Skip to content
llm-review-panel

OpenCode + Ollama

Auth: none (local). Cost: free. Output: plain text.

Install

Terminal window
# 1. install Ollama and pull a model
curl https://ollama.ai/install.sh | sh
ollama pull qwen2.5-coder:14b


# 2. install OpenCode (see https://opencode.ai for current instructions)

Confirm with:

Terminal window
opencode run -m ollama/qwen2.5-coder:14b "hello"

Config entry

{
  "id": "opencode-ollama",
  "enabled": true,
  "command": "opencode",
  "args": ["run", "-m", "{model}", "{prompt}"],
  "prompt_via": "arg",
  "model": "ollama/qwen2.5-coder:14b",
  "result_path": null,
  "timeout": 600,
  "paid": false
}

Notes

  • result_path: null because OpenCode prints the answer as plain text.
  • Longer timeout because local inference is slower than hosted.
  • Pick whatever Ollama-served model you prefer. Larger code-tuned models give better reviews; 14B is a reasonable starting point.
  • This is often the most useful reviewer to keep on the panel even when you have hosted ones, because it brings a genuinely different model family.