Currently, the RAG SDK only supports hosted models. If we could enable the use of local models, similar to web-llm, that would be great. The only issue is that while they are OpenAI-compatible, they don’t provide an endpoint to query from. I believe we could write a simple HTTP server using bun.js and trigger it only if the user decides to use one of the local LLMs.
For example:
- Start the model.
- Hook it up to a simple server.
- Generate a predefined URL and provide it to our base LLM client. And, add it like any other model.
The rest should be straightforward.
Currently, the RAG SDK only supports hosted models. If we could enable the use of local models, similar to web-llm, that would be great. The only issue is that while they are OpenAI-compatible, they don’t provide an endpoint to query from. I believe we could write a simple HTTP server using bun.js and trigger it only if the user decides to use one of the local LLMs.
For example:
The rest should be straightforward.