How to configure LibreChat to discuss with secured self hosted Ollama?

[SansGuidon]

correct. I also tested that I can reach the Ollama api hosted on Cloudron using that same Bearer
curl -v https://ollama-api.<REDACTED>/v1/models -H "Authorization: Bearer <REDACTED>"
it does not return anything more than {"object":"list","data":null} useful but at least it's a 200 and a good enough test for me.

[james]

Hello @SansGuidon

Your API request returned {"object":"list","data":null} which indicated no models found.

For testing, in the web terminal of the Ollama app I send this command:

ollama run tinyllama

After that is done, now I get:

curl https://ollama-api.cloudron.dev/v1/models -H "Authorization: Bearer eyJhbGciOiJIUzI1NiIsInR5cCI6IkpXVCJ9.eyJzdWIiOiJhZG1pbiIsImlhdCI6MTc2MTgxNjM1OSwiZXhwIjoxNzYxODE5OTU5fQ.YLZtAuIjqnApthTBfuoPyyyJ5a7N2wywn2GW9dTqUeU"
{"object":"list","data":[{"id":"tinyllama:latest","object":"model","created":1761817528,"owned_by":"library"}]}

After reproducing your setup I also get the HTTP error 401.
I suspect that the apiKey: config is wrong since we use bearer auth.
And yes, that was it.

Here is my working config for LibreChat:

version: 1.2.8

endpoints:
  custom:
    - name: "Ollama"
      baseURL: "https://ollama-api.cloudron.dev/v1/" 
      models:
        default: [
          "tinyllama:latest"
          ]
        fetch: true
      titleConvo: true
      titleModel: "current_model"
      summarize: false
      summaryModel: "current_model"
      forcePrompt: false
      modelDisplayLabel: "Ollama"
      headers:
        Authorization: "Bearer ${OLLAMA_BEARER_TOKEN}"

Always happy to help.

[james]

Oh, wait no. Missing the apiKey config it defaulted back to my configured GPT.
Sorry, I will look into it.

[james]

The LibreChat logs state on restart:

Oct 30 11:23:19 2025-10-30 10:23:19 error: [indexSync] error fetch failed

But in the Ollama logs for this timestamp I can see no request.
But when trying to send a prompt, I get a request:

Oct 30 11:24:15 172.18.0.1 - - [30/Oct/2025:10:24:15 +0000] "GET /api/tags HTTP/1.1" 401 188 "-" "axios/1.12.1"
Oct 30 11:24:17 172.18.0.1 - - [30/Oct/2025:10:24:17 +0000] "POST /api/chat HTTP/1.1" 401 188 "-" "ollama-js/0.5.14 (x64 linux Node.js/v22.14.0)"

So, something is astray here.

[SansGuidon]

Thanks @james

[james]

I was able to capture the headers that are send to Ollama.

// request made when re-trying the prompt ⚠️the auth bearer header is missing
POST /mirror HTTP/1.1
X-Original-URI: /api/chat
Host: 127.0.0.1:6677
Connection: close
Content-Length: 114
Content-Type: application/json
Accept: application/json
User-Agent: ollama-js/0.5.14 (x64 linux Node.js/v22.14.0)
accept-language: *
sec-fetch-mode: cors
accept-encoding: br, gzip, deflate

// manual curl reuqest - header is there
GET /mirror HTTP/1.1
X-Original-URI: /v1/models
Host: 127.0.0.1:6677
Connection: close
user-agent: curl/8.5.0
accept: */*
authorization: Bearer eyJhbGciOiJIUzI1NiIsInR5cCI6IkpXVCJ9.eyJzdWIiOiJhZG1pbiIsImlhdCI6MTc2MTgxNjM1OSwiZXhwIjoxNzYxODE5OTU5fQ.YLZtAuIjqnApthTBfuoPyyyJ5a7N2wywn2GW9dTqUeU

So for some reason the header is not send.

[SansGuidon]

Thanks @james for confirming my suspicion. But Ollama is accepting the Header so I guess the problem is on LibreChat, right?

[james]

Hello @SansGuidon
Yes, that is what we assume now as well.
We have created a bug report on GitHub for this.

[nebulon]

Just to keep everyone updated, the github issue in question is https://github.com/danny-avila/LibreChat/issues/10311

[james]

There was just a comment.

Thanks, this is specific to Ollama as it still uses some legacy code where headers are not expected.

You can already work around it by changing the custom endpoint name from "ollama" to something else, but I'm pushing a simple fix.

Tested this, config:

version: 1.2.8

endpoints:
  custom:
    - name: "CloudronOllama"
      apiKey: "ollama"
      baseURL: "https://ollama-api.cloudron.dev/v1/"
      models:
        default: ["tinyllama:latest"]
        fetch: true
      titleConvo: true
      titleModel: "current_model"
      summarize: false
      summaryModel: "current_model"
      forcePrompt: false
      modelDisplayLabel: "Ollama"
      headers:
        Content-Type: "application/json"
        Authorization: "Bearer eyJhbGciOiJIUzI1NiIsInR5cCI6IkpXVCJ9.eyJzdWIiOiJhZG1pbiIsImlhdCI6MTc2MTgxNjM1OSwiZXhwIjoxNzYxODE5OTU5fQ.YLZtAuIjqnApthTBfuoPyyyJ5a7N2wywn2GW9dTqUeU"

And indeed, this worked! But danny-avila already created a PR to fix this issue.
So, to get it working right now, just change - name: "Ollama" to e.g. - name: "CloudronOllama"

[SansGuidon]

That's working too for me by changing the name ! Thanks @james