[GH-ISSUE #3394] [FEAT]: Adding custom chunks for RAG? #2189

Closed
opened 2026-02-22 18:28:32 -05:00 by yindo · 2 comments
Owner

Originally created by @hamiz-ahmed on GitHub (Mar 4, 2025).
Original GitHub issue: https://github.com/Mintplex-Labs/anything-llm/issues/3394

What would you like to see?

Is it possible that users can upload custom chunks for rags. This would mean that the chunks are generated independently for RAG, outside anythingLLM and then later integrated in to the application.
The problem right now is that current chunking in Anything LLM doesn't display the metadata in the citations. If somehow, each chunk could have the metadata embedded, and also displayed in citations, it would be absolutely great.

Originally created by @hamiz-ahmed on GitHub (Mar 4, 2025). Original GitHub issue: https://github.com/Mintplex-Labs/anything-llm/issues/3394 ### What would you like to see? Is it possible that users can upload custom chunks for rags. This would mean that the chunks are generated independently for RAG, outside anythingLLM and then later integrated in to the application. The problem right now is that current chunking in Anything LLM doesn't display the metadata in the citations. If somehow, each chunk could have the metadata embedded, and also displayed in citations, it would be absolutely great.
yindo added the enhancementfeature request labels 2026-02-22 18:28:32 -05:00
yindo closed this issue 2026-02-22 18:28:32 -05:00
Author
Owner

@timothycarambat commented on GitHub (Mar 4, 2025):

Couldnt each chunk just be a document? You could also do this by pushing data into the workspace in the vector DB manually. This is possible in practice by crafting these "documents" with metadata via the API /v1/document/raw-text endpoint. Otherwise, I think its unlikely we allow random chunking upserts in the UI since most people wont be utilizing that & we tracking and managing those vectors will be quite tough or costly to track in DB

@timothycarambat commented on GitHub (Mar 4, 2025): Couldnt each chunk just be a document? You could also do this by pushing data into the workspace in the vector DB manually. This is possible in practice by crafting these "documents" with metadata via the API `/v1/document/raw-text` endpoint. Otherwise, I think its unlikely we allow random chunking upserts in the UI since most people wont be utilizing that & we tracking and managing those vectors will be quite tough or costly to track in DB
Author
Owner

@hamiz-ahmed commented on GitHub (Mar 5, 2025):

would the raw_text endpoint create embedding for the metadata of the document also? During RAG retrieval, would this metadata be provided to LLM as context if the api endpoint is used?
Each chunk as a document would be a really ugly solution in the end. Imagine a book containing 300 pages and each document can only consist of 1000 characters. This would ultimatelty result in a lot of documents.

@hamiz-ahmed commented on GitHub (Mar 5, 2025): would the raw_text endpoint create embedding for the metadata of the document also? During RAG retrieval, would this metadata be provided to LLM as context if the api endpoint is used? Each chunk as a document would be a really ugly solution in the end. Imagine a book containing 300 pages and each document can only consist of 1000 characters. This would ultimatelty result in a lot of documents.
yindo changed title from [FEAT]: Adding custom chunks for RAG? to [GH-ISSUE #3394] [FEAT]: Adding custom chunks for RAG? 2026-06-05 14:45:05 -04:00
Sign in to join this conversation.
1 Participants
Notifications
Due Date
No due date set.
Dependencies

No dependencies set.

Reference: Mintplex-Labs/anything-llm#2189