Issues with Rag files for agents configured through modelSpecs in librechat.yml

[toxic frigate]

Hello!

I have configured a default agent through the modelSpecs: prop in my librechat.yml file

modelSpecs:
enforce: false
prioritize: true
addedEndpoints:
- agents
- Mistral
- anthropic
list:
- name:

But I can't find a way to give it access to a knowloedge database. It's not able to "see" the files uploaded in my account.
Through the UI when I upload files into an agent configuration tab it's all fine. So I guess that I'm missing something on how to tell the predefined model/agent what files and where they are.

Any idea?

Thanks!

[grim forge]

I looked into this — here’s what I found:
• The modelSpecs section in librechat.yml is only for defining models and agents (endpoint, name, prompts, etc.).
• It does not support pointing directly to RAG files or a knowledge base.
• RAG functionality is always handled via the RAG API and the files you upload through the UI. These get indexed into the vector database automatically.
• To make sure your agent can “see” those files, you need to configure the RAG API correctly in your .env (e.g. RAG_API_URL, EMBEDDINGS_PROVIDER, HF_TOKEN, or RAG_AZURE_OPENAI_API_VERSION if you’re using Azure).
• Once the RAG API is running and configured, uploaded files will be available to the agent without any extra YAML settings.

So in short:
• modelSpecs → just agent/model definitions.
• RAG → controlled by the RAG API + .env config + UI file uploads.

Hope that clears things up!