMMARY\n\n" + summary_text
-        return enhanced, await asyncio.to_thread(embed_text, enhanced)
-
-    # Standard summary (Unstructured / LlamaCloud / others)
-    meta = {
-        "file_name": file_name,
-        "etl_service": etl_service,
-        "document_type": "File Document",
-    }
-    return await generate_document_summary(markdown_content, user_llm, meta)
-
-
 # ---------------------------------------------------------------------------
 # Unified save function
 # ---------------------------------------------------------------------------
@@ -90,7 +32,6 @@ async def save_file_document(
     user_id: str,
     etl_service: str,
     connector: dict | None = None,
-    enable_summary: bool = True,
 ) -> Document | None:
     """
     Process and store a file document with deduplication and migration support.
@@ -106,7 +47,6 @@ async def save_file_document(
         user_id: ID of the user
         etl_service: Name of the ETL service (UNSTRUCTURED, LLAMACLOUD, DOCLING)
         connector: Optional connector info for Google Drive files
-        enable_summary: Whether to generate an AI summary
 
     Returns:
         Document object if successful, None if duplicate detected
@@ -133,24 +73,16 @@ async def save_file_document(
             if should_skip:
                 return doc
 
-        user_llm = await get_user_long_context_llm(session, user_id, search_space_id)
+        document_content = f"File: {file_name}\n\n{markdown_content[:4000]}"
-        if not user_llm:
+        document_embedding = embed_text(document_content)
-            raise RuntimeError(
-                f"No long context LLM configured for user {user_id} "
-                f"in search space {search_space_id}"
-            )
-
-        summary_content, summary_embedding = await _generate_summary(
-            markdown_content, file_name, etl_service, user_llm, enable_summary
-        )
         chunks = await create_document_chunks(markdown_content)
         doc_metadata = {"FILE_NAME": file_name, "ETL_SERVICE": etl_service}
 
         if existing_document:
             existing_document.title = file_name
-            existing_document.content = summary_content
+            existing_document.content = document_content
             existing_document.content_hash = content_hash
-            existing_document.embedding = summary_embedding
+            existing_document.embedding = document_embedding
             existing_document.document_metadata = doc_metadata
             await safe_set_chunks(session, existing_document, chunks)
             existing_document.source_markdown = markdown_content
@@ -171,8 +103,8 @@ async def save_file_document(
             title=file_name,
             document_type=doc_type,
             document_metadata=doc_metadata,
-            content=summary_content,
+            content=document_content,
-            embedding=summary_embedding,
+            embedding=document_embedding,
             chunks=chunks,
             content_hash=content_hash,
             unique_identifier_hash=primary_hash,

surfsense_backend/app/tasks/document_processors/circleback_processor.py

@@ -25,11 +25,10 @@ from app.db import (
     SearchSourceConnectorType,
     SearchSpace,
 )
-from app.services.llm_service import get_document_summary_llm
 from app.utils.document_converters import (
     create_document_chunks,
+    embed_text,
     generate_content_hash,
-    generate_document_summary,
     generate_unique_identifier_hash,
 )
 
@@ -176,34 +175,8 @@ async def add_circleback_meeting_document(
         # PHASE 3: Process the document content
         # =======================================================================
 
-        # Get LLM for generating summary
+        summary_content = markdown_content
-        llm = await get_document_summary_llm(session, search_space_id)
+        summary_embedding = embed_text(summary_content)
-        if not llm:
-            logger.warning(
-                f"No LLM configured for search space {search_space_id}. Using content as summary."
-            )
-            # Use first 1000 chars as summary if no LLM available
-            summary_content = (
-                markdown_content[:1000] + "..."
-                if len(markdown_content) > 1000
-                else markdown_content
-            )
-            summary_embedding = None
-        else:
-            # Generate summary with metadata
-            summary_metadata = {
-                "meeting_name": meeting_name,
-                "meeting_id": meeting_id,
-                "document_type": "Circleback Meeting",
-                **{
-                    k: v
-                    for k, v in metadata.items()
-                    if isinstance(v, str | int | float | bool)
-                },
-            }
-            summary_content, summary_embedding = await generate_document_summary(
-                markdown_content, llm, summary_metadata
-            )
 
         # Process chunks
         chunks = await create_document_chunks(markdown_content)
@@ -224,8 +197,7 @@ async def add_circleback_meeting_document(
         document.title = meeting_name
         document.content = summary_content
         document.content_hash = content_hash
-        if summary_embedding is not None:
+        document.embedding = summary_embedding
-            document.embedding = summary_embedding
         document.document_metadata = document_metadata
         await safe_set_chunks(session, document, chunks)
         document.source_markdown = markdown_content

surfsense_backend/app/tasks/document_processors/extension_processor.py

@@ -9,12 +9,11 @@ from sqlalchemy.ext.asyncio import AsyncSession
 
 from app.db import Document, DocumentType
 from app.schemas import ExtensionDocumentContent
-from app.services.llm_service import get_user_long_context_llm
 from app.services.task_logging_service import TaskLoggingService
 from app.utils.document_converters import (
     create_document_chunks,
+    embed_text,
     generate_content_hash,
-    generate_document_summary,
     generate_unique_identifier_hash,
 )
 
@@ -123,26 +122,8 @@ async def add_extension_received_document(
                     f"Content changed for URL {content.metadata.VisitedWebPageURL}. Updating document."
                 )
 
-        # Get user's long context LLM (needed for both create and update)
+        summary_content = combined_document_string
-        user_llm = await get_user_long_context_llm(session, user_id, search_space_id)
+        summary_embedding = embed_text(summary_content)
-        if not user_llm:
-            raise RuntimeError(
-                f"No long context LLM configured for user {user_id} in search space {search_space_id}"
-            )
-
-        # Generate summary with metadata
-        document_metadata = {
-            "session_id": content.metadata.BrowsingSessionId,
-            "url": content.metadata.VisitedWebPageURL,
-            "title": content.metadata.VisitedWebPageTitle,
-            "referrer": content.metadata.VisitedWebPageReffererURL,
-            "timestamp": content.metadata.VisitedWebPageDateWithTimeInISOString,
-            "duration_ms": content.metadata.VisitedWebPageVisitDurationInMilliseconds,
-            "document_type": "Browser Extension Capture",
-        }
-        summary_content, summary_embedding = await generate_document_summary(
-            combined_document_string, user_llm, document_metadata
-        )
 
         # Process chunks
         chunks = await create_document_chunks(content.pageContent)

surfsense_backend/app/tasks/document_processors/file_processors.py

@@ -10,7 +10,7 @@ from __future__ import annotations
 import contextlib
 import logging
 import os
-from dataclasses import dataclass, field
+from dataclasses import dataclass
 
 from fastapi import HTTPException
 from sqlalchemy.ext.asyncio import AsyncSession
@@ -48,12 +48,6 @@ class _ProcessingContext:
     notification: Notification | None = None
     use_vision_llm: bool = False
     processing_mode: str = "basic"
-    enable_summary: bool = field(init=False)
-
-    def __post_init__(self) -> None:
-        self.enable_summary = (
-            self.connector.get("enable_summary", True) if self.connector else True
-        )
 
 
 # ---------------------------------------------------------------------------
@@ -261,7 +255,6 @@ async def _process_document_upload(ctx: _ProcessingContext) -> Document | None:
         ctx.user_id,
         etl_result.etl_service,
         ctx.connector,
-        enable_summary=ctx.enable_summary,
     )
 
     if result:
@@ -466,7 +459,6 @@ async def process_file_in_background_with_document(
     log_entry: Log,
     connector: dict | None = None,
     notification: Notification | None = None,
-    should_summarize: bool = False,
     use_vision_llm: bool = False,
     processing_mode: str = "basic",
 ) -> Document | None:
@@ -482,7 +474,6 @@ async def process_file_in_background_with_document(
     from app.indexing_pipeline.adapters.file_upload_adapter import (
         UploadDocumentAdapter,
     )
-    from app.services.llm_service import get_user_long_context_llm
     from app.utils.document_converters import generate_content_hash
 
     from .base import check_duplicate_document
@@ -522,8 +513,6 @@ async def process_file_in_background_with_document(
                 stage="chunking",
             )
 
-        user_llm = await get_user_long_context_llm(session, user_id, search_space_id)
-
         adapter = UploadDocumentAdapter(session)
         await adapter.index(
             markdown_content=markdown_content,
@@ -531,8 +520,6 @@ async def process_file_in_background_with_document(
             etl_service=etl_service,
             search_space_id=search_space_id,
             user_id=user_id,
-            llm=user_llm,
-            should_summarize=should_summarize,
         )
 
         if billable_pages > 0:

surfsense_backend/app/tasks/document_processors/markdown_processor.py

@@ -8,12 +8,11 @@ from sqlalchemy.exc import SQLAlchemyError
 from sqlalchemy.ext.asyncio import AsyncSession
 
 from app.db import Document, DocumentStatus, DocumentType
-from app.services.llm_service import get_user_long_context_llm
 from app.services.task_logging_service import TaskLoggingService
 from app.utils.document_converters import (
     create_document_chunks,
+    embed_text,
     generate_content_hash,
-    generate_document_summary,
 )
 
 from ._helpers import (
@@ -183,21 +182,8 @@ async def add_received_markdown_file_document(
                 return doc
             # Content changed - continue to update
 
-        # Get user's long context LLM (needed for both create and update)
+        summary_content = f"File: {file_name}\n\n{file_in_markdown[:4000]}"
-        user_llm = await get_user_long_context_llm(session, user_id, search_space_id)
+        summary_embedding = embed_text(summary_content)
-        if not user_llm:
-            raise RuntimeError(
-                f"No long context LLM configured for user {user_id} in search space {search_space_id}"
-            )
-
-        # Generate summary with metadata
-        document_metadata = {
-            "file_name": file_name,
-            "document_type": "Markdown File Document",
-        }
-        summary_content, summary_embedding = await generate_document_summary(
-            file_in_markdown, user_llm, document_metadata
-        )
 
         # Process chunks
         chunks = await create_document_chunks(file_in_markdown)

surfsense_backend/app/tasks/document_processors/youtube_processor.py

@@ -17,12 +17,11 @@ from sqlalchemy.ext.asyncio import AsyncSession
 from youtube_transcript_api import YouTubeTranscriptApi
 
 from app.db import Document, DocumentStatus, DocumentType
-from app.services.llm_service import get_user_long_context_llm
 from app.services.task_logging_service import TaskLoggingService
 from app.utils.document_converters import (
     create_document_chunks,
+    embed_text,
     generate_content_hash,
-    generate_document_summary,
     generate_unique_identifier_hash,
 )
 from app.utils.proxy_config import get_requests_proxies
@@ -355,40 +354,8 @@ async def add_youtube_video_document(
             await session.commit()
             return document
 
-        # Get LLM for summary generation
+        summary_content = combined_document_string
-        await task_logger.log_task_progress(
+        summary_embedding = embed_text(summary_content)
-            log_entry,
-            f"Preparing for summary generation: {video_data.get('title', 'YouTube Video')}",
-            {"stage": "llm_setup"},
-        )
-
-        # Get user's long context LLM
-        user_llm = await get_user_long_context_llm(session, user_id, search_space_id)
-        if not user_llm:
-            raise RuntimeError(
-                f"No long context LLM configured for user {user_id} in search space {search_space_id}"
-            )
-
-        # Generate summary
-        await task_logger.log_task_progress(
-            log_entry,
-            f"Generating summary for video: {video_data.get('title', 'YouTube Video')}",
-            {"stage": "summary_generation"},
-        )
-
-        # Generate summary with metadata
-        document_metadata_for_summary = {
-            "url": url,
-            "video_id": video_id,
-            "title": video_data.get("title", "YouTube Video"),
-            "author": video_data.get("author_name", "Unknown"),
-            "thumbnail": video_data.get("thumbnail_url", ""),
-            "document_type": "YouTube Video Document",
-            "has_transcript": "No captions available" not in transcript_text,
-        }
-        summary_content, summary_embedding = await generate_document_summary(
-            combined_document_string, user_llm, document_metadata_for_summary
-        )
 
         # Process chunks
         await task_logger.log_task_progress(

surfsense_backend/app/utils/document_converters.py

@@ -9,7 +9,6 @@ from litellm import get_model_info, token_counter
 
 from app.config import config
 from app.db import Chunk, DocumentType
-from app.prompts import SUMMARY_PROMPT_TEMPLATE
 
 logger = logging.getLogger(__name__)
 
@@ -176,57 +175,6 @@ def optimize_content_for_context_window(
     return optimized_content
 
 
-async def generate_document_summary(
-    content: str,
-    user_llm,
-    document_metadata: dict | None = None,
-) -> tuple[str, list[float]]:
-    """
-    Generate summary and embedding for document content with metadata.
-
-    Args:
-        content: Document content
-        user_llm: User's LLM instance
-        document_metadata: Optional metadata dictionary to include in summary
-
-    Returns:
-        Tuple of (enhanced_summary_content, summary_embedding)
-    """
-    # Get model name from user_llm for token counting
-    model_name = getattr(user_llm, "model", "gpt-3.5-turbo")  # Fallback to default
-
-    # Optimize content to fit within context window
-    optimized_content = optimize_content_for_context_window(
-        content, document_metadata, model_name
-    )
-
-    summary_chain = SUMMARY_PROMPT_TEMPLATE | user_llm
-    content_with_metadata = f"<DOCUMENT><DOCUMENT_METADATA>\n\n{document_metadata}\n\n</DOCUMENT_METADATA>\n\n<DOCUMENT_CONTENT>\n\n{optimized_content}\n\n</DOCUMENT_CONTENT></DOCUMENT>"
-    summary_result = await summary_chain.ainvoke({"document": content_with_metadata})
-    summary_content = summary_result.content
-
-    # Combine summary with metadata if provided
-    if document_metadata:
-        metadata_parts = []
-        metadata_parts.append("# DOCUMENT METADATA")
-
-        for key, value in document_metadata.items():
-            if value:  # Only include non-empty values
-                formatted_key = key.replace("_", " ").title()
-                metadata_parts.append(f"**{formatted_key}:** {value}")
-
-        metadata_section = "\n".join(metadata_parts)
-        enhanced_summary_content = (
-            f"{metadata_section}\n\n# DOCUMENT SUMMARY\n\n{summary_content}"
-        )
-    else:
-        enhanced_summary_content = summary_content
-
-    summary_embedding = await asyncio.to_thread(embed_text, enhanced_summary_content)
-
-    return enhanced_summary_content, summary_embedding
-
-
 async def create_document_chunks(content: str) -> list[Chunk]:
     """
     Create chunks from document content.

surfsense_backend/tests/e2e/fakes/llm.py

@@ -7,13 +7,13 @@ The production indexing pipeline summarizes documents with:
     summary_content = summary_result.content
 
 The `llm` parameter is supplied per-document by
-`app.services.llm_service.get_user_long_context_llm`. We patch THAT
+`app.services.llm_service.get_agent_llm`. We patch THAT
 function to return a langchain-native FakeListChatModel so the rest of
 the chain works unchanged. No real LLM provider package is touched.
 
 Run-backend / run-celery use unittest.mock.patch.start() to install
 this at every binding site (the source module + every consumer that
-did `from app.services.llm_service import get_user_long_context_llm`
+did `from app.services.llm_service import get_agent_llm`
 at module load time).
 """
 
@@ -42,7 +42,7 @@ def _make_fake_llm() -> FakeListChatModel:
     return fake
 
 
-async def fake_get_user_long_context_llm(*args: Any, **kwargs: Any) -> Any:
+async def fake_get_agent_llm(*args: Any, **kwargs: Any) -> Any:
-    """Drop-in replacement for app.services.llm_service.get_user_long_context_llm."""
+    """Drop-in replacement for app.services.llm_service.get_agent_llm."""
     logger.info("[fake-llm] returning FakeListChatModel for E2E indexing")
     return _make_fake_llm()

surfsense_backend/tests/e2e/run_backend.py

@@ -206,23 +206,23 @@ def _patch_llm_bindings() -> None:
         fake_create_chat_litellm_from_agent_config,
         fake_create_chat_litellm_from_config,
     )
-    from tests.e2e.fakes.llm import fake_get_user_long_context_llm
+    from tests.e2e.fakes.llm import fake_get_agent_llm
 
     targets = [
-        "app.services.llm_service.get_user_long_context_llm",
+        "app.services.llm_service.get_agent_llm",
-        "app.tasks.connector_indexers.confluence_indexer.get_user_long_context_llm",
+        "app.tasks.connector_indexers.confluence_indexer.get_agent_llm",
-        "app.tasks.connector_indexers.google_drive_indexer.get_user_long_context_llm",
+        "app.tasks.connector_indexers.google_drive_indexer.get_agent_llm",
-        "app.tasks.connector_indexers.google_gmail_indexer.get_user_long_context_llm",
+        "app.tasks.connector_indexers.google_gmail_indexer.get_agent_llm",
-        "app.tasks.connector_indexers.notion_indexer.get_user_long_context_llm",
+        "app.tasks.connector_indexers.notion_indexer.get_agent_llm",
-        "app.tasks.connector_indexers.onedrive_indexer.get_user_long_context_llm",
+        "app.tasks.connector_indexers.onedrive_indexer.get_agent_llm",
-        "app.tasks.connector_indexers.dropbox_indexer.get_user_long_context_llm",
+        "app.tasks.connector_indexers.dropbox_indexer.get_agent_llm",
-        "app.tasks.connector_indexers.local_folder_indexer.get_user_long_context_llm",
+        "app.tasks.connector_indexers.local_folder_indexer.get_agent_llm",
-        "app.tasks.document_processors._save.get_user_long_context_llm",
+        "app.tasks.document_processors._save.get_agent_llm",
-        "app.tasks.document_processors.markdown_processor.get_user_long_context_llm",
+        "app.tasks.document_processors.markdown_processor.get_agent_llm",
     ]
     for target in targets:
         try:
-            p = patch(target, fake_get_user_long_context_llm)
+            p = patch(target, fake_get_agent_llm)
             p.start()
             _active_patches.append(p)
             logger.info("[fake-llm] patched %s", target)

surfsense_backend/tests/e2e/run_celery.py

@@ -183,23 +183,23 @@ def _patch_llm_bindings() -> None:
         fake_create_chat_litellm_from_agent_config,
         fake_create_chat_litellm_from_config,
     )
-    from tests.e2e.fakes.llm import fake_get_user_long_context_llm
+    from tests.e2e.fakes.llm import fake_get_agent_llm
 
     targets = [
-        "app.services.llm_service.get_user_long_context_llm",
+        "app.services.llm_service.get_agent_llm",
-        "app.tasks.connector_indexers.confluence_indexer.get_user_long_context_llm",
+        "app.tasks.connector_indexers.confluence_indexer.get_agent_llm",
-        "app.tasks.connector_indexers.google_drive_indexer.get_user_long_context_llm",
+        "app.tasks.connector_indexers.google_drive_indexer.get_agent_llm",
-        "app.tasks.connector_indexers.google_gmail_indexer.get_user_long_context_llm",
+        "app.tasks.connector_indexers.google_gmail_indexer.get_agent_llm",
-        "app.tasks.connector_indexers.notion_indexer.get_user_long_context_llm",
+        "app.tasks.connector_indexers.notion_indexer.get_agent_llm",
-        "app.tasks.connector_indexers.onedrive_indexer.get_user_long_context_llm",
+        "app.tasks.connector_indexers.onedrive_indexer.get_agent_llm",
-        "app.tasks.connector_indexers.dropbox_indexer.get_user_long_context_llm",
+        "app.tasks.connector_indexers.dropbox_indexer.get_agent_llm",
-        "app.tasks.connector_indexers.local_folder_indexer.get_user_long_context_llm",
+        "app.tasks.connector_indexers.local_folder_indexer.get_agent_llm",
-        "app.tasks.document_processors._save.get_user_long_context_llm",
+        "app.tasks.document_processors._save.get_agent_llm",
-        "app.tasks.document_processors.markdown_processor.get_user_long_context_llm",
+        "app.tasks.document_processors.markdown_processor.get_agent_llm",
     ]
     for target in targets:
         try:
-            p = patch(target, fake_get_user_long_context_llm)
+            p = patch(target, fake_get_agent_llm)
             p.start()
             _active_patches.append(p)
             logger.info("[fake-llm] patched %s in celery worker", target)

surfsense_backend/tests/integration/chat/test_thread_visibility.py

@@ -0,0 +1,279 @@
+"""Integration tests for new-chat thread visibility invariants.
+
+These tests exercise the route handlers directly with real DB-backed
+users, memberships, and permissions. The important contract is that a
+thread shared with a search space stays shared across normal metadata
+updates until the creator explicitly makes it private again.
+"""
+
+from __future__ import annotations
+
+import uuid
+
+import pytest
+import pytest_asyncio
+from fastapi import HTTPException
+from sqlalchemy import select
+from sqlalchemy.ext.asyncio import AsyncSession
+
+from app.db import (
+    ChatVisibility,
+    SearchSpace,
+    SearchSpaceMembership,
+    SearchSpaceRole,
+    User,
+)
+from app.routes import new_chat_routes
+from app.schemas.new_chat import (
+    NewChatThreadCreate,
+    NewChatThreadUpdate,
+    NewChatThreadVisibilityUpdate,
+)
+
+pytestmark = pytest.mark.integration
+
+
+@pytest_asyncio.fixture
+async def db_member(db_session: AsyncSession, db_search_space: SearchSpace) -> User:
+    member = User(
+        id=uuid.uuid4(),
+        email="member@surfsense.net",
+        hashed_password="hashed",
+        is_active=True,
+        is_superuser=False,
+        is_verified=True,
+    )
+    db_session.add(member)
+    await db_session.flush()
+
+    role = (
+        (
+            await db_session.execute(
+                select(SearchSpaceRole).where(
+                    SearchSpaceRole.search_space_id == db_search_space.id,
+                    SearchSpaceRole.name == "Editor",
+                )
+            )
+        )
+        .scalars()
+        .one()
+    )
+    db_session.add(
+        SearchSpaceMembership(
+            user_id=member.id,
+            search_space_id=db_search_space.id,
+            role_id=role.id,
+            is_owner=False,
+        )
+    )
+    await db_session.flush()
+    return member
+
+
+async def _create_thread(
+    db_session: AsyncSession,
+    db_user: User,
+    db_search_space: SearchSpace,
+    *,
+    title: str = "Visibility Invariant Chat",
+):
+    return await new_chat_routes.create_thread(
+        NewChatThreadCreate(
+            title=title,
+            archived=False,
+            search_space_id=db_search_space.id,
+            visibility=ChatVisibility.PRIVATE,
+        ),
+        session=db_session,
+        user=db_user,
+    )
+
+
+def _active_thread_ids(response) -> set[int]:
+    return {thread.id for thread in response.threads}
+
+
+def _search_thread_ids(response) -> set[int]:
+    return {thread.id for thread in response}
+
+
+async def test_private_thread_is_hidden_from_other_search_space_member(
+    db_session: AsyncSession,
+    db_user: User,
+    db_member: User,
+    db_search_space: SearchSpace,
+):
+    thread = await _create_thread(db_session, db_user, db_search_space)
+
+    member_threads = await new_chat_routes.list_threads(
+        search_space_id=db_search_space.id,
+        session=db_session,
+        user=db_member,
+    )
+    member_search = await new_chat_routes.search_threads(
+        search_space_id=db_search_space.id,
+        title="Visibility",
+        session=db_session,
+        user=db_member,
+    )
+
+    assert thread.id not in _active_thread_ids(member_threads)
+    assert thread.id not in _search_thread_ids(member_search)
+    with pytest.raises(HTTPException) as exc_info:
+        await new_chat_routes.get_thread_full(
+            thread_id=thread.id,
+            session=db_session,
+            user=db_member,
+        )
+    assert exc_info.value.status_code == 403
+
+
+async def test_creator_can_share_thread_and_member_can_list_search_read_it(
+    db_session: AsyncSession,
+    db_user: User,
+    db_member: User,
+    db_search_space: SearchSpace,
+):
+    thread = await _create_thread(db_session, db_user, db_search_space)
+
+    updated = await new_chat_routes.update_thread_visibility(
+        thread_id=thread.id,
+        visibility_update=NewChatThreadVisibilityUpdate(
+            visibility=ChatVisibility.SEARCH_SPACE,
+        ),
+        session=db_session,
+        user=db_user,
+    )
+
+    member_threads = await new_chat_routes.list_threads(
+        search_space_id=db_search_space.id,
+        session=db_session,
+        user=db_member,
+    )
+    member_search = await new_chat_routes.search_threads(
+        search_space_id=db_search_space.id,
+        title="Visibility",
+        session=db_session,
+        user=db_member,
+    )
+    full_thread = await new_chat_routes.get_thread_full(
+        thread_id=thread.id,
+        session=db_session,
+        user=db_member,
+    )
+
+    assert updated.visibility == ChatVisibility.SEARCH_SPACE
+    assert thread.id in _active_thread_ids(member_threads)
+    assert thread.id in _search_thread_ids(member_search)
+    assert full_thread["id"] == thread.id
+    assert full_thread["visibility"] == ChatVisibility.SEARCH_SPACE
+
+
+async def test_rename_and_archive_do_not_reset_shared_visibility(
+    db_session: AsyncSession,
+    db_user: User,
+    db_search_space: SearchSpace,
+):
+    thread = await _create_thread(db_session, db_user, db_search_space)
+    await new_chat_routes.update_thread_visibility(
+        thread_id=thread.id,
+        visibility_update=NewChatThreadVisibilityUpdate(
+            visibility=ChatVisibility.SEARCH_SPACE,
+        ),
+        session=db_session,
+        user=db_user,
+    )
+
+    renamed = await new_chat_routes.update_thread(
+        thread_id=thread.id,
+        thread_update=NewChatThreadUpdate(title="Renamed Shared Chat"),
+        session=db_session,
+        user=db_user,
+    )
+    archived = await new_chat_routes.update_thread(
+        thread_id=thread.id,
+        thread_update=NewChatThreadUpdate(archived=True),
+        session=db_session,
+        user=db_user,
+    )
+
+    assert renamed.visibility == ChatVisibility.SEARCH_SPACE
+    assert archived.visibility == ChatVisibility.SEARCH_SPACE
+    assert archived.archived is True
+
+
+async def test_non_creator_cannot_change_shared_thread_back_to_private(
+    db_session: AsyncSession,
+    db_user: User,
+    db_member: User,
+    db_search_space: SearchSpace,
+):
+    thread = await _create_thread(db_session, db_user, db_search_space)
+    await new_chat_routes.update_thread_visibility(
+        thread_id=thread.id,
+        visibility_update=NewChatThreadVisibilityUpdate(
+            visibility=ChatVisibility.SEARCH_SPACE,
+        ),
+        session=db_session,
+        user=db_user,
+    )
+
+    with pytest.raises(HTTPException) as exc_info:
+        await new_chat_routes.update_thread_visibility(
+            thread_id=thread.id,
+            visibility_update=NewChatThreadVisibilityUpdate(
+                visibility=ChatVisibility.PRIVATE,
+            ),
+            session=db_session,
+            user=db_member,
+        )
+
+    assert exc_info.value.status_code == 403
+
+
+async def test_creator_can_make_shared_thread_private_again(
+    db_session: AsyncSession,
+    db_user: User,
+    db_member: User,
+    db_search_space: SearchSpace,
+):
+    thread = await _create_thread(db_session, db_user, db_search_space)
+    await new_chat_routes.update_thread_visibility(
+        thread_id=thread.id,
+        visibility_update=NewChatThreadVisibilityUpdate(
+            visibility=ChatVisibility.SEARCH_SPACE,
+        ),
+        session=db_session,
+        user=db_user,
+    )
+
+    private_again = await new_chat_routes.update_thread_visibility(
+        thread_id=thread.id,
+        visibility_update=NewChatThreadVisibilityUpdate(
+            visibility=ChatVisibility.PRIVATE,
+        ),
+        session=db_session,
+        user=db_user,
+    )
+    member_threads = await new_chat_routes.list_threads(
+        search_space_id=db_search_space.id,
+        session=db_session,
+        user=db_member,
+    )
+    member_search = await new_chat_routes.search_threads(
+        search_space_id=db_search_space.id,
+        title="Visibility",
+        session=db_session,
+        user=db_member,
+    )
+
+    assert private_again.visibility == ChatVisibility.PRIVATE
+    assert thread.id not in _active_thread_ids(member_threads)
+    assert thread.id not in _search_thread_ids(member_search)
+    with pytest.raises(HTTPException) as exc_info:
+        await new_chat_routes.get_thread_full(
+            thread_id=thread.id,
+            session=db_session,
+            user=db_member,
+        )
+    assert exc_info.value.status_code == 403

surfsense_backend/tests/integration/conftest.py

@@ -1,7 +1,7 @@
 import importlib
 import sys
 import uuid
-from unittest.mock import AsyncMock, MagicMock
+from unittest.mock import MagicMock
 
 import pytest
 import pytest_asyncio
@@ -123,26 +123,6 @@ async def db_search_space(db_session: AsyncSession, db_user: User) -> SearchSpac
     return space
 
 
-@pytest.fixture
-def patched_summarize(monkeypatch) -> AsyncMock:
-    mock = AsyncMock(return_value="Mocked summary.")
-    monkeypatch.setattr(
-        "app.indexing_pipeline.indexing_pipeline_service.summarize_document",
-        mock,
-    )
-    return mock
-
-
-@pytest.fixture
-def patched_summarize_raises(monkeypatch) -> AsyncMock:
-    mock = AsyncMock(side_effect=RuntimeError("LLM unavailable"))
-    monkeypatch.setattr(
-        "app.indexing_pipeline.indexing_pipeline_service.summarize_document",
-        mock,
-    )
-    return mock
-
-
 @pytest.fixture
 def patched_embed_texts(monkeypatch) -> MagicMock:
     mock = MagicMock(side_effect=lambda texts: [[0.1] * _EMBEDDING_DIM for _ in texts])
@@ -153,6 +133,16 @@ def patched_embed_texts(monkeypatch) -> MagicMock:
     return mock
 
 
+@pytest.fixture
+def patched_embed_texts_raises(monkeypatch) -> MagicMock:
+    mock = MagicMock(side_effect=RuntimeError("Embedding unavailable"))
+    monkeypatch.setattr(
+        "app.indexing_pipeline.indexing_pipeline_service.embed_texts",
+        mock,
+    )
+    return mock
+
+
 @pytest.fixture
 def patched_chunk_text(monkeypatch) -> MagicMock:
     mock = MagicMock(return_value=["Test chunk content."])

surfsense_backend/tests/integration/document_upload/conftest.py

@@ -68,7 +68,6 @@ class InlineTaskDispatcher:
         filename: str,
         search_space_id: int,
         user_id: str,
-        should_summarize: bool = False,
         use_vision_llm: bool = False,
         processing_mode: str = "basic",
     ) -> None:
@@ -83,7 +82,6 @@ class InlineTaskDispatcher:
                 filename,
                 search_space_id,
                 user_id,
-                should_summarize=should_summarize,
                 use_vision_llm=use_vision_llm,
                 processing_mode=processing_mode,
             )
@@ -266,10 +264,6 @@ async def page_limits():
 @pytest.fixture(autouse=True)
 def _mock_external_apis(monkeypatch):
     """Mock LLM, embedding, and chunking — these are external API boundaries."""
-    monkeypatch.setattr(
-        "app.indexing_pipeline.indexing_pipeline_service.summarize_document",
-        AsyncMock(return_value="Mocked summary."),
-    )
     monkeypatch.setattr(
         "app.indexing_pipeline.indexing_pipeline_service.embed_texts",
         MagicMock(side_effect=lambda texts: [[0.1] * _EMBEDDING_DIM for _ in texts]),

surfsense_backend/tests/integration/indexing_pipeline/adapters/test_file_upload_adapter.py

@@ -8,7 +8,7 @@ pytestmark = pytest.mark.integration
 
 
 @pytest.mark.usefixtures(
-    "patched_summarize", "patched_embed_texts", "patched_chunk_text"
+"patched_embed_texts", "patched_chunk_text"
 )
 async def test_sets_status_ready(db_session, db_search_space, db_user, mocker):
     """Document status is READY after successful indexing."""
@@ -19,7 +19,6 @@ async def test_sets_status_ready(db_session, db_search_space, db_user, mocker):
         etl_service="UNSTRUCTURED",
         search_space_id=db_search_space.id,
         user_id=str(db_user.id),
-        llm=mocker.Mock(),
     )
 
     result = await db_session.execute(
@@ -31,10 +30,10 @@ async def test_sets_status_ready(db_session, db_search_space, db_user, mocker):
 
 
 @pytest.mark.usefixtures(
-    "patched_summarize", "patched_embed_texts", "patched_chunk_text"
+"patched_embed_texts", "patched_chunk_text"
 )
-async def test_content_is_summary(db_session, db_search_space, db_user, mocker):
+async def test_content_is_source_markdown(db_session, db_search_space, db_user, mocker):
-    """Document content is set to the LLM-generated summary."""
+    """Document content is set to the extracted source markdown."""
     adapter = UploadDocumentAdapter(db_session)
     await adapter.index(
         markdown_content="## Hello\n\nSome content.",
@@ -42,8 +41,6 @@ async def test_content_is_summary(db_session, db_search_space, db_user, mocker):
         etl_service="UNSTRUCTURED",
         search_space_id=db_search_space.id,
         user_id=str(db_user.id),
-        llm=mocker.Mock(),
-        should_summarize=True,
     )
 
     result = await db_session.execute(
@@ -51,11 +48,11 @@ async def test_content_is_summary(db_session, db_search_space, db_user, mocker):
     )
     document = result.scalars().first()
 
-    assert document.content == "Mocked summary."
+    assert document.content == "## Hello\n\nSome content."
 
 
 @pytest.mark.usefixtures(
-    "patched_summarize", "patched_embed_texts", "patched_chunk_text"
+"patched_embed_texts", "patched_chunk_text"
 )
 async def test_chunks_written_to_db(db_session, db_search_space, db_user, mocker):
     """Chunks derived from the source markdown are persisted in the DB."""
@@ -66,7 +63,6 @@ async def test_chunks_written_to_db(db_session, db_search_space, db_user, mocker
         etl_service="UNSTRUCTURED",
         search_space_id=db_search_space.id,
         user_id=str(db_user.id),
-        llm=mocker.Mock(),
     )
 
     result = await db_session.execute(
@@ -83,9 +79,7 @@ async def test_chunks_written_to_db(db_session, db_search_space, db_user, mocker
     assert chunks[0].content == "Test chunk content."
 
 
-@pytest.mark.usefixtures(
+@pytest.mark.usefixtures("patched_embed_texts_raises", "patched_chunk_text")
-    "patched_summarize_raises", "patched_embed_texts", "patched_chunk_text"
-)
 async def test_raises_on_indexing_failure(db_session, db_search_space, db_user, mocker):
     """RuntimeError is raised when the indexing step fails so the caller can fire a failure notification."""
     adapter = UploadDocumentAdapter(db_session)
@@ -96,8 +90,6 @@ async def test_raises_on_indexing_failure(db_session, db_search_space, db_user,
             etl_service="UNSTRUCTURED",
             search_space_id=db_search_space.id,
             user_id=str(db_user.id),
-            llm=mocker.Mock(),
-            should_summarize=True,
         )
 
 
@@ -107,10 +99,10 @@ async def test_raises_on_indexing_failure(db_session, db_search_space, db_user,
 
 
 @pytest.mark.usefixtures(
-    "patched_summarize", "patched_embed_texts", "patched_chunk_text"
+"patched_embed_texts", "patched_chunk_text"
 )
 async def test_reindex_updates_content(db_session, db_search_space, db_user, mocker):
-    """Document content is updated to the new summary after reindexing."""
+    """Document content is updated to the new source markdown after reindexing."""
     adapter = UploadDocumentAdapter(db_session)
     await adapter.index(
         markdown_content="## Original\n\nOriginal content.",
@@ -118,7 +110,6 @@ async def test_reindex_updates_content(db_session, db_search_space, db_user, moc
         etl_service="UNSTRUCTURED",
         search_space_id=db_search_space.id,
         user_id=str(db_user.id),
-        llm=mocker.Mock(),
     )
 
     result = await db_session.execute(
@@ -129,14 +120,14 @@ async def test_reindex_updates_content(db_session, db_search_space, db_user, moc
     document.source_markdown = "## Edited\n\nNew content after user edit."
     await db_session.flush()
 
-    await adapter.reindex(document=document, llm=mocker.Mock())
+    await adapter.reindex(document=document)
 
     await db_session.refresh(document)
-    assert document.content == "Mocked summary."
+    assert document.content == "## Edited\n\nNew content after user edit."
 
 
 @pytest.mark.usefixtures(
-    "patched_summarize", "patched_embed_texts", "patched_chunk_text"
+"patched_embed_texts", "patched_chunk_text"
 )
 async def test_reindex_updates_content_hash(
     db_session, db_search_space, db_user, mocker
@@ -149,7 +140,6 @@ async def test_reindex_updates_content_hash(
         etl_service="UNSTRUCTURED",
         search_space_id=db_search_space.id,
         user_id=str(db_user.id),
-        llm=mocker.Mock(),
     )
 
     result = await db_session.execute(
@@ -161,14 +151,14 @@ async def test_reindex_updates_content_hash(
     document.source_markdown = "## Edited\n\nNew content after user edit."
     await db_session.flush()
 
-    await adapter.reindex(document=document, llm=mocker.Mock())
+    await adapter.reindex(document=document)
 
     await db_session.refresh(document)
     assert document.content_hash != original_hash
 
 
 @pytest.mark.usefixtures(
-    "patched_summarize", "patched_embed_texts", "patched_chunk_text"
+"patched_embed_texts", "patched_chunk_text"
 )
 async def test_reindex_sets_status_ready(db_session, db_search_space, db_user, mocker):
     """Document status is READY after successful reindexing."""
@@ -179,7 +169,6 @@ async def test_reindex_sets_status_ready(db_session, db_search_space, db_user, m
         etl_service="UNSTRUCTURED",
         search_space_id=db_search_space.id,
         user_id=str(db_user.id),
-        llm=mocker.Mock(),
     )
 
     result = await db_session.execute(
@@ -190,13 +179,13 @@ async def test_reindex_sets_status_ready(db_session, db_search_space, db_user, m
     document.source_markdown = "## Edited\n\nNew content after user edit."
     await db_session.flush()
 
-    await adapter.reindex(document=document, llm=mocker.Mock())
+    await adapter.reindex(document=document)
 
     await db_session.refresh(document)
     assert DocumentStatus.is_state(document.status, DocumentStatus.READY)
 
 
-@pytest.mark.usefixtures("patched_summarize", "patched_embed_texts")
+@pytest.mark.usefixtures("patched_embed_texts")
 async def test_reindex_replaces_chunks(db_session, db_search_space, db_user, mocker):
     """Reindexing replaces old chunks with new content rather than appending."""
     mocker.patch(
@@ -211,7 +200,6 @@ async def test_reindex_replaces_chunks(db_session, db_search_space, db_user, moc
         etl_service="UNSTRUCTURED",
         search_space_id=db_search_space.id,
         user_id=str(db_user.id),
-        llm=mocker.Mock(),
     )
 
     result = await db_session.execute(
@@ -223,7 +211,7 @@ async def test_reindex_replaces_chunks(db_session, db_search_space, db_user, moc
     document.source_markdown = "## Edited\n\nNew content after user edit."
     await db_session.flush()
 
-    await adapter.reindex(document=document, llm=mocker.Mock())
+    await adapter.reindex(document=document)
 
     chunks_result = await db_session.execute(
         select(Chunk).filter(Chunk.document_id == document_id)
@@ -235,7 +223,7 @@ async def test_reindex_replaces_chunks(db_session, db_search_space, db_user, moc
 
 
 @pytest.mark.usefixtures(
-    "patched_summarize", "patched_embed_texts", "patched_chunk_text"
+"patched_embed_texts", "patched_chunk_text"
 )
 async def test_reindex_clears_reindexing_flag(
     db_session, db_search_space, db_user, mocker
@@ -248,7 +236,6 @@ async def test_reindex_clears_reindexing_flag(
         etl_service="UNSTRUCTURED",
         search_space_id=db_search_space.id,
         user_id=str(db_user.id),
-        llm=mocker.Mock(),
     )
 
     result = await db_session.execute(
@@ -260,19 +247,17 @@ async def test_reindex_clears_reindexing_flag(
     document.content_needs_reindexing = True
     await db_session.flush()
 
-    await adapter.reindex(document=document, llm=mocker.Mock())
+    await adapter.reindex(document=document)
 
     await db_session.refresh(document)
     assert document.content_needs_reindexing is False
 
 
 @pytest.mark.usefixtures("patched_embed_texts", "patched_chunk_text")
-async def test_reindex_raises_on_failure(db_session, db_search_space, db_user, mocker):
+async def test_reindex_raises_on_failure(
+    db_session, db_search_space, db_user, patched_embed_texts, mocker
+):
     """RuntimeError is raised when reindexing fails so the caller can handle it."""
-    mocker.patch(
-        "app.indexing_pipeline.indexing_pipeline_service.summarize_document",
-        return_value="Mocked summary.",
-    )
 
     adapter = UploadDocumentAdapter(db_session)
     await adapter.index(
@@ -281,7 +266,6 @@ async def test_reindex_raises_on_failure(db_session, db_search_space, db_user, m
         etl_service="UNSTRUCTURED",
         search_space_id=db_search_space.id,
         user_id=str(db_user.id),
-        llm=mocker.Mock(),
     )
 
     result = await db_session.execute(
@@ -292,13 +276,10 @@ async def test_reindex_raises_on_failure(db_session, db_search_space, db_user, m
     document.source_markdown = "## Edited\n\nNew content after user edit."
     await db_session.flush()
 
-    mocker.patch(
+    patched_embed_texts.side_effect = RuntimeError("Embedding unavailable")
-        "app.indexing_pipeline.indexing_pipeline_service.summarize_document",
-        side_effect=RuntimeError("LLM unavailable"),
-    )
 
     with pytest.raises(RuntimeError, match=r"Embedding failed|Reindexing failed"):
-        await adapter.reindex(document=document, llm=mocker.Mock())
+        await adapter.reindex(document=document)
 
 
 async def test_reindex_raises_on_empty_source_markdown(
@@ -323,4 +304,4 @@ async def test_reindex_raises_on_empty_source_markdown(
     adapter = UploadDocumentAdapter(db_session)
 
     with pytest.raises(RuntimeError, match="no source_markdown"):
-        await adapter.reindex(document=document, llm=mocker.Mock())
+        await adapter.reindex(document=document)

surfsense_backend/tests/integration/indexing_pipeline/test_calendar_pipeline.py

@@ -25,8 +25,6 @@ def _cal_doc(
         search_space_id=search_space_id,
         connector_id=connector_id,
         created_by_id=user_id,
-        should_summarize=True,
-        fallback_summary=f"Calendar: Event {unique_id}",
         metadata={
             "event_id": unique_id,
             "start_time": "2025-01-15T10:00:00",
@@ -37,7 +35,7 @@ def _cal_doc(
 
 
 @pytest.mark.usefixtures(
-    "patched_summarize", "patched_embed_texts", "patched_chunk_text"
+"patched_embed_texts", "patched_chunk_text"
 )
 async def test_calendar_pipeline_creates_ready_document(
     db_session, db_search_space, db_connector, db_user, mocker
@@ -55,7 +53,7 @@ async def test_calendar_pipeline_creates_ready_document(
     prepared = await service.prepare_for_indexing([doc])
     assert len(prepared) == 1
 
-    await service.index(prepared[0], doc, llm=mocker.Mock())
+    await service.index(prepared[0], doc)
 
     result = await db_session.execute(
         select(Document).filter(Document.search_space_id == space_id)
@@ -68,7 +66,7 @@ async def test_calendar_pipeline_creates_ready_document(
 
 
 @pytest.mark.usefixtures(
-    "patched_summarize", "patched_embed_texts", "patched_chunk_text"
+"patched_embed_texts", "patched_chunk_text"
 )
 async def test_calendar_legacy_doc_migrated(
     db_session, db_search_space, db_connector, db_user, mocker

surfsense_backend/tests/integration/indexing_pipeline/test_drive_pipeline.py

@@ -25,8 +25,6 @@ def _drive_doc(
         search_space_id=search_space_id,
         connector_id=connector_id,
         created_by_id=user_id,
-        should_summarize=True,
-        fallback_summary=f"File: {unique_id}.pdf",
         metadata={
             "google_drive_file_id": unique_id,
             "google_drive_file_name": f"{unique_id}.pdf",
@@ -36,7 +34,7 @@ def _drive_doc(
 
 
 @pytest.mark.usefixtures(
-    "patched_summarize", "patched_embed_texts", "patched_chunk_text"
+"patched_embed_texts", "patched_chunk_text"
 )
 async def test_drive_pipeline_creates_ready_document(
     db_session, db_search_space, db_connector, db_user, mocker
@@ -54,7 +52,7 @@ async def test_drive_pipeline_creates_ready_document(
     prepared = await service.prepare_for_indexing([doc])
     assert len(prepared) == 1
 
-    await service.index(prepared[0], doc, llm=mocker.Mock())
+    await service.index(prepared[0], doc)
 
     result = await db_session.execute(
         select(Document).filter(Document.search_space_id == space_id)
@@ -67,7 +65,7 @@ async def test_drive_pipeline_creates_ready_document(
 
 
 @pytest.mark.usefixtures(
-    "patched_summarize", "patched_embed_texts", "patched_chunk_text"
+"patched_embed_texts", "patched_chunk_text"
 )
 async def test_drive_legacy_doc_migrated(
     db_session, db_search_space, db_connector, db_user, mocker

surfsense_backend/tests/integration/indexing_pipeline/test_dropbox_pipeline.py

@@ -24,8 +24,6 @@ def _dropbox_doc(
         search_space_id=search_space_id,
         connector_id=connector_id,
         created_by_id=user_id,
-        should_summarize=True,
-        fallback_summary=f"File: {unique_id}.docx",
         metadata={
             "dropbox_file_id": unique_id,
             "dropbox_file_name": f"{unique_id}.docx",
@@ -35,7 +33,7 @@ def _dropbox_doc(
 
 
 @pytest.mark.usefixtures(
-    "patched_summarize", "patched_embed_texts", "patched_chunk_text"
+"patched_embed_texts", "patched_chunk_text"
 )
 async def test_dropbox_pipeline_creates_ready_document(
     db_session, db_search_space, db_connector, db_user, mocker
@@ -53,7 +51,7 @@ async def test_dropbox_pipeline_creates_ready_document(
     prepared = await service.prepare_for_indexing([doc])
     assert len(prepared) == 1
 
-    await service.index(prepared[0], doc, llm=mocker.Mock())
+    await service.index(prepared[0], doc)
 
     result = await db_session.execute(
         select(Document).filter(Document.search_space_id == space_id)
@@ -66,7 +64,7 @@ async def test_dropbox_pipeline_creates_ready_document(
 
 
 @pytest.mark.usefixtures(
-    "patched_summarize", "patched_embed_texts", "patched_chunk_text"
+"patched_embed_texts", "patched_chunk_text"
 )
 async def test_dropbox_duplicate_content_skipped(
     db_session, db_search_space, db_connector, db_user, mocker
@@ -86,7 +84,7 @@ async def test_dropbox_duplicate_content_skipped(
 
     prepared = await service.prepare_for_indexing([doc])
     assert len(prepared) == 1
-    await service.index(prepared[0], doc, llm=mocker.Mock())
+    await service.index(prepared[0], doc)
 
     result = await db_session.execute(
         select(Document).filter(Document.search_space_id == space_id)

surfsense_backend/tests/integration/indexing_pipeline/test_gmail_pipeline.py

@@ -28,8 +28,6 @@ def _gmail_doc(
         search_space_id=search_space_id,
         connector_id=connector_id,
         created_by_id=user_id,
-        should_summarize=True,
-        fallback_summary=f"Gmail: Subject for {unique_id}",
         metadata={
             "message_id": unique_id,
             "from": "sender@example.com",
@@ -39,7 +37,7 @@ def _gmail_doc(
 
 
 @pytest.mark.usefixtures(
-    "patched_summarize", "patched_embed_texts", "patched_chunk_text"
+"patched_embed_texts", "patched_chunk_text"
 )
 async def test_gmail_pipeline_creates_ready_document(
     db_session, db_search_space, db_connector, db_user, mocker
@@ -57,7 +55,7 @@ async def test_gmail_pipeline_creates_ready_document(
     prepared = await service.prepare_for_indexing([doc])
     assert len(prepared) == 1
 
-    await service.index(prepared[0], doc, llm=mocker.Mock())
+    await service.index(prepared[0], doc)
 
     result = await db_session.execute(
         select(Document).filter(Document.search_space_id == space_id)
@@ -71,7 +69,7 @@ async def test_gmail_pipeline_creates_ready_document(
 
 
 @pytest.mark.usefixtures(
-    "patched_summarize", "patched_embed_texts", "patched_chunk_text"
+"patched_embed_texts", "patched_chunk_text"
 )
 async def test_gmail_legacy_doc_migrated_then_reused(
     db_session, db_search_space, db_connector, db_user, mocker

surfsense_backend/tests/integration/indexing_pipeline/test_index_batch.py

@@ -10,7 +10,7 @@ pytestmark = pytest.mark.integration
 
 
 @pytest.mark.usefixtures(
-    "patched_summarize", "patched_embed_texts", "patched_chunk_text"
+"patched_embed_texts", "patched_chunk_text"
 )
 async def test_index_batch_creates_ready_documents(
     db_session, db_search_space, make_connector_document, mocker
@@ -33,7 +33,7 @@ async def test_index_batch_creates_ready_documents(
     ]
 
     service = IndexingPipelineService(session=db_session)
-    results = await service.index_batch(docs, llm=mocker.Mock())
+    results = await service.index_batch(docs)
 
     assert len(results) == 2
 
@@ -50,10 +50,10 @@ async def test_index_batch_creates_ready_documents(
 
 
 @pytest.mark.usefixtures(
-    "patched_summarize", "patched_embed_texts", "patched_chunk_text"
+"patched_embed_texts", "patched_chunk_text"
 )
 async def test_index_batch_empty_returns_empty(db_session, mocker):
     """index_batch with empty input returns an empty list."""
     service = IndexingPipelineService(session=db_session)
-    results = await service.index_batch([], llm=mocker.Mock())
+    results = await service.index_batch([])
     assert results == []

surfsense_backend/tests/integration/indexing_pipeline/test_index_document.py

@@ -10,9 +10,7 @@ _EMBEDDING_DIM = app_config.embedding_model_instance.dimension
 pytestmark = pytest.mark.integration
 
 
-@pytest.mark.usefixtures(
+@pytest.mark.usefixtures("patched_embed_texts", "patched_chunk_text")
-    "patched_summarize", "patched_embed_texts", "patched_chunk_text"
-)
 async def test_sets_status_ready(
     db_session,
     db_search_space,
@@ -27,7 +25,7 @@ async def test_sets_status_ready(
     document = prepared[0]
     document_id = document.id
 
-    await service.index(document, connector_doc, llm=mocker.Mock())
+    await service.index(document, connector_doc)
 
     result = await db_session.execute(
         select(Document).filter(Document.id == document_id)
@@ -37,16 +35,14 @@ async def test_sets_status_ready(
     assert DocumentStatus.is_state(reloaded.status, DocumentStatus.READY)
 
 
-@pytest.mark.usefixtures(
+@pytest.mark.usefixtures("patched_embed_texts", "patched_chunk_text")
-    "patched_summarize", "patched_embed_texts", "patched_chunk_text"
+async def test_content_is_source_markdown_by_default(
-)
-async def test_content_is_summary_when_should_summarize_true(
     db_session,
     db_search_space,
     make_connector_document,
     mocker,
 ):
-    """Document content is set to the LLM-generated summary when should_summarize=True."""
+    """Document content is set to source_markdown by default."""
     connector_doc = make_connector_document(search_space_id=db_search_space.id)
     service = IndexingPipelineService(session=db_session)
 
@@ -54,28 +50,25 @@ async def test_content_is_summary_when_should_summarize_true(
     document = prepared[0]
     document_id = document.id
 
-    await service.index(document, connector_doc, llm=mocker.Mock())
+    await service.index(document, connector_doc)
 
     result = await db_session.execute(
         select(Document).filter(Document.id == document_id)
     )
     reloaded = result.scalars().first()
 
-    assert reloaded.content == "Mocked summary."
+    assert reloaded.content == connector_doc.source_markdown
 
 
-@pytest.mark.usefixtures(
+@pytest.mark.usefixtures("patched_embed_texts", "patched_chunk_text")
-    "patched_summarize", "patched_embed_texts", "patched_chunk_text"
+async def test_content_is_source_markdown_when_custom_content(
-)
-async def test_content_is_source_markdown_when_should_summarize_false(
     db_session,
     db_search_space,
     make_connector_document,
 ):
-    """Document content is set to source_markdown verbatim when should_summarize=False."""
+    """Document content is set to source_markdown verbatim."""
     connector_doc = make_connector_document(
         search_space_id=db_search_space.id,
-        should_summarize=False,
         source_markdown="## Raw content",
     )
     service = IndexingPipelineService(session=db_session)
@@ -84,7 +77,7 @@ async def test_content_is_source_markdown_when_should_summarize_false(
     document = prepared[0]
     document_id = document.id
 
-    await service.index(document, connector_doc, llm=None)
+    await service.index(document, connector_doc)
 
     result = await db_session.execute(
         select(Document).filter(Document.id == document_id)
@@ -94,9 +87,7 @@ async def test_content_is_source_markdown_when_should_summarize_false(
     assert reloaded.content == "## Raw content"
 
 
-@pytest.mark.usefixtures(
+@pytest.mark.usefixtures("patched_embed_texts", "patched_chunk_text")
-    "patched_summarize", "patched_embed_texts", "patched_chunk_text"
-)
 async def test_chunks_written_to_db(
     db_session,
     db_search_space,
@@ -111,7 +102,7 @@ async def test_chunks_written_to_db(
     document = prepared[0]
     document_id = document.id
 
-    await service.index(document, connector_doc, llm=mocker.Mock())
+    await service.index(document, connector_doc)
 
     result = await db_session.execute(
         select(Chunk).filter(Chunk.document_id == document_id)
@@ -122,9 +113,7 @@ async def test_chunks_written_to_db(
     assert chunks[0].content == "Test chunk content."
 
 
-@pytest.mark.usefixtures(
+@pytest.mark.usefixtures("patched_embed_texts", "patched_chunk_text")
-    "patched_summarize", "patched_embed_texts", "patched_chunk_text"
-)
 async def test_embedding_written_to_db(
     db_session,
     db_search_space,
@@ -139,7 +128,7 @@ async def test_embedding_written_to_db(
     document = prepared[0]
     document_id = document.id
 
-    await service.index(document, connector_doc, llm=mocker.Mock())
+    await service.index(document, connector_doc)
 
     result = await db_session.execute(
         select(Document).filter(Document.id == document_id)
@@ -150,9 +139,7 @@ async def test_embedding_written_to_db(
     assert len(reloaded.embedding) == _EMBEDDING_DIM
 
 
-@pytest.mark.usefixtures(
+@pytest.mark.usefixtures("patched_embed_texts", "patched_chunk_text")
-    "patched_summarize", "patched_embed_texts", "patched_chunk_text"
-)
 async def test_updated_at_advances_after_indexing(
     db_session,
     db_search_space,
@@ -172,7 +159,7 @@ async def test_updated_at_advances_after_indexing(
     )
     updated_at_pending = result.scalars().first().updated_at
 
-    await service.index(document, connector_doc, llm=mocker.Mock())
+    await service.index(document, connector_doc)
 
     result = await db_session.execute(
         select(Document).filter(Document.id == document_id)
@@ -182,18 +169,15 @@ async def test_updated_at_advances_after_indexing(
     assert updated_at_ready > updated_at_pending
 
 
-@pytest.mark.usefixtures(
+@pytest.mark.usefixtures("patched_embed_texts", "patched_chunk_text")
-    "patched_summarize", "patched_embed_texts", "patched_chunk_text"
-)
 async def test_no_llm_falls_back_to_source_markdown(
     db_session,
     db_search_space,
     make_connector_document,
 ):
-    """When llm=None and no fallback_summary, content falls back to source_markdown."""
+    """Content stays deterministic source markdown without an LLM."""
     connector_doc = make_connector_document(
         search_space_id=db_search_space.id,
-        should_summarize=True,
         source_markdown="## Fallback content",
     )
     service = IndexingPipelineService(session=db_session)
@@ -202,7 +186,7 @@ async def test_no_llm_falls_back_to_source_markdown(
     document = prepared[0]
     document_id = document.id
 
-    await service.index(document, connector_doc, llm=None)
+    await service.index(document, connector_doc)
 
     result = await db_session.execute(
         select(Document).filter(Document.id == document_id)
@@ -213,27 +197,23 @@ async def test_no_llm_falls_back_to_source_markdown(
     assert reloaded.content == "## Fallback content"
 
 
-@pytest.mark.usefixtures(
+@pytest.mark.usefixtures("patched_embed_texts", "patched_chunk_text")
-    "patched_summarize", "patched_embed_texts", "patched_chunk_text"
+async def test_source_markdown_used_without_preview(
-)
-async def test_fallback_summary_used_when_llm_unavailable(
     db_session,
     db_search_space,
     make_connector_document,
 ):
-    """fallback_summary is used as content when llm=None and should_summarize=True."""
+    """Source markdown is used without fallback preview fields."""
     connector_doc = make_connector_document(
         search_space_id=db_search_space.id,
-        should_summarize=True,
         source_markdown="## Full raw content",
-        fallback_summary="Short pre-built summary.",
     )
     service = IndexingPipelineService(session=db_session)
 
     prepared = await service.prepare_for_indexing([connector_doc])
     document_id = prepared[0].id
 
-    await service.index(prepared[0], connector_doc, llm=None)
+    await service.index(prepared[0], connector_doc)
 
     result = await db_session.execute(
         select(Document).filter(Document.id == document_id)
@@ -241,12 +221,10 @@ async def test_fallback_summary_used_when_llm_unavailable(
     reloaded = result.scalars().first()
 
     assert DocumentStatus.is_state(reloaded.status, DocumentStatus.READY)
-    assert reloaded.content == "Short pre-built summary."
+    assert reloaded.content == "## Full raw content"
 
 
-@pytest.mark.usefixtures(
+@pytest.mark.usefixtures("patched_embed_texts", "patched_chunk_text")
-    "patched_summarize", "patched_embed_texts", "patched_chunk_text"
-)
 async def test_reindex_replaces_old_chunks(
     db_session,
     db_search_space,
@@ -264,14 +242,14 @@ async def test_reindex_replaces_old_chunks(
     document = prepared[0]
     document_id = document.id
 
-    await service.index(document, connector_doc, llm=mocker.Mock())
+    await service.index(document, connector_doc)
 
     updated_doc = make_connector_document(
         search_space_id=db_search_space.id,
         source_markdown="## v2",
     )
     re_prepared = await service.prepare_for_indexing([updated_doc])
-    await service.index(re_prepared[0], updated_doc, llm=mocker.Mock())
+    await service.index(re_prepared[0], updated_doc)
 
     result = await db_session.execute(
         select(Chunk).filter(Chunk.document_id == document_id)
@@ -281,16 +259,14 @@ async def test_reindex_replaces_old_chunks(
     assert len(chunks) == 1
 
 
-@pytest.mark.usefixtures(
+@pytest.mark.usefixtures("patched_embed_texts_raises", "patched_chunk_text")
-    "patched_summarize_raises", "patched_embed_texts", "patched_chunk_text"
+async def test_embedding_error_sets_status_failed(
-)
-async def test_llm_error_sets_status_failed(
     db_session,
     db_search_space,
     make_connector_document,
     mocker,
 ):
-    """Document status is FAILED when the LLM raises during indexing."""
+    """Document status is FAILED when embedding raises during indexing."""
     connector_doc = make_connector_document(search_space_id=db_search_space.id)
     service = IndexingPipelineService(session=db_session)
 
@@ -298,7 +274,7 @@ async def test_llm_error_sets_status_failed(
     document = prepared[0]
     document_id = document.id
 
-    await service.index(document, connector_doc, llm=mocker.Mock())
+    await service.index(document, connector_doc)
 
     result = await db_session.execute(
         select(Document).filter(Document.id == document_id)
@@ -308,10 +284,8 @@ async def test_llm_error_sets_status_failed(
     assert DocumentStatus.is_state(reloaded.status, DocumentStatus.FAILED)
 
 
-@pytest.mark.usefixtures(
+@pytest.mark.usefixtures("patched_embed_texts_raises", "patched_chunk_text")
-    "patched_summarize_raises", "patched_embed_texts", "patched_chunk_text"
+async def test_embedding_error_leaves_no_partial_data(
-)
-async def test_llm_error_leaves_no_partial_data(
     db_session,
     db_search_space,
     make_connector_document,
@@ -325,7 +299,7 @@ async def test_llm_error_leaves_no_partial_data(
     document = prepared[0]
     document_id = document.id
 
-    await service.index(document, connector_doc, llm=mocker.Mock())
+    await service.index(document, connector_doc)
 
     result = await db_session.execute(
         select(Document).filter(Document.id == document_id)

surfsense_backend/tests/integration/indexing_pipeline/test_local_folder_pipeline.py

@@ -21,7 +21,6 @@ from app.db import (
 pytestmark = pytest.mark.integration
 
 UNIFIED_FIXTURES = (
-    "patched_summarize",
     "patched_embed_texts",
     "patched_chunk_text",
 )
@@ -787,7 +786,7 @@ class TestPipelineIntegration:
         assert len(prepared) == 1
 
         db_doc = prepared[0]
-        result = await service.index(db_doc, doc, llm=mocker.Mock())
+        result = await service.index(db_doc, doc)
         assert result is not None
 
         docs = (
@@ -1272,7 +1271,7 @@ class TestIndexingProgressFlag:
         original_index = IndexingPipelineService.index
         flag_observed = []
 
-        async def patched_index(self_pipe, document, connector_doc, llm):
+        async def patched_index(self_pipe, document, connector_doc):
             folder = (
                 await db_session.execute(
                     select(Folder).where(
@@ -1284,7 +1283,7 @@ class TestIndexingProgressFlag:
             if folder:
                 meta = folder.folder_metadata or {}
                 flag_observed.append(meta.get("indexing_in_progress", False))
-            return await original_index(self_pipe, document, connector_doc, llm)
+            return await original_index(self_pipe, document, connector_doc)
 
         IndexingPipelineService.index = patched_index
         try:

surfsense_backend/tests/integration/indexing_pipeline/test_onedrive_pipeline.py

@@ -24,8 +24,6 @@ def _onedrive_doc(
         search_space_id=search_space_id,
         connector_id=connector_id,
         created_by_id=user_id,
-        should_summarize=True,
-        fallback_summary=f"File: {unique_id}.docx",
         metadata={
             "onedrive_file_id": unique_id,
             "onedrive_file_name": f"{unique_id}.docx",
@@ -35,7 +33,7 @@ def _onedrive_doc(
 
 
 @pytest.mark.usefixtures(
-    "patched_summarize", "patched_embed_texts", "patched_chunk_text"
+"patched_embed_texts", "patched_chunk_text"
 )
 async def test_onedrive_pipeline_creates_ready_document(
     db_session, db_search_space, db_connector, db_user, mocker
@@ -53,7 +51,7 @@ async def test_onedrive_pipeline_creates_ready_document(
     prepared = await service.prepare_for_indexing([doc])
     assert len(prepared) == 1
 
-    await service.index(prepared[0], doc, llm=mocker.Mock())
+    await service.index(prepared[0], doc)
 
     result = await db_session.execute(
         select(Document).filter(Document.search_space_id == space_id)
@@ -66,7 +64,7 @@ async def test_onedrive_pipeline_creates_ready_document(
 
 
 @pytest.mark.usefixtures(
-    "patched_summarize", "patched_embed_texts", "patched_chunk_text"
+"patched_embed_texts", "patched_chunk_text"
 )
 async def test_onedrive_duplicate_content_skipped(
     db_session, db_search_space, db_connector, db_user, mocker
@@ -86,7 +84,7 @@ async def test_onedrive_duplicate_content_skipped(
 
     prepared = await service.prepare_for_indexing([doc])
     assert len(prepared) == 1
-    await service.index(prepared[0], doc, llm=mocker.Mock())
+    await service.index(prepared[0], doc)
 
     result = await db_session.execute(
         select(Document).filter(Document.search_space_id == space_id)

surfsense_backend/tests/integration/indexing_pipeline/test_prepare_for_indexing.py

@@ -33,7 +33,7 @@ async def test_new_document_is_persisted_with_pending_status(
 
 
 @pytest.mark.usefixtures(
-    "patched_summarize", "patched_embed_texts", "patched_chunk_text"
+"patched_embed_texts", "patched_chunk_text"
 )
 async def test_unchanged_ready_document_is_skipped(
     db_session,
@@ -47,7 +47,7 @@ async def test_unchanged_ready_document_is_skipped(
 
     # Index fully so the document reaches ready state
     prepared = await service.prepare_for_indexing([doc])
-    await service.index(prepared[0], doc, llm=mocker.Mock())
+    await service.index(prepared[0], doc)
 
     # Same content on the next run — a ready document must be skipped
     results = await service.prepare_for_indexing([doc])
@@ -56,7 +56,7 @@ async def test_unchanged_ready_document_is_skipped(
 
 
 @pytest.mark.usefixtures(
-    "patched_summarize", "patched_embed_texts", "patched_chunk_text"
+"patched_embed_texts", "patched_chunk_text"
 )
 async def test_title_only_change_updates_title_in_db(
     db_session,
@@ -72,7 +72,7 @@ async def test_title_only_change_updates_title_in_db(
 
     prepared = await service.prepare_for_indexing([original])
     document_id = prepared[0].id
-    await service.index(prepared[0], original, llm=mocker.Mock())
+    await service.index(prepared[0], original)
 
     renamed = make_connector_document(
         search_space_id=db_search_space.id, title="Updated Title"
@@ -338,9 +338,7 @@ async def test_same_content_from_different_source_is_skipped(
     assert len(result.scalars().all()) == 1
 
 
-@pytest.mark.usefixtures(
+@pytest.mark.usefixtures("patched_embed_texts_raises", "patched_chunk_text")
-    "patched_summarize_raises", "patched_embed_texts", "patched_chunk_text"
-)
 async def test_failed_document_with_unchanged_content_is_requeued(
     db_session,
     db_search_space,
@@ -351,10 +349,10 @@ async def test_failed_document_with_unchanged_content_is_requeued(
     doc = make_connector_document(search_space_id=db_search_space.id)
     service = IndexingPipelineService(session=db_session)
 
-    # First run: document is created and indexing crashes → status = failed
+    # First run: document is created and indexing crashes, so status becomes failed.
     prepared = await service.prepare_for_indexing([doc])
     document_id = prepared[0].id
-    await service.index(prepared[0], doc, llm=mocker.Mock())
+    await service.index(prepared[0], doc)
 
     result = await db_session.execute(
         select(Document).filter(Document.id == document_id)

surfsense_backend/tests/unit/agents/new_chat/tools/test_resume_page_limits.py

@@ -101,7 +101,7 @@ async def test_generate_resume_defaults_to_one_page_target(monkeypatch) -> None:
     llm = SimpleNamespace(ainvoke=AsyncMock(side_effect=_llm_invoke))
     monkeypatch.setattr(
         resume_tool,
-        "get_document_summary_llm",
+        "get_agent_llm",
         AsyncMock(return_value=llm),
     )
     monkeypatch.setattr(resume_tool, "_compile_typst", lambda _source: b"pdf")
@@ -130,7 +130,7 @@ async def test_generate_resume_compresses_when_over_limit(monkeypatch) -> None:
     llm = SimpleNamespace(ainvoke=AsyncMock(side_effect=responses))
     monkeypatch.setattr(
         resume_tool,
-        "get_document_summary_llm",
+        "get_agent_llm",
         AsyncMock(return_value=llm),
     )
     monkeypatch.setattr(resume_tool, "_compile_typst", lambda _source: b"pdf")
@@ -165,7 +165,7 @@ async def test_generate_resume_returns_ready_when_target_not_met(monkeypatch) ->
     llm = SimpleNamespace(ainvoke=AsyncMock(side_effect=responses))
     monkeypatch.setattr(
         resume_tool,
-        "get_document_summary_llm",
+        "get_agent_llm",
         AsyncMock(return_value=llm),
     )
     monkeypatch.setattr(resume_tool, "_compile_typst", lambda _source: b"pdf")
@@ -198,7 +198,7 @@ async def test_generate_resume_fails_when_hard_limit_exceeded(monkeypatch) -> No
     llm = SimpleNamespace(ainvoke=AsyncMock(side_effect=responses))
     monkeypatch.setattr(
         resume_tool,
-        "get_document_summary_llm",
+        "get_agent_llm",
         AsyncMock(return_value=llm),
     )
     monkeypatch.setattr(resume_tool, "_compile_typst", lambda _source: b"pdf")

surfsense_backend/tests/unit/automations/actions/builtin/agent_task/test_dependencies.py

@@ -14,8 +14,8 @@ from typing import Any
 
 import pytest
 
-import app.automations.actions.agent_task.dependencies as deps_mod
+import app.automations.actions.builtin.agent_task.dependencies as deps_mod
-from app.automations.actions.agent_task.dependencies import (
+from app.automations.actions.builtin.agent_task.dependencies import (
     DependencyError,
     build_dependencies,
 )

surfsense_backend/tests/unit/connector_indexers/test_confluence_parallel.py

@@ -71,7 +71,6 @@ async def test_build_connector_doc_produces_correct_fields():
         connector_id=_CONNECTOR_ID,
         search_space_id=_SEARCH_SPACE_ID,
         user_id=_USER_ID,
-        enable_summary=True,
     )
 
     assert doc.title == "Engineering Handbook"
@@ -81,7 +80,6 @@ async def test_build_connector_doc_produces_correct_fields():
     assert doc.search_space_id == _SEARCH_SPACE_ID
     assert doc.connector_id == _CONNECTOR_ID
     assert doc.created_by_id == _USER_ID
-    assert doc.should_summarize is True
     assert doc.metadata["page_id"] == "abc-123"
     assert doc.metadata["page_title"] == "Engineering Handbook"
     assert doc.metadata["space_id"] == "ENG"
@@ -89,21 +87,6 @@ async def test_build_connector_doc_produces_correct_fields():
     assert doc.metadata["connector_id"] == _CONNECTOR_ID
     assert doc.metadata["document_type"] == "Confluence Page"
     assert doc.metadata["connector_type"] == "Confluence"
-    assert doc.fallback_summary is not None
-    assert "Engineering Handbook" in doc.fallback_summary
-    assert markdown in doc.fallback_summary
-
-
-async def test_build_connector_doc_summary_disabled():
-    doc = _build_connector_doc(
-        _make_page(),
-        _to_markdown(_make_page()),
-        connector_id=_CONNECTOR_ID,
-        search_space_id=_SEARCH_SPACE_ID,
-        user_id=_USER_ID,
-        enable_summary=False,
-    )
-    assert doc.should_summarize is False
 
 
 # ---------------------------------------------------------------------------
@@ -111,10 +94,9 @@ async def test_build_connector_doc_summary_disabled():
 # ---------------------------------------------------------------------------
 
 
-def _mock_connector(enable_summary: bool = True):
+def _mock_connector():
     c = MagicMock()
     c.config = {"access_token": "tok"}
-    c.enable_summary = enable_summary
     c.last_indexed_at = None
     return c
 

surfsense_backend/tests/unit/connector_indexers/test_dropbox_parallel.py

@@ -71,7 +71,6 @@ async def test_single_file_returns_one_connector_document(
         connector_id=_CONNECTOR_ID,
         search_space_id=_SEARCH_SPACE_ID,
         user_id=_USER_ID,
-        enable_summary=True,
     )
 
     assert len(docs) == 1
@@ -97,7 +96,6 @@ async def test_multiple_files_all_produce_documents(
         connector_id=_CONNECTOR_ID,
         search_space_id=_SEARCH_SPACE_ID,
         user_id=_USER_ID,
-        enable_summary=True,
     )
 
     assert len(docs) == 3
@@ -125,7 +123,6 @@ async def test_one_download_exception_does_not_block_others(
         connector_id=_CONNECTOR_ID,
         search_space_id=_SEARCH_SPACE_ID,
         user_id=_USER_ID,
-        enable_summary=True,
     )
 
     assert len(docs) == 2
@@ -152,7 +149,6 @@ async def test_etl_error_counts_as_download_failure(
         connector_id=_CONNECTOR_ID,
         search_space_id=_SEARCH_SPACE_ID,
         user_id=_USER_ID,
-        enable_summary=True,
     )
 
     assert len(docs) == 1
@@ -191,7 +187,6 @@ async def test_concurrency_bounded_by_semaphore(
         connector_id=_CONNECTOR_ID,
         search_space_id=_SEARCH_SPACE_ID,
         user_id=_USER_ID,
-        enable_summary=True,
         max_concurrency=2,
     )
 
@@ -231,7 +226,6 @@ async def test_heartbeat_fires_during_parallel_downloads(
         connector_id=_CONNECTOR_ID,
         search_space_id=_SEARCH_SPACE_ID,
         user_id=_USER_ID,
-        enable_summary=True,
         on_heartbeat=_on_heartbeat,
     )
 
@@ -324,7 +318,6 @@ async def _run_full_scan(mocks, monkeypatch, page_files, *, max_files=500):
         mocks["task_logger"],
         mocks["log_entry"],
         max_files,
-        enable_summary=True,
     )
 
 
@@ -434,7 +427,6 @@ async def _run_selected(mocks, file_tuples):
         connector_id=_CONNECTOR_ID,
         search_space_id=_SEARCH_SPACE_ID,
         user_id=_USER_ID,
-        enable_summary=True,
     )
 
 
@@ -569,7 +561,6 @@ async def test_delta_sync_deletions_call_remove_document(monkeypatch):
         mock_task_logger,
         MagicMock(),
         max_files=500,
-        enable_summary=True,
     )
 
     assert sorted(remove_calls) == ["id:del1", "id:del2"]
@@ -608,7 +599,6 @@ async def test_delta_sync_upserts_filtered_and_downloaded(monkeypatch):
         mock_task_logger,
         MagicMock(),
         max_files=500,
-        enable_summary=True,
     )
 
     assert indexed == 2
@@ -670,7 +660,6 @@ async def test_delta_sync_mix_deletions_and_upserts(monkeypatch):
         mock_task_logger,
         MagicMock(),
         max_files=500,
-        enable_summary=True,
     )
 
     assert sorted(remove_calls) == ["id:del1", "id:del2"]
@@ -704,7 +693,6 @@ async def test_delta_sync_returns_new_cursor(monkeypatch):
         mock_task_logger,
         MagicMock(),
         max_files=500,
-        enable_summary=True,
     )
 
     assert cursor == "brand-new-cursor-xyz"
@@ -725,7 +713,7 @@ def orchestrator_mocks(monkeypatch):
     mock_connector = MagicMock()
     mock_connector.config = {"_token_encrypted": False}
     mock_connector.last_indexed_at = None
-    mock_connector.enable_summary = True
+    mock_connector.enable_vision_llm = True
 
     monkeypatch.setattr(
         _mod,

surfsense_backend/tests/unit/connector_indexers/test_google_drive_parallel.py

@@ -66,7 +66,6 @@ async def test_single_file_returns_one_connector_document(
         connector_id=_CONNECTOR_ID,
         search_space_id=_SEARCH_SPACE_ID,
         user_id=_USER_ID,
-        enable_summary=True,
     )
 
     assert len(docs) == 1
@@ -91,7 +90,6 @@ async def test_multiple_files_all_produce_documents(
         connector_id=_CONNECTOR_ID,
         search_space_id=_SEARCH_SPACE_ID,
         user_id=_USER_ID,
-        enable_summary=True,
     )
 
     assert len(docs) == 3
@@ -119,7 +117,6 @@ async def test_one_download_exception_does_not_block_others(
         connector_id=_CONNECTOR_ID,
         search_space_id=_SEARCH_SPACE_ID,
         user_id=_USER_ID,
-        enable_summary=True,
     )
 
     assert len(docs) == 2
@@ -146,7 +143,6 @@ async def test_etl_error_counts_as_download_failure(
         connector_id=_CONNECTOR_ID,
         search_space_id=_SEARCH_SPACE_ID,
         user_id=_USER_ID,
-        enable_summary=True,
     )
 
     assert len(docs) == 1
@@ -186,7 +182,6 @@ async def test_concurrency_bounded_by_semaphore(
         connector_id=_CONNECTOR_ID,
         search_space_id=_SEARCH_SPACE_ID,
         user_id=_USER_ID,
-        enable_summary=True,
         max_concurrency=2,
     )
 
@@ -226,7 +221,6 @@ async def test_heartbeat_fires_during_parallel_downloads(
         connector_id=_CONNECTOR_ID,
         search_space_id=_SEARCH_SPACE_ID,
         user_id=_USER_ID,
-        enable_summary=True,
         on_heartbeat=_on_heartbeat,
     )
 
@@ -300,12 +294,6 @@ def full_scan_mocks(mock_drive_client, monkeypatch):
         MagicMock(return_value=pipeline_mock),
     )
 
-    monkeypatch.setattr(
-        _mod,
-        "get_user_long_context_llm",
-        AsyncMock(return_value=MagicMock()),
-    )
-
     return {
         "drive_client": mock_drive_client,
         "session": mock_session,
@@ -333,7 +321,6 @@ async def _run_full_scan(mocks, *, max_files=500, include_subfolders=False):
         mocks["log_entry"],
         max_files,
         include_subfolders=include_subfolders,
-        enable_summary=True,
     )
 
 
@@ -487,12 +474,6 @@ async def test_delta_sync_removals_serial_rest_parallel(monkeypatch):
         "IndexingPipelineService",
         MagicMock(return_value=pipeline_mock),
     )
-    monkeypatch.setattr(
-        _mod,
-        "get_user_long_context_llm",
-        AsyncMock(return_value=MagicMock()),
-    )
-
     mock_session, _ = _make_page_limit_session()
     mock_task_logger = MagicMock()
     mock_task_logger.log_task_progress = AsyncMock()
@@ -509,7 +490,6 @@ async def test_delta_sync_removals_serial_rest_parallel(monkeypatch):
         mock_task_logger,
         MagicMock(),
         max_files=500,
-        enable_summary=True,
     )
 
     assert sorted(remove_calls) == ["del1", "del2", "trash1"]
@@ -577,7 +557,6 @@ async def _run_selected(mocks, file_ids):
         connector_id=_CONNECTOR_ID,
         search_space_id=_SEARCH_SPACE_ID,
         user_id=_USER_ID,
-        enable_summary=True,
     )
 
 

surfsense_backend/tests/unit/connector_indexers/test_linear_parallel.py

@@ -70,7 +70,6 @@ async def test_build_connector_doc_produces_correct_fields():
         connector_id=_CONNECTOR_ID,
         search_space_id=_SEARCH_SPACE_ID,
         user_id=_USER_ID,
-        enable_summary=True,
     )
 
     assert doc.title == "ENG-42: Fix login bug"
@@ -80,7 +79,6 @@ async def test_build_connector_doc_produces_correct_fields():
     assert doc.search_space_id == _SEARCH_SPACE_ID
     assert doc.connector_id == _CONNECTOR_ID
     assert doc.created_by_id == _USER_ID
-    assert doc.should_summarize is True
     assert doc.metadata["issue_id"] == "abc-123"
     assert doc.metadata["issue_identifier"] == "ENG-42"
     assert doc.metadata["issue_title"] == "Fix login bug"
@@ -90,24 +88,6 @@ async def test_build_connector_doc_produces_correct_fields():
     assert doc.metadata["connector_id"] == _CONNECTOR_ID
     assert doc.metadata["document_type"] == "Linear Issue"
     assert doc.metadata["connector_type"] == "Linear"
-    assert doc.fallback_summary is not None
-    assert "ENG-42" in doc.fallback_summary
-    assert markdown in doc.fallback_summary
-
-
-async def test_build_connector_doc_summary_disabled():
-    """When enable_summary is False, should_summarize is False."""
-    doc = _build_connector_doc(
-        _make_issue(),
-        _make_formatted_issue(),
-        "# content",
-        connector_id=_CONNECTOR_ID,
-        search_space_id=_SEARCH_SPACE_ID,
-        user_id=_USER_ID,
-        enable_summary=False,
-    )
-
-    assert doc.should_summarize is False
 
 
 # ---------------------------------------------------------------------------
@@ -115,10 +95,9 @@ async def test_build_connector_doc_summary_disabled():
 # ---------------------------------------------------------------------------
 
 
-def _mock_connector(enable_summary: bool = True):
+def _mock_connector():
     c = MagicMock()
     c.config = {"access_token": "tok"}
-    c.enable_summary = enable_summary
     c.last_indexed_at = None
     return c
 

surfsense_backend/tests/unit/connector_indexers/test_notion_parallel.py

@@ -41,7 +41,6 @@ async def test_build_connector_doc_produces_correct_fields():
         connector_id=_CONNECTOR_ID,
         search_space_id=_SEARCH_SPACE_ID,
         user_id=_USER_ID,
-        enable_summary=True,
     )
 
     assert doc.title == "My Notion Page"
@@ -51,29 +50,11 @@ async def test_build_connector_doc_produces_correct_fields():
     assert doc.search_space_id == _SEARCH_SPACE_ID
     assert doc.connector_id == _CONNECTOR_ID
     assert doc.created_by_id == _USER_ID
-    assert doc.should_summarize is True
     assert doc.metadata["page_title"] == "My Notion Page"
     assert doc.metadata["page_id"] == "abc-123"
     assert doc.metadata["connector_id"] == _CONNECTOR_ID
     assert doc.metadata["document_type"] == "Notion Page"
     assert doc.metadata["connector_type"] == "Notion"
-    assert doc.fallback_summary is not None
-    assert "My Notion Page" in doc.fallback_summary
-    assert markdown in doc.fallback_summary
-
-
-async def test_build_connector_doc_summary_disabled():
-    """When enable_summary is False, should_summarize is False."""
-    doc = _build_connector_doc(
-        _make_page(),
-        "# content",
-        connector_id=_CONNECTOR_ID,
-        search_space_id=_SEARCH_SPACE_ID,
-        user_id=_USER_ID,
-        enable_summary=False,
-    )
-
-    assert doc.should_summarize is False
 
 
 # ---------------------------------------------------------------------------
@@ -81,10 +62,9 @@ async def test_build_connector_doc_summary_disabled():
 # ---------------------------------------------------------------------------
 
 
-def _mock_connector(enable_summary: bool = True):
+def _mock_connector():
     c = MagicMock()
     c.config = {"access_token": "tok"}
-    c.enable_summary = enable_summary
     c.last_indexed_at = None
     return c
 

surfsense_backend/tests/unit/connector_indexers/test_onedrive_parallel.py

@@ -65,7 +65,6 @@ async def test_single_file_returns_one_connector_document(
         connector_id=_CONNECTOR_ID,
         search_space_id=_SEARCH_SPACE_ID,
         user_id=_USER_ID,
-        enable_summary=True,
     )
 
     assert len(docs) == 1
@@ -91,7 +90,6 @@ async def test_multiple_files_all_produce_documents(
         connector_id=_CONNECTOR_ID,
         search_space_id=_SEARCH_SPACE_ID,
         user_id=_USER_ID,
-        enable_summary=True,
     )
 
     assert len(docs) == 3
@@ -119,7 +117,6 @@ async def test_one_download_exception_does_not_block_others(
         connector_id=_CONNECTOR_ID,
         search_space_id=_SEARCH_SPACE_ID,
         user_id=_USER_ID,
-        enable_summary=True,
     )
 
     assert len(docs) == 2
@@ -146,7 +143,6 @@ async def test_etl_error_counts_as_download_failure(
         connector_id=_CONNECTOR_ID,
         search_space_id=_SEARCH_SPACE_ID,
         user_id=_USER_ID,
-        enable_summary=True,
     )
 
     assert len(docs) == 1
@@ -185,7 +181,6 @@ async def test_concurrency_bounded_by_semaphore(
         connector_id=_CONNECTOR_ID,
         search_space_id=_SEARCH_SPACE_ID,
         user_id=_USER_ID,
-        enable_summary=True,
         max_concurrency=2,
     )
 
@@ -225,7 +220,6 @@ async def test_heartbeat_fires_during_parallel_downloads(
         connector_id=_CONNECTOR_ID,
         search_space_id=_SEARCH_SPACE_ID,
         user_id=_USER_ID,
-        enable_summary=True,
         on_heartbeat=_on_heartbeat,
     )
 

surfsense_backend/tests/unit/connector_indexers/test_page_limits.py

@@ -180,7 +180,6 @@ async def _run_gdrive_selected(mocks, file_ids):
         connector_id=_CONNECTOR_ID,
         search_space_id=_SEARCH_SPACE_ID,
         user_id=_USER_ID,
-        enable_summary=True,
     )
 
 
@@ -336,10 +335,6 @@ def gdrive_full_scan_mocks(monkeypatch):
     monkeypatch.setattr(
         _mod, "IndexingPipelineService", MagicMock(return_value=pipeline_mock)
     )
-    monkeypatch.setattr(
-        _mod, "get_user_long_context_llm", AsyncMock(return_value=MagicMock())
-    )
-
     return {
         "mod": _mod,
         "session": session,
@@ -366,7 +361,6 @@ async def _run_gdrive_full_scan(mocks, max_files=500):
         MagicMock(),
         max_files,
         include_subfolders=False,
-        enable_summary=True,
     )
 
 
@@ -454,10 +448,6 @@ async def test_gdrive_delta_sync_skips_over_quota(monkeypatch):
     monkeypatch.setattr(
         _mod, "IndexingPipelineService", MagicMock(return_value=pipeline_mock)
     )
-    monkeypatch.setattr(
-        _mod, "get_user_long_context_llm", AsyncMock(return_value=MagicMock())
-    )
-
     mock_task_logger = MagicMock()
     mock_task_logger.log_task_progress = AsyncMock()
 
@@ -473,7 +463,6 @@ async def test_gdrive_delta_sync_skips_over_quota(monkeypatch):
         mock_task_logger,
         MagicMock(),
         max_files=500,
-        enable_summary=True,
     )
 
     call_files = download_mock.call_args[0][1]
@@ -539,7 +528,6 @@ async def _run_onedrive_selected(mocks, file_ids):
         connector_id=_CONNECTOR_ID,
         search_space_id=_SEARCH_SPACE_ID,
         user_id=_USER_ID,
-        enable_summary=True,
     )
 
 
@@ -641,7 +629,6 @@ async def _run_dropbox_selected(mocks, file_paths):
         connector_id=_CONNECTOR_ID,
         search_space_id=_SEARCH_SPACE_ID,
         user_id=_USER_ID,
-        enable_summary=True,
     )
 
 

surfsense_backend/tests/unit/gateway/test_webhook_routes.py

@@ -69,6 +69,13 @@ def _signed_slack_request(payload: dict, *, secret: str = "signing-secret") -> R
     )
 
 
+def _enable_slack_gateway(monkeypatch):
+    monkeypatch.setattr(routes.config, "GATEWAY_SLACK_ENABLED", True)
+    monkeypatch.setattr(routes.config, "GATEWAY_SLACK_CLIENT_ID", "client-id")
+    monkeypatch.setattr(routes.config, "GATEWAY_SLACK_CLIENT_SECRET", "client-secret")
+    monkeypatch.setattr(routes.config, "GATEWAY_SLACK_SIGNING_SECRET", "signing-secret")
+
+
 async def _call_webhook(*, request: RequestStub, account_id: int, session):
     return await routes.telegram_webhook(
         request=request,
@@ -207,7 +214,7 @@ def test_verify_slack_signature_accepts_valid_signature():
 
 @pytest.mark.asyncio
 async def test_slack_webhook_url_verification(monkeypatch, mocker):
-    monkeypatch.setattr(routes.config, "GATEWAY_SLACK_SIGNING_SECRET", "signing-secret")
+    _enable_slack_gateway(monkeypatch)
     request = _signed_slack_request({"type": "url_verification", "challenge": "abc123"})
 
     response = await routes.slack_webhook(request=request, session=mocker.AsyncMock())
@@ -218,7 +225,7 @@ async def test_slack_webhook_url_verification(monkeypatch, mocker):
 
 @pytest.mark.asyncio
 async def test_slack_webhook_persists_event(monkeypatch, mocker):
-    monkeypatch.setattr(routes.config, "GATEWAY_SLACK_SIGNING_SECRET", "signing-secret")
+    _enable_slack_gateway(monkeypatch)
     session = mocker.AsyncMock()
     monkeypatch.setattr(routes, "get_slack_account_by_team", mocker.AsyncMock(return_value=_slack_account()))
     persist = mocker.AsyncMock(return_value=100)
@@ -248,7 +255,7 @@ async def test_slack_webhook_persists_event(monkeypatch, mocker):
 
 @pytest.mark.asyncio
 async def test_slack_webhook_ignores_self_event(monkeypatch, mocker):
-    monkeypatch.setattr(routes.config, "GATEWAY_SLACK_SIGNING_SECRET", "signing-secret")
+    _enable_slack_gateway(monkeypatch)
     session = mocker.AsyncMock()
     monkeypatch.setattr(routes, "get_slack_account_by_team", mocker.AsyncMock(return_value=_slack_account()))
     persist = mocker.AsyncMock(return_value=100)
@@ -275,7 +282,7 @@ async def test_slack_webhook_ignores_self_event(monkeypatch, mocker):
 
 
 @pytest.mark.asyncio
-async def test_discord_gateway_install_returns_oauth_url(monkeypatch):
+async def test_discord_gateway_install_returns_oauth_url(monkeypatch, mocker):
     monkeypatch.setattr(routes.config, "DISCORD_CLIENT_ID", "discord-client")
     monkeypatch.setattr(
         routes.config,
@@ -283,10 +290,12 @@ async def test_discord_gateway_install_returns_oauth_url(monkeypatch):
         "http://localhost:8000/api/v1/gateway/discord/callback",
     )
     monkeypatch.setattr(routes.config, "SECRET_KEY", "test-secret")
+    monkeypatch.setattr(routes, "check_search_space_access", mocker.AsyncMock())
 
     response = await routes.install_discord_gateway(
         search_space_id=123,
         user=SimpleNamespace(id="00000000-0000-0000-0000-000000000001"),
+        session=mocker.AsyncMock(),
     )
 
     assert response["auth_url"].startswith("https://discord.com/api/oauth2/authorize?")

surfsense_backend/tests/unit/indexing_pipeline/test_connector_document.py

@@ -18,7 +18,6 @@ def test_valid_document_created_with_required_fields():
         connector_id=42,
         created_by_id="00000000-0000-0000-0000-000000000001",
     )
-    assert doc.should_summarize is True
     assert doc.should_use_code_chunker is False
     assert doc.metadata == {}
     assert doc.connector_id == 42

surfsense_backend/tests/unit/indexing_pipeline/test_document_summarizer.py

@@ -1,41 +0,0 @@
-from unittest.mock import MagicMock
-
-import pytest
-
-from app.indexing_pipeline.document_summarizer import summarize_document
-
-pytestmark = pytest.mark.unit
-
-
-@pytest.mark.usefixtures("patched_summarizer_chain")
-async def test_without_metadata_returns_raw_summary():
-    """Summarizer returns the LLM output directly when no metadata is provided."""
-    result = await summarize_document("# Content", llm=MagicMock(model="gpt-4"))
-
-    assert result == "The summary."
-
-
-@pytest.mark.usefixtures("patched_summarizer_chain")
-async def test_with_metadata_includes_metadata_values_in_output():
-    """Non-empty metadata values are prepended to the summary output."""
-    result = await summarize_document(
-        "# Content",
-        llm=MagicMock(model="gpt-4"),
-        metadata={"author": "Alice", "source": "Notion"},
-    )
-
-    assert "Alice" in result
-    assert "Notion" in result
-
-
-@pytest.mark.usefixtures("patched_summarizer_chain")
-async def test_with_metadata_omits_empty_fields_from_output():
-    """Empty metadata fields are omitted from the summary output."""
-    result = await summarize_document(
-        "# Content",
-        llm=MagicMock(model="gpt-4"),
-        metadata={"author": "Alice", "description": ""},
-    )
-
-    assert "Alice" in result
-    assert "description" not in result.lower()

surfsense_backend/tests/unit/indexing_pipeline/test_index_batch.py

@@ -37,12 +37,10 @@ async def test_calls_prepare_then_index_per_document(pipeline, make_connector_do
     orm2 = MagicMock(spec=Document)
     orm2.unique_identifier_hash = compute_unique_identifier_hash(doc2)
 
-    mock_llm = MagicMock()
-
     pipeline.prepare_for_indexing = AsyncMock(return_value=[orm1, orm2])
-    pipeline.index = AsyncMock(side_effect=lambda doc, cdoc, llm: doc)
+    pipeline.index = AsyncMock(side_effect=lambda doc, cdoc: doc)
 
-    results = await pipeline.index_batch([doc1, doc2], mock_llm)
+    results = await pipeline.index_batch([doc1, doc2])
 
     pipeline.prepare_for_indexing.assert_awaited_once_with([doc1, doc2])
     assert pipeline.index.await_count == 2
@@ -53,7 +51,7 @@ async def test_empty_input_returns_empty(pipeline):
     """Empty connector_docs list returns empty result."""
     pipeline.prepare_for_indexing = AsyncMock(return_value=[])
 
-    results = await pipeline.index_batch([], MagicMock())
+    results = await pipeline.index_batch([])
 
     assert results == []
 
@@ -74,7 +72,7 @@ async def test_skips_document_without_matching_connector_doc(
     pipeline.prepare_for_indexing = AsyncMock(return_value=[orphan_orm])
     pipeline.index = AsyncMock()
 
-    results = await pipeline.index_batch([doc1], MagicMock())
+    results = await pipeline.index_batch([doc1])
 
     pipeline.index.assert_not_awaited()
     assert results == []

surfsense_backend/tests/unit/indexing_pipeline/test_index_batch_parallel.py

@@ -51,11 +51,6 @@ async def test_index_calls_embed_and_chunk_via_to_thread(
         return await original_to_thread(func, *args, **kwargs)
 
     monkeypatch.setattr(asyncio, "to_thread", tracking_to_thread)
-
-    monkeypatch.setattr(
-        "app.indexing_pipeline.indexing_pipeline_service.summarize_document",
-        AsyncMock(return_value="Summary."),
-    )
     mock_chunk_hybrid = MagicMock(return_value=["chunk1"])
     mock_chunk_hybrid.__name__ = "chunk_text_hybrid"
     monkeypatch.setattr(
@@ -85,7 +80,7 @@ async def test_index_calls_embed_and_chunk_via_to_thread(
     document.id = 1
     document.status = DocumentStatus.pending()
 
-    await pipeline.index(document, connector_doc, llm=MagicMock())
+    await pipeline.index(document, connector_doc)
 
     # Either chunker entry point satisfies the "chunking runs off the event
     # loop" contract this test guards. Routing between the two is verified
@@ -104,10 +99,6 @@ async def test_non_code_documents_use_hybrid_chunker(
     mid-row. Only documents flagged with ``should_use_code_chunker=True``
     should take the ``chunk_text`` path.
     """
-    monkeypatch.setattr(
-        "app.indexing_pipeline.indexing_pipeline_service.summarize_document",
-        AsyncMock(return_value="Summary."),
-    )
     mock_chunk_hybrid = MagicMock(return_value=["chunk1"])
     mock_chunk_hybrid.__name__ = "chunk_text_hybrid"
     monkeypatch.setattr(
@@ -139,7 +130,7 @@ async def test_non_code_documents_use_hybrid_chunker(
     document.id = 1
     document.status = DocumentStatus.pending()
 
-    await pipeline.index(document, connector_doc, llm=MagicMock())
+    await pipeline.index(document, connector_doc)
 
     mock_chunk_hybrid.assert_called_once()
     mock_chunk_code.assert_not_called()
@@ -192,19 +183,14 @@ async def test_batch_parallel_indexes_all_documents(
 
     index_calls = []
 
-    async def fake_index(self, document, connector_doc, llm):
+    async def fake_index(self, document, connector_doc):
         index_calls.append(document.id)
         document.status = DocumentStatus.ready()
         return document
 
     monkeypatch.setattr(IndexingPipelineService, "index", fake_index)
 
-    async def mock_get_llm(session):
+    _, indexed, failed = await pipeline.index_batch_parallel(docs, max_concurrency=2)
-        return MagicMock()
-
-    _, indexed, failed = await pipeline.index_batch_parallel(
-        docs, mock_get_llm, max_concurrency=2
-    )
 
     assert indexed == 3
     assert failed == 0
@@ -233,20 +219,15 @@ async def test_batch_parallel_one_failure_does_not_affect_others(
         _mock_session_factory(orm_by_id),
     )
 
-    async def failing_index(self, document, connector_doc, llm):
+    async def failing_index(self, document, connector_doc):
         if document.id == 2:
-            raise RuntimeError("LLM exploded")
+            raise RuntimeError("Indexing exploded")
         document.status = DocumentStatus.ready()
         return document
 
     monkeypatch.setattr(IndexingPipelineService, "index", failing_index)
 
-    async def mock_get_llm(session):
+    _, indexed, failed = await pipeline.index_batch_parallel(docs, max_concurrency=4)
-        return MagicMock()
-
-    _, indexed, failed = await pipeline.index_batch_parallel(
-        docs, mock_get_llm, max_concurrency=4
-    )
 
     assert indexed == 2
     assert failed == 1

surfsense_backend/tests/unit/tasks/chat/streaming/test_parallel_refactor_parity.py

@@ -246,6 +246,8 @@ def test_new_chat_runtime_context_prefers_accepted_folder_ids() -> None:
         mentioned_document_ids=[1, 2],
         accepted_folder_ids=[10],
         mentioned_folder_ids=[20, 30],
+        mentioned_connector_ids=None,
+        mentioned_connectors=None,
         request_id="req",
         turn_id="t1",
     )
@@ -263,6 +265,8 @@ def test_new_chat_runtime_context_falls_back_to_mentioned_folder_ids() -> None:
         mentioned_document_ids=None,
         accepted_folder_ids=[],
         mentioned_folder_ids=[20, 30],
+        mentioned_connector_ids=None,
+        mentioned_connectors=None,
         request_id=None,
         turn_id="t2",
     )

surfsense_evals/README.md

@@ -137,15 +137,14 @@ Notes:
 - `--skip-unanswerable` (run) — drop unanswerable questions
 - `--docs <a.pdf>,<b.pdf>` (run) — scope to specific docs
 
-## Ingestion knobs (vision LLM, processing mode, summarize)
+## Ingestion knobs (vision LLM, processing mode)
 
-The harness exposes `POST /api/v1/documents/fileupload`'s three knobs on every `ingest` subcommand:
+The harness exposes `POST /api/v1/documents/fileupload`'s ingest knobs on every `ingest` subcommand:
 
 | Flag pair                                  | Effect                                                                                  |
 |--------------------------------------------|-----------------------------------------------------------------------------------------|
 | `--use-vision-llm` / `--no-vision-llm`     | Walk every embedded image in the PDF and inline image-derived text at the image's position (see below). |
 | `--processing-mode {basic,premium}`        | `premium` carries a 10× page multiplier and routes to a stronger ETL (e.g. LlamaCloud). |
-| `--should-summarize` / `--no-summarize`    | Generate a per-document summary at ingest.                                              |
 
 The "Default ingest" column in the benchmarks table is what runs if you don't pass any flag. Whatever was actually used is recorded as a `__settings__` header in the doc map (`data/<suite>/maps/<benchmark>_*_map.jsonl`) and as `extra.ingest_settings` in `run_artifact.json`, then surfaced in the report — no need to hunt through CLI history.
 

surfsense_evals/src/surfsense_evals/core/clients/documents.py

@@ -110,7 +110,6 @@ class DocumentsClient:
         files: Iterable[Path],
         *,
         search_space_id: int,
-        should_summarize: bool = False,
         use_vision_llm: bool = False,
         processing_mode: str = "basic",
     ) -> FileUploadResult:
@@ -149,7 +148,6 @@ class DocumentsClient:
                 f"{self._base}/api/v1/documents/fileupload",
                 data={
                     "search_space_id": str(search_space_id),
-                    "should_summarize": "true" if should_summarize else "false",
                     "use_vision_llm": "true" if use_vision_llm else "false",
                     "processing_mode": processing_mode,
                 },

surfsense_evals/src/surfsense_evals/core/clients/search_space.py

@@ -83,7 +83,6 @@ class LlmPreferences:
     """
 
     agent_llm_id: int | None
-    document_summary_llm_id: int | None
     image_generation_config_id: int | None
     vision_llm_config_id: int | None
     agent_llm: dict[str, Any] | None
@@ -93,7 +92,6 @@ class LlmPreferences:
     def from_payload(cls, payload: dict[str, Any]) -> LlmPreferences:
         return cls(
             agent_llm_id=payload.get("agent_llm_id"),
-            document_summary_llm_id=payload.get("document_summary_llm_id"),
             image_generation_config_id=payload.get("image_generation_config_id"),
             vision_llm_config_id=payload.get("vision_llm_config_id"),
             agent_llm=payload.get("agent_llm"),
@@ -154,7 +152,6 @@ class SearchSpaceClient:
         search_space_id: int,
         *,
         agent_llm_id: int | None = None,
-        document_summary_llm_id: int | None = None,
         image_generation_config_id: int | None = None,
         vision_llm_config_id: int | None = None,
     ) -> LlmPreferences:
@@ -167,8 +164,6 @@ class SearchSpaceClient:
         body: dict[str, Any] = {}
         if agent_llm_id is not None:
             body["agent_llm_id"] = agent_llm_id
-        if document_summary_llm_id is not None:
-            body["document_summary_llm_id"] = document_summary_llm_id
         if image_generation_config_id is not None:
             body["image_generation_config_id"] = image_generation_config_id
         if vision_llm_config_id is not None:

surfsense_evals/src/surfsense_evals/core/ingest_settings.py

@@ -8,15 +8,13 @@ exactly three knobs (verified at
 * ``processing_mode``     — ``"basic"`` (default) | ``"premium"``
 * ``use_vision_llm``      — ``bool`` (run vision LLM during ingest to
                             extract image content / captions / tables)
-* ``should_summarize``    — ``bool`` (generate document summary)
 
 This module gives every benchmark a uniform way to:
 
 1. Receive sensible per-benchmark defaults (text-only benchmarks
    default vision off; image-bearing benchmarks default vision on).
 2. Accept CLI overrides (``--use-vision-llm`` / ``--no-vision-llm``,
-   ``--processing-mode {basic,premium}``,
+   ``--processing-mode {basic,premium}``).
-   ``--should-summarize`` / ``--no-summarize``).
 3. Persist the *actual* settings used into the doc-map manifest and
    the run artifact so reports can show "vision=ON, mode=premium →
    65% accuracy" head-to-head with "vision=OFF, mode=basic → 52%".
@@ -71,13 +69,11 @@ class IngestSettings:
 
     use_vision_llm: bool = False
     processing_mode: str = "basic"
-    should_summarize: bool = False
 
     def to_dict(self) -> dict[str, Any]:
         return {
             "use_vision_llm": self.use_vision_llm,
             "processing_mode": self.processing_mode,
-            "should_summarize": self.should_summarize,
         }
 
     @classmethod
@@ -87,14 +83,13 @@ class IngestSettings:
         ``opts`` is the kwargs dict built by ``core.cli`` from the
         argparse namespace (see ``_cmd_ingest`` / ``_cmd_run``). Keys
         we look for: ``use_vision_llm`` (bool or None), ``processing_mode``
-        (str or None), ``should_summarize`` (bool or None). Anything
+        (str or None). Anything
         else is ignored so benchmarks can pass through their own opts.
         """
 
         return cls(
             use_vision_llm=_coerce_bool(opts.get("use_vision_llm"), defaults.use_vision_llm),
             processing_mode=_coerce_mode(opts.get("processing_mode"), defaults.processing_mode),
-            should_summarize=_coerce_bool(opts.get("should_summarize"), defaults.should_summarize),
         )
 
     def render_label(self) -> str:
@@ -102,8 +97,7 @@ class IngestSettings:
 
         return (
             f"vision={'on' if self.use_vision_llm else 'off'}, "
-            f"mode={self.processing_mode}, "
+            f"mode={self.processing_mode}"
-            f"summarize={'on' if self.should_summarize else 'off'}"
         )
 
 
@@ -179,14 +173,14 @@ def add_ingest_settings_args(
     *,
     defaults: IngestSettings,
 ) -> None:
-    """Attach the three ingest-settings flag pairs to ``parser``.
+    """Attach ingest-settings flags to ``parser``.
 
-    Each bool exposes a mutually exclusive ``--foo`` / ``--no-foo``
+    The vision bool exposes a mutually exclusive ``--foo`` / ``--no-foo``
-    pair so an operator can flip either direction without restating
+    pair so an operator can flip either direction without restating every
-    every flag. Default is ``None`` so that "operator didn't pass the
+    flag. Default is ``None`` so that "operator didn't pass the flag" is
-    flag" is distinguishable from "operator explicitly passed false"
+    distinguishable from "operator explicitly passed false" —
-    — ``IngestSettings.merge`` then folds in the benchmark default
+    ``IngestSettings.merge`` then folds in the benchmark default only when
-    only when the operator was silent.
+    the operator was silent.
     """
 
     settings_group = parser.add_argument_group(
@@ -217,18 +211,6 @@ def add_ingest_settings_args(
             f"Default for this benchmark: {defaults.processing_mode!r}."
         ),
     )
-    _add_bool_pair(
-        settings_group,
-        dest="should_summarize",
-        on_flag="--should-summarize",
-        off_flag="--no-summarize",
-        on_help=(
-            "Have SurfSense generate a document summary at ingest "
-            f"(default for this benchmark: "
-            f"{'on' if defaults.should_summarize else 'off'})."
-        ),
-        off_help="Skip per-document summary generation.",
-    )
 
 
 # ---------------------------------------------------------------------------
@@ -292,10 +274,9 @@ def format_ingest_settings_md(settings: Any) -> str:
         return "- SurfSense ingest settings: (not recorded — re-ingest to capture)"
     vision = "on" if settings.get("use_vision_llm") else "off"
     mode = settings.get("processing_mode") or "basic"
-    summarize = "on" if settings.get("should_summarize") else "off"
     return (
         f"- SurfSense ingest settings: vision_llm=`{vision}`, "
-        f"processing_mode=`{mode}`, summarize=`{summarize}`"
+        f"processing_mode=`{mode}`"
     )
 
 

surfsense_evals/src/surfsense_evals/suites/medical/cure/ingest.py

@@ -160,8 +160,7 @@ async def run_ingest(
             upload_result = await docs_client.upload(
                 files=[b.path for b in batches],
                 search_space_id=ctx.search_space_id,
-                should_summarize=settings.should_summarize,
+                    use_vision_llm=settings.use_vision_llm,
-                use_vision_llm=settings.use_vision_llm,
                 processing_mode=settings.processing_mode,
             )
             new_doc_ids = list(upload_result.document_ids)

surfsense_evals/src/surfsense_evals/suites/medical/cure/runner.py

@@ -63,7 +63,6 @@ _DESCRIPTION = "CUREv1 retrieval (single-arm SurfSense): Recall@k / MRR / nDCG@1
 _DEFAULT_INGEST_SETTINGS = IngestSettings(
     use_vision_llm=False,
     processing_mode="basic",
-    should_summarize=False,
 )
 
 

surfsense_evals/src/surfsense_evals/suites/medical/medxpertqa/ingest.py

@@ -208,7 +208,6 @@ async def _upload_pdfs(
         result = await docs_client.upload(
             files=batch,
             search_space_id=ctx.search_space_id,
-            should_summarize=settings.should_summarize,
             use_vision_llm=settings.use_vision_llm,
             processing_mode=settings.processing_mode,
         )

surfsense_evals/src/surfsense_evals/suites/medical/medxpertqa/runner.py

@@ -169,7 +169,6 @@ _DESCRIPTION = (
 _DEFAULT_INGEST_SETTINGS = IngestSettings(
     use_vision_llm=True,
     processing_mode="basic",
-    should_summarize=False,
 )
 
 

surfsense_evals/src/surfsense_evals/suites/medical/mirage/ingest.py

@@ -480,7 +480,6 @@ async def run_ingest(
     upload_result = await docs_client.upload(
         files=[b.path for b in batches],
         search_space_id=ctx.search_space_id,
-        should_summarize=settings.should_summarize,
         use_vision_llm=settings.use_vision_llm,
         processing_mode=settings.processing_mode,
     )

surfsense_evals/src/surfsense_evals/suites/medical/mirage/runner.py

@@ -48,7 +48,6 @@ _DESCRIPTION = "MIRAGE (7,663 medical MCQs) — single-arm SurfSense per-task ac
 _DEFAULT_INGEST_SETTINGS = IngestSettings(
     use_vision_llm=False,
     processing_mode="basic",
-    should_summarize=False,
 )
 
 

surfsense_evals/src/surfsense_evals/suites/multimodal_doc/mmlongbench/ingest.py

@@ -225,7 +225,6 @@ async def _upload_pdfs(
         result = await docs_client.upload(
             files=batch,
             search_space_id=ctx.search_space_id,
-            should_summarize=settings.should_summarize,
             use_vision_llm=settings.use_vision_llm,
             processing_mode=settings.processing_mode,
         )

surfsense_evals/src/surfsense_evals/suites/multimodal_doc/mmlongbench/runner.py

@@ -178,7 +178,6 @@ _TEXT_ONLY_HINTS = ("gpt-5.4-mini", "gpt-3.5", "text-only", "instruct-")
 _DEFAULT_INGEST_SETTINGS = IngestSettings(
     use_vision_llm=True,
     processing_mode="basic",
-    should_summarize=False,
 )