[BUG] Trouble with Ollama

[sodaken]

Is there an existing issue for this?

• I have searched the existing issues

Vercel Runtime Logs

• I have checked the Vercel Runtime Logs for errors (if applicable)

Current Behavior

Can't perform search with models loaded through Ollama.

Error Log:
Attempting to fetch models from: http://localhost:3000/config/models.json⁠

Successfully loaded models from URL

Stream execution error: [Error: Unable to connect. Is the computer able to access the url?] {

code: 'ConnectionRefused',

path: 'http://localhost:11434/api/chat',

errno: 0

}

Expected Behavior

Morphic runs search with the configured model

Steps To Reproduce

I'm using Searxng as search engine.

Ollama is running in the background and I've set up models.json as follows:

"id": "Qwen3:4b-instruct",
"name": "Qwen3:4b-instruct",
"provider": "Ollama",
"providerId": "ollama",
"enabled": true,
"toolCallType": "manual",
"toolCallModel": "phi4"

Environment

- OS: MacOS
- Browser: Safari

Anything else?

No response

[dan-and]

code: 'ConnectionRefused'

looks like morphic cannot reach ollama. Do you run morphic inside a container?

[miurla]

Hi @samuelgodoy,

This issue is likely caused by Docker networking — containers can't reach localhost on the host machine.

In v1.2.0, we've added Ollama support with a dynamic model selector. You no longer need to manually edit models.json. To use Ollama with Docker:

1. Set OLLAMA_BASE_URL=http://host.docker.internal:11434 in your .env.local
2. Run docker compose up -d
3. Select your Ollama model from the UI model selector

Note: Models must support tool calling to work with Morphic. The server validates this automatically on startup.

See the updated Docker guide for details.