Skip to content

Added support for external RAG ingestion and retrieval - #55

Open
cableman wants to merge 1 commit into
upstream/v0.11.0from
feature/external-ingest-rag-ii
Open

Added support for external RAG ingestion and retrieval#55
cableman wants to merge 1 commit into
upstream/v0.11.0from
feature/external-ingest-rag-ii

Conversation

@cableman

@cableman cableman commented Jul 28, 2026

Copy link
Copy Markdown

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.

See https://github.com/AarhusAI/ingestion-service and https://github.com/AarhusAI/retrieval-agent.

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.
@cableman
cableman marked this pull request as ready for review July 28, 2026 11:56
@cableman cableman mentioned this pull request Jul 28, 2026
@cableman
cableman requested a review from jekuaitk July 28, 2026 12:01
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Projects

None yet

Development

Successfully merging this pull request may close these issues.

2 participants