Is there Ollama integration?

[Wanchufley]

I was just wondering if there was a way to connect Ollama to opencode, I am trying to find a reference to it in the documentation, i can see that it has the support for example for llama models, but I can't seem to make it work while trying the config files I find, just getting: Error: agent coder not found

This is just for me to know if I should keep looking into this, I can also provide a my config file in case is useful.

Thanks

[harpomaxx]

hey,
same question here. Have you find a solution on that?

[Wanchufley]

So far I got to connect to my Ollama instance and make it see and choose between the models I have downloaded, but then when I try to prompt the following happens:

1. If you just prompt a question you might get the answer you may not.
2. It is not capable to see the code of the repo, even if the context percentage is at 100%, which you can see when you initialize the repo in Opencode for the first time. Or at least if he sees it it ignores it and doesn't also edit anything in the code base.
3. My Ollama instance is not the most performant ever but I get enough token/s to make it usable for me via openwebui, now, if we put Opencode in the formula, then you are talking waiting 5 minutes to complete a single prompt in which you don't know if you even going to get an answer.

The performance issue might not be from Opencode side but from mine, maybe as I am not using performant enough models, or I might miss something to really configure in the Opencode config as well. But for what is worth I will share it here so at least you can get it to work and then test with your own models.

Note that you should set the following environment variable in your terminal session before running Opencode:
export LOCAL_ENDPOINT=http://your-ollama-ip:11434/v1/

This is the my Opencode config:

  "data": {
    "directory": ".opencode"
  },
  "providers": {
  },
  "agents": {
    "coder": {
      "model": "local.qwen2.5-coder:3b",
      "reasoningEffort": "high",
      "maxTokens": 5000
    }
  },
  "shell": {
    "path": "/bin/bash",
    "args": ["-l"]
  },
  "debug": false,
  "autoCompact": true
}

I still need a lot more testing in my config as I need to understand how to add the different models for the different agents, maybe somebody in the comments can point me into the right direction.

[richardudovich]

same, doesn't work with local models