opencode-local

OpenCode plugin for local inference endpoint management. Port of pi-local.

Features

• Register local inference servers (oMLX, LM Studio, OpenAI-compatible)
• Discover and cache available models across connections
• Load/unload models via TUI dialogs with real-time state refresh
• Cached models appear in /model without network access
• macOS keychain storage for API keys (opt-in)

Configuration

Connections are stored in ~/.config/opencode/opencode-local-connections.json.

API keys support three formats:

• Direct key: sk-1234567890abcdef (stored in config or macOS keychain)
• Environment variable: $MY_API_KEY
• Command substitution: !security find-generic-password ...

Commands

/local-endpoints

Manage local inference endpoints:

• Add — prompts for base URL and optional API key. On macOS, offers to store direct keys in the system keychain.
• Remove — deletes a configured endpoint

/local-model

Browse and manage models from configured endpoints:

• List — shows available models with size, context window, type, and reasoning info
• Load model — → (right arrow) to load selected model
• Unload model — ← (left arrow) to unload selected model
• Filter — type to search models by name

Models are loaded/unloaded via the endpoint's API. After loading, select the model from /model to use it.

Installation

Via npm package:

opencode plugin -g @monroewilliams/opencode-local

Or add to ~/.config/opencode/tui.json and opencode.jsonc:

{ "plugins": ["@monroewilliams/opencode-local"] }

Limitations

• Active model selection requires /model after loading via /local-model (no plugin API to set the active model directly)
• Models in /model reflect the cache built at startup or last live query