@danmademe/pi-provider-litellm
Pi agent extension for LiteLLM proxy auto-discovery and model configuration
Package details
Install @danmademe/pi-provider-litellm from npm and Pi will load the resources declared by the package manifest.
$ pi install npm:@danmademe/pi-provider-litellm- Package
@danmademe/pi-provider-litellm- Version
0.2.0- Published
- May 29, 2026
- Downloads
- not available
- Author
- danmademe
- License
- MIT
- Types
- extension
- Size
- 29.9 KB
- Dependencies
- 0 dependencies · 2 peers
Pi manifest JSON
{
"extensions": [
"./src/index.ts"
]
}Security note
Pi packages can execute code and influence agent behavior. Review the source before installing third-party packages.
README
pi-provider-litellm
[!NOTE] This extension requires pi and a running LiteLLM proxy.
Pi agent extension that auto-discovers and registers models, MCP tools, and skills from a LiteLLM proxy.
Features
- Model discovery — queries
/healthand/model/infoendpoints to discover available models, registers them under thelitellmprovider with cost, context window, and capability metadata - MCP tools — discovers tools from LiteLLM's MCP REST gateway (
/mcp-rest/tools/list) and registers them as pi tools with JSON schema to TypeBox mapping - Skills management —
skill_list,skill_use,skill_register,skill_enable, andskill_disabletools for managing LiteLLM skills - Skills injection — injects enabled skills as structured context into the agent's system prompt on session start, with 60-second caching
All discovery runs in parallel with a 30s timeout. Each step fails gracefully — a failed model discovery won't block MCP tools or skills.
Installation
# From npm (once published)
pi install npm:@danmademe/pi-provider-litellm
# From local path
pi install npm:@danmademe/pi-provider-litellm
Configuration
Provide your LiteLLM proxy URL and API key via environment variables or settings.
Environment variables (recommended)
export LITELLM_URL="https://your-litellm-proxy.example.com"
export LITELLM_KEY="sk-your-api-key"
Optional: override the provider name (defaults to litellm):
export LITELLM_PROVIDER_ID="my-proxy"
Settings file
Add to ~/.pi/agent/settings.json:
{
"pi-provider-litellm": {
"url": "https://your-litellm-proxy.example.com",
"token": "sk-your-api-key"
}
}
Optional: override the provider name:
{
"pi-provider-litellm": {
"url": "https://your-litellm-proxy.example.com",
"token": "sk-your-api-key",
"providerId": "my-proxy"
}
}
Environment variables take precedence over the settings file.
Enabling models
Models registered by this extension use the configured provider ID (default: litellm). To include them in your model cycling, add them to enabledModels in settings.json:
{
"enabledModels": [
"litellm/qwen/qwen3.6-27b",
"litellm/anthropic/claude-sonnet"
]
}
[!TIP] The model ID in pi is
<providerId>/<model_name>, where<providerId>defaults tolitellmand<model_name>is the ID as reported by your LiteLLM proxy (e.g.,qwen/qwen3.6-27b,anthropic/claude-sonnet).
Tools
Once loaded, the extension registers the following tools:
| Tool | Description |
|---|---|
skill_list |
List all available skills on the proxy |
skill_use |
Fetch full content of a skill by name |
skill_register |
Register a new skill from a git repository |
skill_enable |
Enable a skill by name |
skill_disable |
Disable a skill by name |
mcp_<server>_<tool> |
Auto-discovered MCP tools from the proxy |
Development
npm install
npm run typecheck
npm run test:run
License
MIT