RAG API don't use openai /embeddings

[pasengier-smartwork]

What happened?

I'm trying to get RAG_API working with llamacpp server wich offer a openai compatible api.

Parameters are :
RAG_API_URL=http://127.0.0.1:8080
EMBEDDINGS_PROVIDER=openai

The local API is called but with URL /embed and not data in the body.

logs from api :
srv log_server_r: done request: POST /embed 127.0.0.1 404
srv log_server_r: request:
srv log_server_r: response: {"error":{"message":"File Not Found","type":"not_found_error","code":404}}

I tried rewriting the url with nginx, but the body is still empty.

embedding tested with curl :
curl --request POST --url http://localhost:8080/embeddings --header "Content-Type: application/json" --data '{"input": "Hello embeddings"}' --silent
[{"index":0,"embedding":[[0.06244239956140518,0.03484318032860756,-0.016207346692681313,0.04565286263823509,-0.014398101717233658,-0.017415624111890793,-0.0071379863657057285,0.004529181402176619,0.012829050421714783,0.01....

Thanks for LibreChat !!

Version Information

7cf8b84

Steps to Reproduce

1. Use this parameters :
   RAG_API_URL=http://127.0.0.1:8080
   EMBEDDINGS_PROVIDER=openai

2. try uploading a file

What browsers are you seeing the problem on?

Firefox

Relevant log output

From Llama cpp 

srv  log_server_r: done request: POST /embed 127.0.0.1 404
srv  log_server_r: request:  
srv  log_server_r: response: {"error":{"message":"File Not Found","type":"not_found_error","code":404}}

From Reverse proxy : 
[23/Apr/2026:12:56:32 +0000] "POST /api/files HTTP/1.1" 500 35 "https://xxxxx/c/e1343516-091a-4f77-8f84-376ae4a95343" "Mozilla/5.0 (X11; Linux x86_64; rv:148.0) Gecko/20100101 Firefox/148.0" "-"

Screenshots

No response

Code of Conduct

• I agree to follow this project's Code of Conduct

[danny-avila]

You've mixed up two different URLs.

RAG_API_URL is LibreChat's setting telling it where to find the RAG API service (this repo), not the embeddings backend. By pointing it at your llama.cpp server, LibreChat is POSTing the RAG API's /embed route to llama.cpp, which has no such route, hence the 404.

To use llama.cpp as the embeddings backend, configure it on the RAG API side:

# LibreChat -> RAG API service
RAG_API_URL=http://127.0.0.1:8000   # wherever rag_api itself listens

# RAG API -> llama.cpp (OpenAI-compatible)
EMBEDDINGS_PROVIDER=openai
RAG_OPENAI_BASEURL=http://127.0.0.1:8080/v1
RAG_OPENAI_API_KEY=sk-anything
EMBEDDINGS_MODEL=<model-loaded-in-llamacpp>

RAG_OPENAI_BASEURL is passed to langchain's OpenAIEmbeddings(openai_api_base=...), which will then hit {baseurl}/embeddings on llama.cpp (the same endpoint your curl test confirmed works).

[pasengier-smartwork]

Look like i'm missing something, I used the setup you send me and here is log of librechat :

npm[109013]: 2026-04-23 21:49:13 warn: RAG API is either not running or not reachable at http://127.0.0.1:8000, you may experience errors with file uploads.

and no surprise :

npm[109013]: 2026-04-23 21:52:08 error: Error uploading vectors No response received for POST http://127.0.0.1:8000/embed: connect ECONNREFUSED 127.0.0.1:8000

What is supposed to listen on port 8000 ?

[JulianHandrup]

What is supposed to listen on port 8000 ?

Port 8000 is the default port for the RAG-API to listen on, as you can see in the rag_api docs:
https://github.com/danny-avila/rag_api/blob/main/README.md?plain=1#L85

You can override it in your .env-file if you need to.

Check for issues in the rag-api container logs if it has trouble starting.
Because your log clearly says that LibreChat cannot reach it, which is usually a strong indicator that it didn't start at all.

Of course that's all under the assumption that you're using a docker setup.