-
-
Notifications
You must be signed in to change notification settings - Fork 5.3k
Description
What would you like to see?
Description:
Currently, reranking models in AnythingLLM seem to be coupled with LanceDB for vector storage and retrieval. I would like to request the ability to use reranking models with other vector databases supported by AnythingLLM (e.g., Pinecone, Weaviate, FAISS, or others).
Use Case:
This feature would benefit users who:
Prefer or already use vector DBs other than LanceDB for their workflows
Want to leverage the power of reranking models without switching their existing vector DB infrastructure
Need more flexibility to choose vector DBs based on cost, performance, or feature set, while still utilizing reranking for improved retrieval quality
Suggested Implementation:
Decouple reranking model logic from LanceDB-specific code
Provide a standardized interface or adapter layer so reranking can work regardless of the underlying vector DB
Ensure reranking modules accept query results from any supported vector DB and then perform the reranking step accordingly
Benefits:
Greater flexibility in deployment and integration
Improved user experience by combining best-of-breed vector DBs with reranking capabilities
Makes AnythingLLM more extensible and future-proof
Thank you for considering this enhancement!