An LLM plugin for OpenCode Go subscription models.

Since the plugin is not yet on PyPI, install it directly from GitHub using uv tool install:

uv tool install llm --with "llm-opencode @ git+https://github.com/brendonmackenzie/llm-opencode.git"

If you already have llm installed with other plugins, add it to the existing install. For example:

uv tool install llm \
  --with "llm-opencode @ git+https://github.com/brendonmackenzie/llm-opencode.git" \
  --with llm-git --with llm-openrouter \ 
  --with llm-plugin-generator --with llm-uv-tool \
  --with toml --reinstall

To install from a local checkout during development:

uv tool install llm --with /path/to/llm-opencode --reinstall

Set your OpenCode Go API key:

llm keys set opencode

Paste your key when prompted. You can get an API key from the OpenCode console.

Alternatively, set the OPENCODE_KEY environment variable.

List available models:

llm opencode models

List all models registered with LLM:

llm models | grep opencode-go

Run a prompt with an OpenAI-protocol model:

llm -m opencode-go/deepseek-v4-flash "Explain quantum computing in one paragraph"

Run a prompt with an Anthropic-protocol model:

llm -m opencode-go/minimax-m3 "Write a haiku about programming"

Start an interactive chat:

llm chat -m opencode-go/glm-5.1

The model list is fetched dynamically from the OpenCode Go API. Run llm opencode models for the current list.

Available models broadly fall into two protocol groups:

Anthropic-protocol models (MiniMax, Qwen) support the following options:

• -o max_tokens N — Maximum number of tokens to generate (default: 4096)
• -o temperature F — Temperature for sampling (0.0–1.0)

Example:

llm -m opencode-go/minimax-m3 -o max_tokens 100 -o temperature 0.7 "Hello"

Set up a local development environment with uv:

cd llm-opencode
uv venv
uv pip install -e '.[test]'

Run tests with coverage:

uv run pytest --cov=.

Reinstall the plugin into the llm tool after changes:

uv tool install llm --with /path/to/llm-opencode --reinstall