fix: make Ollama base URL configurable via OLLAMA_BASE_URL env var

[arnavgoel17]

Fixes Issue

Fixes #291

Description

The Ollama base URL was hardcoded as http://localhost:11434 across three files. When deployed (e.g. on Vercel), localhost resolves to the server's machine — not the user's — so local models are unreachable regardless of any server-side env var.

What this PR does

Two-layer fix:

1. Server-side default (env var) — OLLAMA_BASE_URL in Config sets the fallback for self-hosted / Docker / local-dev setups. Zero changes needed for existing deployments.

2. Per-request ollama_url param — every Ollama-touching endpoint now accepts a user-supplied URL that takes precedence over the server default. The frontend can pass a tunnel address (e.g. ngrok) so the backend proxies the request to the user's local Ollama — bypassing browser mixed-content and CORS restrictions entirely.

Endpoint	How to pass ollama_url
GET /models	query param ?ollama_url=
POST /model_info	JSON body field ollama_url
POST /chat	form field ollama_url
POST /chat/stream	form field ollama_url

If ollama_url is omitted, the server falls back to OLLAMA_BASE_URL (default http://localhost:11434), so local dev needs no changes.

Type of change

• 🐛 Bug fix

Checklist

• I am a ECSoC'26 contributor
• My code follows the project's style guidelines
• I have added comments in areas that may be hard to understand
• I have NOT included package.json or package-lock.json in this PR

Packages Added (if any)

None

How to test

Prerequisites: Ollama installed and at least one model pulled (e.g. ollama pull gemma3:1b).

Test 1 — Default behaviour (no env var, no ollama_url param)

1. Start the server without setting OLLAMA_BASE_URL.
2. Open the app, select a local model, send a message.
3. Response comes back normally — default http://localhost:11434 still works.

Test 2 — Custom URL via env var (self-hosted / Docker)

1. Set OLLAMA_BASE_URL to your Ollama instance address, start the server.
2. Send a message with a local model — same result expected.

Test 3 — Per-request ollama_url (deployed instance → user's local Ollama)

1. Expose your local Ollama via ngrok: ngrok http 11434
2. Call the models endpoint with the tunnel URL:

   GET /models?ollama_url=https://<your-ngrok-id>.ngrok-free.app
   

3. Your local models should appear in the response, even though the server is remote.
4. Pass the same ollama_url in a /chat or /chat/stream request — the server proxies through it and returns a reply.

Test 4 — Invalid URL gracefully errors

Set OLLAMA_BASE_URL (or pass ollama_url) to an unreachable address — confirm the app returns an error rather than crashing.

What changed (files)

File	Change
server/api/config.py	OLLAMA_BASE_URL via os.getenv() with localhost default
server/api/services/ollama_services.py	Both functions accept optional ollama_url param
server/api/routes/model_routes.py	/models and /model_info accept and forward ollama_url
server/api/routes/chat_routes.py	/chat and /chat/stream accept and forward ollama_url
server/.env.example	Updated docs explaining server default vs per-request param

[vercel]

@arnavgoel17 is attempting to deploy a commit to the Rucha Ambaliya's projects Team on Vercel.

A member of the Team first needs to authorize it.

[Rucha-Ambaliya]

Follow the pr template please

[arnavgoel17]

Follow the pr template please

Could you please elaborate a lil? I'm new to this repo.

[Rucha-Ambaliya]

https://github.com/Rucha-Ambaliya/PrivGPT-Studio?tab=contributing-ov-file#-mandatory-contribution-rules

[Rucha-Ambaliya]

Also please attach a video of testing, and/or a guide on how to test it

[arnavgoel17]

Heyy @Rucha-Ambaliya, I have edited the pr template as described.

[Rucha-Ambaliya]

Recording.2026-08-06.114033.mp4

local models not available at deployment

check #295 for reference

[Rucha-Ambaliya]

Still not working
Until you don't connect with user's machine directly, assigning direct ollama url most probably won't work