OpenCode provider for Pi Agent. Auto-discovers models from OpenCode registry and uses Pi's native OpenAI-compatible handler.

pi install /path/to/pi-opencode-bridge

Then restart Pi.

/opencode-go-key <your-api-key>   # Set OpenCode Go API key directly
/opencode-go-key                   # Interactive prompt if no key provided
/opencode-status                   # Check authentication status

Keys are saved to two locations:

• ~/.local/share/opencode/auth.json — shared with OpenCode CLI
• ~/.pi/agent/auth.json — Pi's AuthStorage format

After setup, select a model via /model. Models are split into two providers:

Go plan — OpenAI-compatible (oc-sdk-go/):

• oc-sdk-go/deepseek-v4-pro
• oc-sdk-go/deepseek-v4-flash
• oc-sdk-go/kimi-k2.6
• oc-sdk-go/mimo-v2.5-pro
• oc-sdk-go/glm-5.1

Go plan — Anthropic format (oc-sdk-go-anthropic/):

• oc-sdk-go-anthropic/qwen3.7-max
• oc-sdk-go-anthropic/qwen3.7-plus
• oc-sdk-go-anthropic/qwen3.6-plus
• oc-sdk-go-anthropic/qwen3.5-plus
• oc-sdk-go-anthropic/minimax-m3
• oc-sdk-go-anthropic/minimax-m2.7
• oc-sdk-go-anthropic/minimax-m2.5

Zen plan (oc-sdk-zen/ and oc-sdk-zen-anthropic/):

• Models auto-split by API format (OpenAI-compatible vs Anthropic)

1. At startup, reads ~/.cache/opencode/models.json (OpenCode's model registry)
2. Models registered in Pi with context windows, costs, and capabilities
3. Pi's native openai-completions handler bridges requests to OpenCode's API
4. Message conversion, tool calls, thinking/reasoning, and SSE streaming handled by Pi

Pi Agent
  ├─ oc-sdk-go            → https://opencode.ai/zen/go/v1 (OpenAI-compatible)
  ├─ oc-sdk-go-anthropic  → https://opencode.ai/zen/go   (Anthropic messages)
  ├─ oc-sdk-zen           → https://opencode.ai/zen/v1   (OpenAI-compatible)
  └─ oc-sdk-zen-anthropic → https://opencode.ai/zen      (Anthropic messages)

• /opencode-connect — CLI-based login via opencode auth login
• /opencode-zen-key — Set Zen plan API key directly