opencode-plugin-litellm

opencode-litellm

OpenCode plugin for LiteLLM proxy support with auto-detection and dynamic model discovery.

Inspired by opencode-lmstudio.

npm package name: opencode-plugin-litellm (the unscoped opencode-litellm name was taken on npm). Repo name and plugin export remain opencode-litellm.

Features

• Auto-detection — finds a LiteLLM proxy running on common ports (4000, 8000, 8080)
• Dynamic model discovery — queries /v1/models so you never hand-maintain a model list
• OpenAI-compatible — registers under OpenCode as @ai-sdk/openai-compatible
• Vendor-aware — extracts litellm_provider (or the provider/model prefix) into organizationOwner
• Smart formatting — turns anthropic/claude-3-5-sonnet into Claude 3 5 Sonnet in the picker
• Categorization — chat / embedding / image / audio modalities are inferred from the model mode or id
• Non-blocking — discovery is capped at 5s so a slow proxy never blocks OpenCode startup

Installation

npm install opencode-plugin-litellm
# or
bun add opencode-plugin-litellm

Usage

Zero-config (recommended)

If a LiteLLM proxy is already running on localhost:4000, just add the plugin:

{
  "$schema": "https://opencode.ai/config.json",
  "plugin": ["opencode-plugin-litellm@latest"]
}

The plugin will detect the proxy, query /v1/models, and inject every model into your provider list automatically.

Manual configuration

{
  "$schema": "https://opencode.ai/config.json",
  "plugin": ["opencode-plugin-litellm@latest"],
  "provider": {
    "litellm": {
      "npm": "@ai-sdk/openai-compatible",
      "name": "LiteLLM (proxy)",
      "options": {
        "baseURL": "http://localhost:4000/v1",
        "apiKey": "{env:LITELLM_API_KEY}"
      }
    }
  }
}

Any models you predefine under provider.litellm.models are kept as-is; the plugin only adds models it discovers that aren't already configured.

Authentication

If your LiteLLM proxy requires a master key, expose it via either:

• apiKey field inside provider.litellm.options, or
• the LITELLM_API_KEY (or LITELLM_MASTER_KEY) environment variable.

How it works

• On OpenCode startup the config hook fires.
• If a litellm provider is configured, its baseURL is used. Otherwise the plugin probes localhost:4000, :8000, :8080.
• Health check via GET /v1/models. If unreachable, the hook is a no-op.
• Models from /v1/models are turned into OpenCode model entries with sensible name, organizationOwner, and modalities fields.
• Discovered models are merged on top of any user-configured ones.

Requirements

• OpenCode with plugin support (@opencode-ai/plugin ^1.0.166)
• A running LiteLLM proxy (pip install 'litellm[proxy]' && litellm --config config.yaml)

License

MIT