pi-lmstudio

An extension for the Pi coding agent that integrates your local LLMs running in LM Studio directly into your AI-assisted coding workflow.

Prerequisites

1. Pi Coding Agent installed.
2. LM Studio installed and running.
3. LM Studio Local Server enabled:
   • Open LM Studio.
   • Go to the Local Server tab.
   • Turn on the switch.

Installation

pi install npm:pi-lmstudio

Configuration

By default, the extension connects to LM Studio at http://127.0.0.1:1234.

To customize the URL, create or modify ~/.pi/agent/lmstudio.json:

{ "url": "http://127.0.0.1:1234" }

You can also use environment variables (prefix with $):

{ "url": "$LM_STUDIO_URL" }

Multiple servers

To connect to several LM Studio instances at once (e.g. a desktop plus a remote GPU box), use the urls list instead. Each entry needs a name, and its models appear in the picker under lmstudio/<name>:

{
  "urls": [
    { "name": "desktop", "url": "http://127.0.0.1:1234" },
    { "name": "gpu-box", "url": "$GPU_BOX_URL" }
  ]
}

Notes:

• Unreachable servers are skipped and picked up automatically once they come online (refreshed each turn).
• If both url and urls are set, urls takes precedence and url is ignored.

Authentication

LM Studio can be configured to require a Bearer token for API access. When using the urls list form, add a token field to any entry that needs it. The token field accepts either a literal string or an environment variable reference (prefixed with $):

{
  "urls": [
    { "name": "desktop", "url": "http://127.0.0.1:1234" },
    { "name": "gpu-box", "url": "http://10.0.0.5:1234", "token": "my-secret-token" },
    { "name": "remote", "url": "http://10.0.0.6:1234", "token": "$LM_STUDIO_TOKEN" }
  ]
}

The token is sent as an Authorization: Bearer header for both model discovery and chat requests.

Usage

1. Launch Pi: Extensions are automatically loaded on startup.

   pi
   

2. Select a Model:

   • Use the /model command.
   • Or use Ctrl+P (Command Palette) and search for your model.
   • Look for models prefixed with lmstudio.

Notes

Context Length

• The actual usable context length is only reported after the model is fully loaded. The extension reloads the model info to retrieve the accurate value.

License

MIT