From 5d658f7e73ae1a1505fb9b4f0b2bd5164125c1c3 Mon Sep 17 00:00:00 2001 From: Jesper Kristensen Date: Tue, 28 Jul 2026 13:40:54 +0200 Subject: [PATCH] Added support for external RAG ingestion and retrieval Delegates the RAG pipeline to external services instead of running it in-process: - Ingestion: when EXTERNAL_INGESTION_ENGINE=external, uploaded files are sent to the ingestion service (S3 reference or multipart), which handles text extraction, chunking, embedding and vector storage. File deletes notify the service so vectors don't outlive the file (best-effort, never blocks the user's delete). - Retrieval: when RAG_RETRIEVAL_ENGINE=external, document search goes to the retrieval service instead of the built-in vector DB, on both the chat files path and the native function calling path (query_knowledge_files). Optionally bypasses local query generation and forwards the recent conversation so the service can generate its own queries (RAG_EXTERNAL_MESSAGE_COUNT / USER_MESSAGES_ONLY). - Admin UI for both engines under Settings > Documents. Both engines are off by default; behaviour is unchanged from stock open-webui unless enabled. See https://github.com/AarhusAI/ingestion-service and https://github.com/AarhusAI/retrieval-agent. --- backend/open_webui/config.py | 45 +++ .../open_webui/retrieval/external_service.py | 256 ++++++++++++++++++ backend/open_webui/retrieval/utils.py | 35 +++ backend/open_webui/routers/files.py | 30 ++ backend/open_webui/routers/knowledge.py | 44 +++ backend/open_webui/routers/retrieval.py | 229 ++++++++++++++-- backend/open_webui/tools/builtin.py | 69 ++++- backend/open_webui/utils/middleware.py | 66 ++++- backend/open_webui/utils/tools.py | 7 + src/lib/apis/retrieval/index.ts | 15 + .../admin/Settings/Documents.svelte | 176 +++++++++++- 11 files changed, 933 insertions(+), 39 deletions(-) create mode 100644 backend/open_webui/retrieval/external_service.py diff --git a/backend/open_webui/config.py b/backend/open_webui/config.py index ec541f5a336e..8b26eebea84f 100644 --- a/backend/open_webui/config.py +++ b/backend/open_webui/config.py @@ -1033,6 +1033,36 @@ def reachable(host: str, port: int) -> bool: RAG_EXTERNAL_RERANKER_TIMEOUT = os.getenv('RAG_EXTERNAL_RERANKER_TIMEOUT', '') +# --- BEGIN EXTERNAL RETRIEVAL PATCH --- +# External retrieval engine: allows delegating document search to an external HTTP service +RAG_RETRIEVAL_ENGINE = os.getenv('RAG_RETRIEVAL_ENGINE', '') + +RAG_EXTERNAL_RETRIEVAL_URL = os.getenv('RAG_EXTERNAL_RETRIEVAL_URL', '') + +RAG_EXTERNAL_RETRIEVAL_API_KEY = os.getenv('RAG_EXTERNAL_RETRIEVAL_API_KEY', '') + +RAG_EXTERNAL_RETRIEVAL_TIMEOUT = os.getenv('RAG_EXTERNAL_RETRIEVAL_TIMEOUT', '') + +RAG_EXTERNAL_BYPASS_QUERY_GENERATION = os.getenv('RAG_EXTERNAL_BYPASS_QUERY_GENERATION', 'False').lower() == 'true' + +RAG_EXTERNAL_MESSAGE_COUNT = int(os.getenv('RAG_EXTERNAL_MESSAGE_COUNT', '10')) + +RAG_EXTERNAL_USER_MESSAGES_ONLY = os.getenv('RAG_EXTERNAL_USER_MESSAGES_ONLY', 'False').lower() == 'true' +# --- END EXTERNAL RETRIEVAL PATCH --- + +# --- BEGIN EXTERNAL INGESTION PATCH --- +# External ingestion engine: delegates document chunking, embedding, and vector +# storage to an external HTTP service instead of running save_docs_to_vector_db +# in-process. Default off; set EXTERNAL_INGESTION_ENGINE=external to enable. +EXTERNAL_INGESTION_ENGINE = os.getenv('EXTERNAL_INGESTION_ENGINE', '') + +EXTERNAL_INGESTION_URL = os.getenv('EXTERNAL_INGESTION_URL', '') + +EXTERNAL_INGESTION_API_KEY = os.getenv('EXTERNAL_INGESTION_API_KEY', '') + +EXTERNAL_INGESTION_TIMEOUT = os.getenv('EXTERNAL_INGESTION_TIMEOUT', '300') +# --- END EXTERNAL INGESTION PATCH --- + RAG_TEXT_SPLITTER = os.getenv('RAG_TEXT_SPLITTER', '') @@ -2886,6 +2916,21 @@ def feishu_oauth_register(oauth: OAuth): 'rag.external_reranker_url': RAG_EXTERNAL_RERANKER_URL, 'rag.external_reranker_api_key': RAG_EXTERNAL_RERANKER_API_KEY, 'rag.external_reranker_timeout': RAG_EXTERNAL_RERANKER_TIMEOUT, + # --- BEGIN EXTERNAL RETRIEVAL PATCH --- + 'rag.retrieval_engine': RAG_RETRIEVAL_ENGINE, + 'rag.external_retrieval_url': RAG_EXTERNAL_RETRIEVAL_URL, + 'rag.external_retrieval_api_key': RAG_EXTERNAL_RETRIEVAL_API_KEY, + 'rag.external_retrieval_timeout': RAG_EXTERNAL_RETRIEVAL_TIMEOUT, + 'rag.external_bypass_query_generation': RAG_EXTERNAL_BYPASS_QUERY_GENERATION, + 'rag.external_message_count': RAG_EXTERNAL_MESSAGE_COUNT, + 'rag.external_user_messages_only': RAG_EXTERNAL_USER_MESSAGES_ONLY, + # --- END EXTERNAL RETRIEVAL PATCH --- + # --- BEGIN EXTERNAL INGESTION PATCH --- + 'rag.external_ingestion_engine': EXTERNAL_INGESTION_ENGINE, + 'rag.external_ingestion_url': EXTERNAL_INGESTION_URL, + 'rag.external_ingestion_api_key': EXTERNAL_INGESTION_API_KEY, + 'rag.external_ingestion_timeout': EXTERNAL_INGESTION_TIMEOUT, + # --- END EXTERNAL INGESTION PATCH --- 'rag.text_splitter': RAG_TEXT_SPLITTER, 'rag.enable_markdown_header_text_splitter': ENABLE_MARKDOWN_HEADER_TEXT_SPLITTER, 'rag.tiktoken_encoding_name': TIKTOKEN_ENCODING_NAME, diff --git a/backend/open_webui/retrieval/external_service.py b/backend/open_webui/retrieval/external_service.py new file mode 100644 index 000000000000..227ed070b8cb --- /dev/null +++ b/backend/open_webui/retrieval/external_service.py @@ -0,0 +1,256 @@ +# --- BEGIN EXTERNAL RETRIEVAL PATCH --- +# External retrieval engine: delegates document search to an external HTTP service +# instead of querying the built-in vector DB directly. +# +# NOTE: this module is named external_service.py (not external.py) because +# upstream v0.11.0 ships its own retrieval/external.py ("external knowledge": +# per-KB connections straight to Qdrant/Milvus/pgvector). The two features are +# independent and coexist. +# --- END EXTERNAL RETRIEVAL PATCH --- +# --- BEGIN EXTERNAL INGESTION PATCH --- +# External ingestion engine: delegates document chunking, embedding, and vector +# storage to an external HTTP service instead of running save_docs_to_vector_db +# in-process. +# --- END EXTERNAL INGESTION PATCH --- + +import logging +from types import SimpleNamespace + +import requests +from open_webui.env import ENABLE_FORWARD_USER_INFO_HEADERS, REQUESTS_VERIFY +from open_webui.models.config import Config +from open_webui.utils.headers import include_user_info_headers + +log = logging.getLogger(__name__) + + +# Maps our config field names to their storage keys in the per-key Config +# store. The storage keys are unchanged from the pre-v0.11.0 ConfigVar +# declarations, so values persisted before the upgrade carry over via the +# 3ff2c63645b8 (reshape config to per-key rows) migration. +EXTERNAL_RAG_CONFIG_KEYS = { + 'RAG_RETRIEVAL_ENGINE': 'rag.retrieval_engine', + 'RAG_EXTERNAL_RETRIEVAL_URL': 'rag.external_retrieval_url', + 'RAG_EXTERNAL_RETRIEVAL_API_KEY': 'rag.external_retrieval_api_key', + 'RAG_EXTERNAL_RETRIEVAL_TIMEOUT': 'rag.external_retrieval_timeout', + 'RAG_EXTERNAL_BYPASS_QUERY_GENERATION': 'rag.external_bypass_query_generation', + 'RAG_EXTERNAL_MESSAGE_COUNT': 'rag.external_message_count', + 'RAG_EXTERNAL_USER_MESSAGES_ONLY': 'rag.external_user_messages_only', + 'EXTERNAL_INGESTION_ENGINE': 'rag.external_ingestion_engine', + 'EXTERNAL_INGESTION_URL': 'rag.external_ingestion_url', + 'EXTERNAL_INGESTION_API_KEY': 'rag.external_ingestion_api_key', + 'EXTERNAL_INGESTION_TIMEOUT': 'rag.external_ingestion_timeout', +} + + +async def get_external_rag_config() -> SimpleNamespace: + """Read our external-RAG keys from the per-key Config store. + + For call sites that don't already have a RetrievalConfig in scope + (retrieval/utils.py, tools/builtin.py, routers/files.py, + routers/knowledge.py). Missing keys degrade to None. + """ + values = await Config.get_many(*EXTERNAL_RAG_CONFIG_KEYS.values()) + return SimpleNamespace(**{field: values.get(key) for field, key in EXTERNAL_RAG_CONFIG_KEYS.items()}) + + +def trim_messages_for_external(messages: list[dict], *, count: int, user_only: bool) -> list[dict]: + """Trim the conversation before forwarding it to the external retrieval + service, honouring RAG_EXTERNAL_MESSAGE_COUNT / RAG_EXTERNAL_USER_MESSAGES_ONLY. + + Shared by utils/middleware.py (chat_completion_files_handler) and + tools/builtin.py (query_knowledge_files) so both paths trim identically. + """ + candidates = [m for m in messages if m.get('role') == 'user'] if user_only else messages + return [{'role': m.get('role', ''), 'content': m.get('content', '')} for m in candidates[-count:]] + + +def query_external_retrieval( + url: str, + api_key: str, + queries: list[str], + collection_names: list[str], + k: int, + timeout: str | None = None, + user=None, + messages: list[dict] | None = None, +) -> dict | None: + """ + Query an external retrieval service. + + POST {url}/search with queries + collection_names + k. + Optionally includes the chat messages so the external service can + extract/generate its own queries from the conversation. Open WebUI's + QUERY_GENERATION_PROMPT_TEMPLATE is intentionally NOT forwarded — the + external service runs its own query generation. + Returns dict with keys: documents, metadatas, distances (matching internal format). + Returns None on error. + """ + payload = { + 'queries': queries, + 'collection_names': collection_names, + 'k': k, + } + + if messages is not None: + payload['messages'] = messages + + try: + headers = { + 'Content-Type': 'application/json', + 'Authorization': f'Bearer {api_key}', + } + + if ENABLE_FORWARD_USER_INFO_HEADERS and user: + headers = include_user_info_headers(headers, user) + + request_timeout = int(timeout) if timeout else None + + log.info( + f'query_external_retrieval: url={url}, queries={queries}, ' + f'messages={len(messages) if messages else 0}, ' + f'collections={collection_names}, k={k}' + ) + + r = requests.post( + f'{url}/search', + headers=headers, + json=payload, + timeout=request_timeout, + verify=REQUESTS_VERIFY, + ) + r.raise_for_status() + data = r.json() + + if 'documents' in data: + return data + else: + log.error('No documents found in external retrieval response') + return None + + except Exception as e: + log.exception(f'Error in external retrieval: {e}') + return None + + +# --- BEGIN EXTERNAL INGESTION PATCH --- +def process_file_external_ingestion( + url: str, + api_key: str, + file_id: str, + filename: str, + collection_name: str, + user_id: str, + local_file_path: str | None = None, + s3_bucket: str | None = None, + s3_key: str | None = None, + timeout: int | None = 300, +) -> dict | None: + """ + Delegate document ingestion to an external HTTP service. + + PUT {url}/api/v1/ingest with either: + - JSON body containing s3_bucket + s3_key (preferred when storage is S3) + - multipart body containing the file (fallback when storage is local) + + Returns the service's response dict on success + ({"status": True, "collection_name": ..., "chunks_count": ...}) + or None on transport / server error. Caller treats None as failure. + """ + try: + headers = {'Authorization': f'Bearer {api_key}'} + endpoint = f'{url.rstrip("/")}/api/v1/ingest' + + if s3_bucket and s3_key: + payload = { + 's3_bucket': s3_bucket, + 's3_key': s3_key, + 'file_id': file_id, + 'filename': filename, + 'collection_name': collection_name, + 'collection_type': 'file', + 'user_id': user_id, + 'overwrite': True, + } + log.info( + f'process_file_external_ingestion (s3): file_id={file_id}, ' + f'collection={collection_name}, s3={s3_bucket}/{s3_key}' + ) + r = requests.put( + endpoint, + headers={**headers, 'Content-Type': 'application/json'}, + json=payload, + timeout=timeout, + verify=REQUESTS_VERIFY, + ) + elif local_file_path: + log.info( + f'process_file_external_ingestion (multipart): file_id={file_id}, ' + f'collection={collection_name}, path={local_file_path}' + ) + with open(local_file_path, 'rb') as fh: + files = {'file': (filename, fh)} + data = { + 'file_id': file_id, + 'filename': filename, + 'collection_name': collection_name, + 'collection_type': 'file', + 'user_id': user_id, + 'overwrite': 'true', + } + r = requests.put( + endpoint, + headers=headers, + files=files, + data=data, + timeout=timeout, + verify=REQUESTS_VERIFY, + ) + else: + log.error('process_file_external_ingestion: no S3 reference and no local file path') + return None + + r.raise_for_status() + return r.json() + + except Exception as e: + log.exception(f'Error in external ingestion: {e}') + return None + + +def delete_file_external_ingestion( + url: str, + api_key: str, + file_id: str, + timeout: int = 300, +) -> dict | None: + """Tell the external ingestion service to drop a file's vectors. + + DELETE {url}/api/v1/documents/{file_id}. The vector store lives behind the + external service now, so Open WebUI's own vector-DB cleanup no longer reaches + it — this call keeps the two in sync when a file is deleted. + + Best-effort: returns the service's response dict on success, or None on any + transport/server error (logged, never raised). Callers MUST NOT let a failed + cleanup block the user's file deletion. ``file_id`` is the bare file UUID — + the service matches on meta.file_id, not the "file-" collection name. + """ + try: + headers = {'Authorization': f'Bearer {api_key}'} + endpoint = f'{url.rstrip("/")}/api/v1/documents/{file_id}' + log.info(f'delete_file_external_ingestion: file_id={file_id}') + r = requests.delete( + endpoint, + headers=headers, + timeout=timeout, + verify=REQUESTS_VERIFY, + ) + r.raise_for_status() + return r.json() + + except Exception as e: + log.exception(f'Error in external ingestion delete: {e}') + return None + + +# --- END EXTERNAL INGESTION PATCH --- diff --git a/backend/open_webui/retrieval/utils.py b/backend/open_webui/retrieval/utils.py index 952b1e6b2615..b5bb199e5d13 100644 --- a/backend/open_webui/retrieval/utils.py +++ b/backend/open_webui/retrieval/utils.py @@ -45,6 +45,14 @@ from open_webui.retrieval.loaders.youtube import YoutubeLoader from open_webui.retrieval.vector.async_client import ASYNC_VECTOR_DB_CLIENT from open_webui.retrieval.external import retrieve_external_knowledge + +# --- BEGIN EXTERNAL RETRIEVAL PATCH --- +from open_webui.retrieval.external_service import ( + get_external_rag_config, + query_external_retrieval, +) + +# --- END EXTERNAL RETRIEVAL PATCH --- from open_webui.retrieval.vector.factory import VECTOR_DB_CLIENT from open_webui.retrieval.vector.main import GetResult, SearchResult from open_webui.retrieval.web.utils import get_web_loader @@ -1334,10 +1342,18 @@ async def get_sources_from_items( hybrid_search, full_context=False, user: UserModel | None = None, + # --- BEGIN EXTERNAL RETRIEVAL PATCH --- + messages: list | None = None, + # --- END EXTERNAL RETRIEVAL PATCH --- ): log.debug('items: %s %s %s %s %s', items, queries, embedding_function, reranking_function, full_context) bypass_embedding_and_retrieval = await Config.get('rag.bypass_embedding_and_retrieval') + # --- BEGIN EXTERNAL RETRIEVAL PATCH --- + # Read once per request, not per item — used in the collection-query + # fallback below to route search to the external retrieval service. + external_rag_config = await get_external_rag_config() + # --- END EXTERNAL RETRIEVAL PATCH --- extracted_collections = [] query_results = [] folder_items = set() @@ -1622,6 +1638,25 @@ async def get_sources_from_items( # Sync helper makes blocking VECTOR_DB_CLIENT calls; # offload so the async caller's event loop stays free. query_result = await asyncio.to_thread(get_all_items_from_collections, collection_names) + # --- BEGIN EXTERNAL RETRIEVAL PATCH --- + elif external_rag_config.RAG_RETRIEVAL_ENGINE == 'external': + # The external retrieval service runs its own query + # generation, so Open WebUI's QUERY_GENERATION_PROMPT_TEMPLATE + # is intentionally NOT forwarded. + # query_external_retrieval is sync (requests-based); offload + # so the async caller's event loop stays free. + query_result = await asyncio.to_thread( + query_external_retrieval, + url=external_rag_config.RAG_EXTERNAL_RETRIEVAL_URL, + api_key=external_rag_config.RAG_EXTERNAL_RETRIEVAL_API_KEY, + queries=queries, + collection_names=list(collection_names), + k=k, + timeout=external_rag_config.RAG_EXTERNAL_RETRIEVAL_TIMEOUT, + user=user, + messages=messages, + ) + # --- END EXTERNAL RETRIEVAL PATCH --- else: query_result = await query_collection( request, diff --git a/backend/open_webui/routers/files.py b/backend/open_webui/routers/files.py index 6dcad4c421db..ccffc2758ffc 100644 --- a/backend/open_webui/routers/files.py +++ b/backend/open_webui/routers/files.py @@ -40,6 +40,14 @@ from open_webui.models.groups import Groups from open_webui.models.knowledge import Knowledges from open_webui.models.users import Users + +# --- BEGIN EXTERNAL INGESTION PATCH --- +from open_webui.retrieval.external_service import ( + delete_file_external_ingestion, + get_external_rag_config, +) + +# --- END EXTERNAL INGESTION PATCH --- from open_webui.retrieval.vector.async_client import ASYNC_VECTOR_DB_CLIENT from open_webui.routers.audio import transcribe from open_webui.routers.retrieval import ProcessFileForm, process_file @@ -1042,6 +1050,28 @@ async def delete_file_by_id( subject_id=id, data={'filename': file.filename}, ) + + # --- BEGIN EXTERNAL INGESTION PATCH --- + # The vector store lives behind the external ingestion service now, + # so the ASYNC_VECTOR_DB_CLIENT deletes above don't reach it. Notify + # the service so this file's chunks don't outlive the file. Kept + # OUTSIDE the try above (which re-raises as HTTP 400) and best-effort + # — a failed cleanup must never fail the user's delete. + _cfg = await get_external_rag_config() + if _cfg.EXTERNAL_INGESTION_ENGINE == 'external' and _cfg.EXTERNAL_INGESTION_URL: + try: + _timeout = int(_cfg.EXTERNAL_INGESTION_TIMEOUT) if _cfg.EXTERNAL_INGESTION_TIMEOUT else 300 + await asyncio.to_thread( + delete_file_external_ingestion, + url=_cfg.EXTERNAL_INGESTION_URL, + api_key=_cfg.EXTERNAL_INGESTION_API_KEY, + file_id=id, + timeout=_timeout, + ) + except Exception as e: + log.debug(f'external ingestion delete for {id}: {e}') + # --- END EXTERNAL INGESTION PATCH --- + return {'message': 'File deleted successfully'} else: raise HTTPException( diff --git a/backend/open_webui/routers/knowledge.py b/backend/open_webui/routers/knowledge.py index cd98681efd1c..a90a696394a9 100644 --- a/backend/open_webui/routers/knowledge.py +++ b/backend/open_webui/routers/knowledge.py @@ -33,6 +33,14 @@ from open_webui.models.models import ModelForm, Models from open_webui.retrieval.vector.async_client import ASYNC_VECTOR_DB_CLIENT from open_webui.retrieval.external import retrieve_external_knowledge, retrieve_external_knowledge_for_connection + +# --- BEGIN EXTERNAL INGESTION PATCH --- +from open_webui.retrieval.external_service import ( + delete_file_external_ingestion, + get_external_rag_config, +) + +# --- END EXTERNAL INGESTION PATCH --- from open_webui.routers.retrieval import ( BatchProcessFilesForm, ProcessFileForm, @@ -1632,6 +1640,24 @@ async def remove_file_from_knowledge_by_id( # Delete file from database await Files.delete_file_by_id(form_data.file_id, db=db) + # --- BEGIN EXTERNAL INGESTION PATCH --- + # File is permanently deleted here (delete_file branch), so clean up its + # vectors in the external ingestion service too. Best-effort, never raises. + _cfg = await get_external_rag_config() + if _cfg.EXTERNAL_INGESTION_ENGINE == 'external' and _cfg.EXTERNAL_INGESTION_URL: + try: + _timeout = int(_cfg.EXTERNAL_INGESTION_TIMEOUT) if _cfg.EXTERNAL_INGESTION_TIMEOUT else 300 + await asyncio.to_thread( + delete_file_external_ingestion, + url=_cfg.EXTERNAL_INGESTION_URL, + api_key=_cfg.EXTERNAL_INGESTION_API_KEY, + file_id=form_data.file_id, + timeout=_timeout, + ) + except Exception as e: + log.debug(f'external ingestion delete for {form_data.file_id}: {e}') + # --- END EXTERNAL INGESTION PATCH --- + if knowledge: response = KnowledgeFilesResponse( **knowledge.model_dump(), @@ -1969,6 +1995,24 @@ async def sync_knowledge_cleanup( except Exception: pass + # --- BEGIN EXTERNAL INGESTION PATCH --- + # Stale file permanently deleted during sync — clean up its vectors + # in the external ingestion service too. Best-effort, never raises. + _cfg = await get_external_rag_config() + if _cfg.EXTERNAL_INGESTION_ENGINE == 'external' and _cfg.EXTERNAL_INGESTION_URL: + try: + _timeout = int(_cfg.EXTERNAL_INGESTION_TIMEOUT) if _cfg.EXTERNAL_INGESTION_TIMEOUT else 300 + await asyncio.to_thread( + delete_file_external_ingestion, + url=_cfg.EXTERNAL_INGESTION_URL, + api_key=_cfg.EXTERNAL_INGESTION_API_KEY, + file_id=file_id, + timeout=_timeout, + ) + except Exception as e: + log.debug(f'external ingestion delete for {file_id}: {e}') + # --- END EXTERNAL INGESTION PATCH --- + # ── Remove orphaned directories (children before parents) ── for dir_id in reversed(form_data.dir_ids): # Only delete directories that belong to this knowledge base. diff --git a/backend/open_webui/routers/retrieval.py b/backend/open_webui/routers/retrieval.py index 4bb2fbca8823..acc390ae38c7 100644 --- a/backend/open_webui/routers/retrieval.py +++ b/backend/open_webui/routers/retrieval.py @@ -62,6 +62,10 @@ from open_webui.models.knowledge import Knowledges from open_webui.models.config import Config +# --- BEGIN EXTERNAL INGESTION PATCH --- +from open_webui.retrieval.external_service import process_file_external_ingestion + +# --- END EXTERNAL INGESTION PATCH --- # Document loaders from open_webui.retrieval.loaders.youtube import YoutubeLoader from open_webui.retrieval.utils import ( @@ -353,6 +357,21 @@ def get_rf( 'RAG_EXTERNAL_RERANKER_API_KEY': 'rag.external_reranker_api_key', 'RAG_EXTERNAL_RERANKER_TIMEOUT': 'rag.external_reranker_timeout', 'RAG_EXTERNAL_RERANKER_URL': 'rag.external_reranker_url', + # --- BEGIN EXTERNAL RETRIEVAL PATCH --- + 'RAG_RETRIEVAL_ENGINE': 'rag.retrieval_engine', + 'RAG_EXTERNAL_RETRIEVAL_URL': 'rag.external_retrieval_url', + 'RAG_EXTERNAL_RETRIEVAL_API_KEY': 'rag.external_retrieval_api_key', + 'RAG_EXTERNAL_RETRIEVAL_TIMEOUT': 'rag.external_retrieval_timeout', + 'RAG_EXTERNAL_BYPASS_QUERY_GENERATION': 'rag.external_bypass_query_generation', + 'RAG_EXTERNAL_MESSAGE_COUNT': 'rag.external_message_count', + 'RAG_EXTERNAL_USER_MESSAGES_ONLY': 'rag.external_user_messages_only', + # --- END EXTERNAL RETRIEVAL PATCH --- + # --- BEGIN EXTERNAL INGESTION PATCH --- + 'EXTERNAL_INGESTION_ENGINE': 'rag.external_ingestion_engine', + 'EXTERNAL_INGESTION_URL': 'rag.external_ingestion_url', + 'EXTERNAL_INGESTION_API_KEY': 'rag.external_ingestion_api_key', + 'EXTERNAL_INGESTION_TIMEOUT': 'rag.external_ingestion_timeout', + # --- END EXTERNAL INGESTION PATCH --- 'RAG_FULL_CONTEXT': 'rag.full_context', 'RAG_OLLAMA_API_KEY': 'rag.ollama.api_key', 'RAG_OLLAMA_BASE_URL': 'rag.ollama.base_url', @@ -674,6 +693,21 @@ async def get_rag_config(request: Request, user=Depends(get_admin_user)): 'RAG_EXTERNAL_RERANKER_URL': config.RAG_EXTERNAL_RERANKER_URL, 'RAG_EXTERNAL_RERANKER_API_KEY': config.RAG_EXTERNAL_RERANKER_API_KEY, 'RAG_EXTERNAL_RERANKER_TIMEOUT': config.RAG_EXTERNAL_RERANKER_TIMEOUT, + # --- BEGIN EXTERNAL RETRIEVAL PATCH --- + 'RAG_RETRIEVAL_ENGINE': config.RAG_RETRIEVAL_ENGINE, + 'RAG_EXTERNAL_RETRIEVAL_URL': config.RAG_EXTERNAL_RETRIEVAL_URL, + 'RAG_EXTERNAL_RETRIEVAL_API_KEY': config.RAG_EXTERNAL_RETRIEVAL_API_KEY, + 'RAG_EXTERNAL_RETRIEVAL_TIMEOUT': config.RAG_EXTERNAL_RETRIEVAL_TIMEOUT, + 'RAG_EXTERNAL_BYPASS_QUERY_GENERATION': config.RAG_EXTERNAL_BYPASS_QUERY_GENERATION, + 'RAG_EXTERNAL_MESSAGE_COUNT': config.RAG_EXTERNAL_MESSAGE_COUNT, + 'RAG_EXTERNAL_USER_MESSAGES_ONLY': config.RAG_EXTERNAL_USER_MESSAGES_ONLY, + # --- END EXTERNAL RETRIEVAL PATCH --- + # --- BEGIN EXTERNAL INGESTION PATCH --- + 'EXTERNAL_INGESTION_ENGINE': config.EXTERNAL_INGESTION_ENGINE, + 'EXTERNAL_INGESTION_URL': config.EXTERNAL_INGESTION_URL, + 'EXTERNAL_INGESTION_API_KEY': config.EXTERNAL_INGESTION_API_KEY, + 'EXTERNAL_INGESTION_TIMEOUT': config.EXTERNAL_INGESTION_TIMEOUT, + # --- END EXTERNAL INGESTION PATCH --- # Chunking settings 'TEXT_SPLITTER': config.TEXT_SPLITTER, 'RAG_TOKENIZER_MODEL': config.RAG_TOKENIZER_MODEL, @@ -912,6 +946,23 @@ class ConfigForm(BaseModel): RAG_EXTERNAL_RERANKER_API_KEY: str | None = None RAG_EXTERNAL_RERANKER_TIMEOUT: str | None = None + # --- BEGIN EXTERNAL RETRIEVAL PATCH --- + RAG_RETRIEVAL_ENGINE: str | None = None + RAG_EXTERNAL_RETRIEVAL_URL: str | None = None + RAG_EXTERNAL_RETRIEVAL_API_KEY: str | None = None + RAG_EXTERNAL_RETRIEVAL_TIMEOUT: str | None = None + RAG_EXTERNAL_BYPASS_QUERY_GENERATION: bool | None = None + RAG_EXTERNAL_MESSAGE_COUNT: int | None = None + RAG_EXTERNAL_USER_MESSAGES_ONLY: bool | None = None + # --- END EXTERNAL RETRIEVAL PATCH --- + + # --- BEGIN EXTERNAL INGESTION PATCH --- + EXTERNAL_INGESTION_ENGINE: str | None = None + EXTERNAL_INGESTION_URL: str | None = None + EXTERNAL_INGESTION_API_KEY: str | None = None + EXTERNAL_INGESTION_TIMEOUT: str | None = None + # --- END EXTERNAL INGESTION PATCH --- + # Chunking settings TEXT_SPLITTER: str | None = None RAG_TOKENIZER_MODEL: str | None = None @@ -1160,6 +1211,74 @@ async def update_rag_config(request: Request, form_data: ConfigForm, user=Depend else config.RAG_RERANKING_BATCH_SIZE ) + # --- BEGIN EXTERNAL RETRIEVAL PATCH --- + config.RAG_RETRIEVAL_ENGINE = ( + form_data.RAG_RETRIEVAL_ENGINE if form_data.RAG_RETRIEVAL_ENGINE is not None else config.RAG_RETRIEVAL_ENGINE + ) + + config.RAG_EXTERNAL_RETRIEVAL_URL = ( + form_data.RAG_EXTERNAL_RETRIEVAL_URL + if form_data.RAG_EXTERNAL_RETRIEVAL_URL is not None + else config.RAG_EXTERNAL_RETRIEVAL_URL + ) + + config.RAG_EXTERNAL_RETRIEVAL_API_KEY = ( + form_data.RAG_EXTERNAL_RETRIEVAL_API_KEY + if form_data.RAG_EXTERNAL_RETRIEVAL_API_KEY is not None + else config.RAG_EXTERNAL_RETRIEVAL_API_KEY + ) + + config.RAG_EXTERNAL_RETRIEVAL_TIMEOUT = ( + form_data.RAG_EXTERNAL_RETRIEVAL_TIMEOUT + if form_data.RAG_EXTERNAL_RETRIEVAL_TIMEOUT is not None + else config.RAG_EXTERNAL_RETRIEVAL_TIMEOUT + ) + + config.RAG_EXTERNAL_BYPASS_QUERY_GENERATION = ( + form_data.RAG_EXTERNAL_BYPASS_QUERY_GENERATION + if form_data.RAG_EXTERNAL_BYPASS_QUERY_GENERATION is not None + else config.RAG_EXTERNAL_BYPASS_QUERY_GENERATION + ) + + config.RAG_EXTERNAL_MESSAGE_COUNT = ( + form_data.RAG_EXTERNAL_MESSAGE_COUNT + if form_data.RAG_EXTERNAL_MESSAGE_COUNT is not None + else config.RAG_EXTERNAL_MESSAGE_COUNT + ) + + config.RAG_EXTERNAL_USER_MESSAGES_ONLY = ( + form_data.RAG_EXTERNAL_USER_MESSAGES_ONLY + if form_data.RAG_EXTERNAL_USER_MESSAGES_ONLY is not None + else config.RAG_EXTERNAL_USER_MESSAGES_ONLY + ) + # --- END EXTERNAL RETRIEVAL PATCH --- + + # --- BEGIN EXTERNAL INGESTION PATCH --- + config.EXTERNAL_INGESTION_ENGINE = ( + form_data.EXTERNAL_INGESTION_ENGINE + if form_data.EXTERNAL_INGESTION_ENGINE is not None + else config.EXTERNAL_INGESTION_ENGINE + ) + + config.EXTERNAL_INGESTION_URL = ( + form_data.EXTERNAL_INGESTION_URL + if form_data.EXTERNAL_INGESTION_URL is not None + else config.EXTERNAL_INGESTION_URL + ) + + config.EXTERNAL_INGESTION_API_KEY = ( + form_data.EXTERNAL_INGESTION_API_KEY + if form_data.EXTERNAL_INGESTION_API_KEY is not None + else config.EXTERNAL_INGESTION_API_KEY + ) + + config.EXTERNAL_INGESTION_TIMEOUT = ( + form_data.EXTERNAL_INGESTION_TIMEOUT + if form_data.EXTERNAL_INGESTION_TIMEOUT is not None + else config.EXTERNAL_INGESTION_TIMEOUT + ) + # --- END EXTERNAL INGESTION PATCH --- + log.info(f'Updating reranking model: {config.RAG_RERANKING_MODEL} to {form_data.RAG_RERANKING_MODEL}') try: config.RAG_RERANKING_MODEL = ( @@ -1381,6 +1500,21 @@ async def update_rag_config(request: Request, form_data: ConfigForm, user=Depend 'RAG_EXTERNAL_RERANKER_URL': config.RAG_EXTERNAL_RERANKER_URL, 'RAG_EXTERNAL_RERANKER_API_KEY': config.RAG_EXTERNAL_RERANKER_API_KEY, 'RAG_EXTERNAL_RERANKER_TIMEOUT': config.RAG_EXTERNAL_RERANKER_TIMEOUT, + # --- BEGIN EXTERNAL RETRIEVAL PATCH --- + 'RAG_RETRIEVAL_ENGINE': config.RAG_RETRIEVAL_ENGINE, + 'RAG_EXTERNAL_RETRIEVAL_URL': config.RAG_EXTERNAL_RETRIEVAL_URL, + 'RAG_EXTERNAL_RETRIEVAL_API_KEY': config.RAG_EXTERNAL_RETRIEVAL_API_KEY, + 'RAG_EXTERNAL_RETRIEVAL_TIMEOUT': config.RAG_EXTERNAL_RETRIEVAL_TIMEOUT, + 'RAG_EXTERNAL_BYPASS_QUERY_GENERATION': config.RAG_EXTERNAL_BYPASS_QUERY_GENERATION, + 'RAG_EXTERNAL_MESSAGE_COUNT': config.RAG_EXTERNAL_MESSAGE_COUNT, + 'RAG_EXTERNAL_USER_MESSAGES_ONLY': config.RAG_EXTERNAL_USER_MESSAGES_ONLY, + # --- END EXTERNAL RETRIEVAL PATCH --- + # --- BEGIN EXTERNAL INGESTION PATCH --- + 'EXTERNAL_INGESTION_ENGINE': config.EXTERNAL_INGESTION_ENGINE, + 'EXTERNAL_INGESTION_URL': config.EXTERNAL_INGESTION_URL, + 'EXTERNAL_INGESTION_API_KEY': config.EXTERNAL_INGESTION_API_KEY, + 'EXTERNAL_INGESTION_TIMEOUT': config.EXTERNAL_INGESTION_TIMEOUT, + # --- END EXTERNAL INGESTION PATCH --- # Chunking settings 'TEXT_SPLITTER': config.TEXT_SPLITTER, 'RAG_TOKENIZER_MODEL': config.RAG_TOKENIZER_MODEL, @@ -1993,26 +2127,83 @@ async def process_file( await db.commit() # External embedding API takes time (5-60s+). - # Subsequent updates use fresh async sessions. - # NOTE: save_docs_to_vector_db is a sync function that - # calls asyncio.run_coroutine_threadsafe(..., main_loop).result() - # which blocks the calling thread. We MUST run it in a - # worker thread to avoid deadlocking the event loop. - result = await run_in_threadpool( - save_docs_to_vector_db, - request, - docs=docs, - collection_name=collection_name, - config=config, - metadata={ - 'file_id': file.id, - 'name': file.filename, - 'hash': hash, - }, - add=(True if form_data.collection_name else False), - user=user, + + # --- BEGIN EXTERNAL INGESTION PATCH --- + # When EXTERNAL_INGESTION_ENGINE == "external", delegate + # chunk + embed + vector-store to the external ingestion + # service. Pre-extracted content (form_data.content) keeps + # the in-process pipeline; everything else (fresh upload, + # KB add/update, reindex) routes to the external service, + # which is idempotent per file_id via overwrite=true. + _use_external_ingest = ( + config.EXTERNAL_INGESTION_ENGINE == 'external' and not form_data.content and bool(file.path) ) - log.info(f'added {len(docs)} items to collection {collection_name}') + + if _use_external_ingest: + _s3_bucket, _s3_key = None, None + if file.path.startswith('s3://'): + _without_scheme = file.path[len('s3://') :] + if '/' in _without_scheme: + _s3_bucket, _s3_key = _without_scheme.split('/', 1) + + # Reindex / KB-add paths skip the fresh-upload branch + # above, so the local `file_path` variable may be + # unset. Fetch a local copy on demand for multipart + # fallback; S3 mode doesn't need it. + _local_file_path = None + if not _s3_bucket: + try: + _local_file_path = await asyncio.to_thread(Storage.get_file, file.path) + except Exception: + _local_file_path = None + + _timeout_str = config.EXTERNAL_INGESTION_TIMEOUT + _timeout = int(_timeout_str) if _timeout_str else 300 + + _ingest_result = await asyncio.to_thread( + process_file_external_ingestion, + url=config.EXTERNAL_INGESTION_URL, + api_key=config.EXTERNAL_INGESTION_API_KEY, + file_id=file.id, + filename=file.filename, + collection_name=collection_name, + user_id=file.user_id, + local_file_path=_local_file_path, + s3_bucket=_s3_bucket, + s3_key=_s3_key, + timeout=_timeout, + ) + + result = bool(_ingest_result and _ingest_result.get('status')) + if not result: + _err = (_ingest_result or {}).get('error') or 'External ingestion failed' + raise Exception(_err) + log.info( + f'external ingestion completed for file {file.id}: ' + f'chunks={(_ingest_result or {}).get("chunks_count")}' + ) + else: + # Subsequent updates use fresh async sessions. + # NOTE: save_docs_to_vector_db is a sync function that + # calls asyncio.run_coroutine_threadsafe(..., main_loop).result() + # which blocks the calling thread. We MUST run it in a + # worker thread to avoid deadlocking the event loop. + result = await run_in_threadpool( + save_docs_to_vector_db, + request, + docs=docs, + collection_name=collection_name, + config=config, + metadata={ + 'file_id': file.id, + 'name': file.filename, + 'hash': hash, + }, + add=(True if form_data.collection_name else False), + user=user, + ) + log.info(f'added {len(docs)} items to collection {collection_name}') + # --- END EXTERNAL INGESTION PATCH --- if result: # Fresh session for the final update. diff --git a/backend/open_webui/tools/builtin.py b/backend/open_webui/tools/builtin.py index 1b721f19a1b0..0b5c795176ed 100644 --- a/backend/open_webui/tools/builtin.py +++ b/backend/open_webui/tools/builtin.py @@ -30,6 +30,15 @@ from open_webui.models.messages import Message, Messages from open_webui.models.notes import Notes from open_webui.models.users import UserModel + +# --- BEGIN EXTERNAL RETRIEVAL PATCH --- +from open_webui.retrieval.external_service import ( + get_external_rag_config, + query_external_retrieval, + trim_messages_for_external, +) + +# --- END EXTERNAL RETRIEVAL PATCH --- from open_webui.retrieval.utils import get_content_from_url from open_webui.retrieval.vector.async_client import ASYNC_VECTOR_DB_CLIENT from open_webui.routers.images import ( @@ -2996,6 +3005,9 @@ async def query_knowledge_files( __request__: Request = None, __user__: dict = None, __model_knowledge__: list[dict] = None, + # --- BEGIN EXTERNAL RETRIEVAL PATCH --- + __messages__: list = None, + # --- END EXTERNAL RETRIEVAL PATCH --- ) -> str: """ Search knowledge base files using semantic/vector search. Searches across collections (KBs), @@ -3044,8 +3056,15 @@ async def query_knowledge_files( user_role = __user__.get('role', 'user') user_group_ids = [group.id for group in await Groups.get_groups_by_member_id(user_id)] + # --- BEGIN EXTERNAL RETRIEVAL PATCH --- + external_rag_config = await get_external_rag_config() + use_external = external_rag_config.RAG_RETRIEVAL_ENGINE == 'external' + + # The external retrieval engine embeds server-side, so a local embedding + # function is only required for the internal vector-DB path. + # --- END EXTERNAL RETRIEVAL PATCH --- embedding_function = getattr(__request__.app.state, 'EMBEDDING_FUNCTION', None) - if not embedding_function: + if not use_external and not embedding_function: return json.dumps({'error': 'Embedding function not configured'}) user_model = UserModel.model_construct(id=user_id, role=user_role) @@ -3151,13 +3170,47 @@ async def query_knowledge_files( # Query vector collections if any if collection_names: - query_results = await query_collection( - __request__, - collection_names=collection_names, - queries=[query], - embedding_function=lambda queries, prefix: embedding_function(queries, prefix=prefix, user=user_model), - k=count, - ) + # --- BEGIN EXTERNAL RETRIEVAL PATCH --- + if use_external: + # Route to the external retrieval engine, mirroring the patch in + # retrieval/utils.py:get_sources_from_items. The LLM already + # generated `query` via the tool call, but we still forward the + # recent conversation so the external service's agentic pipeline + # has context for its own query generation. Trimming is shared + # with chat_completion_files_handler so both paths honour the + # same RAG_EXTERNAL_MESSAGE_COUNT / USER_MESSAGES_ONLY settings. + messages_for_external = None + if __messages__: + messages_for_external = trim_messages_for_external( + __messages__, + count=external_rag_config.RAG_EXTERNAL_MESSAGE_COUNT, + user_only=external_rag_config.RAG_EXTERNAL_USER_MESSAGES_ONLY, + ) + + # query_external_retrieval is sync (requests-based); offload so + # the async caller's event loop stays free. + query_results = await asyncio.to_thread( + query_external_retrieval, + url=external_rag_config.RAG_EXTERNAL_RETRIEVAL_URL, + api_key=external_rag_config.RAG_EXTERNAL_RETRIEVAL_API_KEY, + queries=[query], + collection_names=collection_names, + k=count, + timeout=external_rag_config.RAG_EXTERNAL_RETRIEVAL_TIMEOUT, + user=__user__, + messages=messages_for_external, + ) + else: + query_results = await query_collection( + __request__, + collection_names=collection_names, + queries=[query], + embedding_function=lambda queries, prefix: embedding_function( + queries, prefix=prefix, user=user_model + ), + k=count, + ) + # --- END EXTERNAL RETRIEVAL PATCH --- if query_results and 'documents' in query_results: documents = query_results.get('documents', [[]])[0] diff --git a/backend/open_webui/utils/middleware.py b/backend/open_webui/utils/middleware.py index 271d072992db..64f9c99b38cc 100644 --- a/backend/open_webui/utils/middleware.py +++ b/backend/open_webui/utils/middleware.py @@ -48,6 +48,11 @@ from open_webui.models.oauth_sessions import OAuthSessions from open_webui.models.users import UserModel, Users from open_webui.events import EVENTS, publish_event + +# --- BEGIN EXTERNAL RETRIEVAL PATCH --- +from open_webui.retrieval.external_service import trim_messages_for_external + +# --- END EXTERNAL RETRIEVAL PATCH --- from open_webui.retrieval.utils import get_sources_from_items from open_webui.routers.images import ( CreateImageForm, @@ -1834,8 +1839,49 @@ async def chat_completion_files_handler( # Check if all files are in full context mode all_full_context = all(item.get('context') == 'full' for item in files) + # One batched SELECT instead of six sequential round trips. + # --- BEGIN EXTERNAL RETRIEVAL PATCH --- + # Moved above the query-generation block (upstream has it just before + # get_sources_from_items) and extended with the external-retrieval keys + # so the bypass decision below can reuse the same round trip. + # --- END EXTERNAL RETRIEVAL PATCH --- + rag_config = await Config.get_many( + 'rag.top_k', + 'rag.top_k_reranker', + 'rag.relevance_threshold', + 'rag.hybrid_bm25_weight', + 'rag.enable_hybrid_search', + 'rag.full_context', + # --- BEGIN EXTERNAL RETRIEVAL PATCH --- + 'rag.retrieval_engine', + 'rag.external_bypass_query_generation', + 'rag.external_message_count', + 'rag.external_user_messages_only', + # --- END EXTERNAL RETRIEVAL PATCH --- + ) + + # --- BEGIN EXTERNAL RETRIEVAL PATCH --- + # When the external retrieval service runs its own query generation, + # skip the LLM query-generation step and forward the (trimmed) raw + # conversation instead. queries stays empty -> falls through to the + # last-user-message fallback below. + bypass_external = rag_config.get('rag.retrieval_engine') == 'external' and rag_config.get( + 'rag.external_bypass_query_generation' + ) + + messages_for_external = ( + trim_messages_for_external( + body['messages'], + count=rag_config.get('rag.external_message_count'), + user_only=rag_config.get('rag.external_user_messages_only'), + ) + if bypass_external + else None + ) + # --- END EXTERNAL RETRIEVAL PATCH --- + queries = [] - if not all_full_context: + if not all_full_context and not bypass_external: try: queries_response = await generate_queries( request, @@ -1880,15 +1926,6 @@ async def chat_completion_files_handler( queries = [get_last_user_message(body['messages']) or ''] try: - # One batched SELECT instead of six sequential round trips. - rag_config = await Config.get_many( - 'rag.top_k', - 'rag.top_k_reranker', - 'rag.relevance_threshold', - 'rag.hybrid_bm25_weight', - 'rag.enable_hybrid_search', - 'rag.full_context', - ) # Directly await async get_sources_from_items (no thread needed - fully async now) sources = await get_sources_from_items( request=request, @@ -1909,6 +1946,9 @@ async def chat_completion_files_handler( hybrid_search=rag_config.get('rag.enable_hybrid_search'), full_context=all_full_context or rag_config.get('rag.full_context'), user=user, + # --- BEGIN EXTERNAL RETRIEVAL PATCH --- + messages=messages_for_external, + # --- END EXTERNAL RETRIEVAL PATCH --- ) except Exception as e: log.exception(e) @@ -2890,6 +2930,12 @@ async def tool_function(**kwargs): { **extra_params, '__event_emitter__': event_emitter, + # --- BEGIN EXTERNAL RETRIEVAL PATCH --- + # Forward the conversation so builtin knowledge tools + # (query_knowledge_files) can pass it to the external + # retrieval service for its own query generation. + '__messages__': form_data['messages'], + # --- END EXTERNAL RETRIEVAL PATCH --- '__skill_ids__': view_skill_ids, }, features, diff --git a/backend/open_webui/utils/tools.py b/backend/open_webui/utils/tools.py index 3f7d2aa094a6..7f331a4d0046 100644 --- a/backend/open_webui/utils/tools.py +++ b/backend/open_webui/utils/tools.py @@ -781,6 +781,13 @@ async def has_user_chat_permission(permission_key: str) -> bool: '__chat_id__': extra_params.get('__chat_id__'), '__message_id__': extra_params.get('__message_id__'), '__model_knowledge__': model_knowledge, + # --- BEGIN EXTERNAL RETRIEVAL PATCH --- + # Forwarded so query_knowledge_files can hand the conversation + # to the external retrieval service (it only reaches the tool + # because the tool declares __messages__ in its signature; see + # get_async_tool_function_and_apply_extra_params). + '__messages__': extra_params.get('__messages__', []), + # --- END EXTERNAL RETRIEVAL PATCH --- }, get_builtin_function_introspection(func), ) diff --git a/src/lib/apis/retrieval/index.ts b/src/lib/apis/retrieval/index.ts index fc5a7e827417..8acc0491fe8f 100644 --- a/src/lib/apis/retrieval/index.ts +++ b/src/lib/apis/retrieval/index.ts @@ -61,6 +61,21 @@ type RAGConfigForm = { web_loader_ssl_verification?: boolean; web?: Record; youtube?: YoutubeConfigForm; + // --- BEGIN EXTERNAL RETRIEVAL PATCH --- + RAG_RETRIEVAL_ENGINE?: string; + RAG_EXTERNAL_RETRIEVAL_URL?: string; + RAG_EXTERNAL_RETRIEVAL_API_KEY?: string; + RAG_EXTERNAL_RETRIEVAL_TIMEOUT?: string; + RAG_EXTERNAL_BYPASS_QUERY_GENERATION?: boolean; + RAG_EXTERNAL_MESSAGE_COUNT?: number; + RAG_EXTERNAL_USER_MESSAGES_ONLY?: boolean; + // --- END EXTERNAL RETRIEVAL PATCH --- + // --- BEGIN EXTERNAL INGESTION PATCH --- + EXTERNAL_INGESTION_ENGINE?: string; + EXTERNAL_INGESTION_URL?: string; + EXTERNAL_INGESTION_API_KEY?: string; + EXTERNAL_INGESTION_TIMEOUT?: string; + // --- END EXTERNAL INGESTION PATCH --- }; export const updateRAGConfig = async (token: string, payload: RAGConfigForm) => { diff --git a/src/lib/components/admin/Settings/Documents.svelte b/src/lib/components/admin/Settings/Documents.svelte index 3f2911ff36cd..ae37904a9375 100644 --- a/src/lib/components/admin/Settings/Documents.svelte +++ b/src/lib/components/admin/Settings/Documents.svelte @@ -395,7 +395,71 @@ {#if RAGConfig}
- + + + + + + + + + + {#if RAGConfig.EXTERNAL_INGESTION_ENGINE === 'external'} +
+ + + + + + +
+ + + + + {/if} + + + + + + + {#if RAGConfig.EXTERNAL_INGESTION_ENGINE !== 'external'} + {/if} + + {/if} + - {#if !RAGConfig.BYPASS_EMBEDDING_AND_RETRIEVAL} + + {#if !RAGConfig.BYPASS_EMBEDDING_AND_RETRIEVAL && RAGConfig.EXTERNAL_INGESTION_ENGINE !== 'external'} + + + {#if RAGConfig.EXTERNAL_INGESTION_ENGINE === 'external'} +
+ {$i18n.t( + 'When external ingestion is enabled, this model is used only to encode queries at retrieval time. It must match the embedding model configured in the ingestion service.' + )} +
+ {/if} + {#if !RAGConfig.RAG_FULL_CONTEXT} + + + + + + + + + {#if RAGConfig.RAG_RETRIEVAL_ENGINE === 'external'} +
+ + + + + + +
+ + + + + + + + + + {#if RAGConfig.RAG_EXTERNAL_BYPASS_QUERY_GENERATION} + + + + + + + + {/if} + {/if} + +