Retrieval requires very large dependencies

[voidwisp]

I've attempted to create an MCP server using AWS Lambda with graphrag libraries. I've used the Lambda layers instead of docker containers for lower cold-start times. However the dependencies required for graphrag are extremely large that they cant even fit into AWS Lambda layers and hit limits. Spacy alone is 500 MB and llama-index-core brings a large amount of dependencies. It would be great to be able to make the retrieval a lot more light-weight for environments like AWS Lambda.

[iansrobinson]

Hi @voidwisp

Thanks for the comment

The size of the install has been a long-standing issue. In large part this is because of the LlamaIndex dependencies. When we started the project we thought we'd get more lift from LlamaIndex than has in fact been the case. Unfortunately, LlamaIndex is prevalent throughout both the indexing and retrieval modules - removing it looks like it would require a significant engineering effort.

No specific plans on the way forwards yet, but we are aware of the issue – it's on our list of things to address as we plan work for this year.

Kind regards

ian