@mlflow/opencode

MLflow tracing plugin for OpenCode.

This plugin automatically traces OpenCode conversations to MLflow, capturing:

• User prompts and assistant responses
• LLM calls with token usage
• Tool invocations and results
• Session metadata

Installation

npm install @mlflow/opencode

Usage

1. Add to your opencode.json:

{
  "plugin": ["@mlflow/opencode"]
}

2. Set environment variables:

export MLFLOW_TRACKING_URI=http://localhost:5000
export MLFLOW_EXPERIMENT_ID=123

3. Run OpenCode normally - traces are created automatically when sessions become idle.

Configuration

The plugin is configured via environment variables:

| Variable | Required | Description | | ----------------------- | -------- | ---------------------------------------------------------- | | MLFLOW_TRACKING_URI | Yes | MLflow tracking server URI (e.g., http://localhost:5000) | | MLFLOW_EXPERIMENT_ID | Yes | MLflow experiment ID | | MLFLOW_OPENCODE_DEBUG | No | Set to true to enable debug logging |

Viewing Traces

Start an MLflow server and view your traces in the UI:

mlflow server
# Open http://localhost:5000

License

Apache-2.0