Sync Pi models

Keep Pi's models.json in sync with the model aliases exposed by an OpenAI-compatible server such as llama-swap.

Run /models-sync when your server gains models. It adds only model IDs Pi does not already know about, so your existing context limits, vision settings, reasoning settings, sampling parameters, names, and other tuning stay intact.

Use once

You do not need to install anything permanently. Run:

pi -e npm:pi-models-sync-extension

Then run this inside Pi:

/models-sync

If models.json has several providers, Pi shows a picker. To target one directly:

/models-sync llama-swap

-e loads the extension for that Pi process only. It does not install the package or change your Pi settings.

Install it permanently (optional)

pi install npm:pi-models-sync-extension

Start Pi normally afterwards and use /models-sync whenever needed.

Set up your provider

The provider must already be configured in Pi's models.json with a baseUrl and api. For example, ~/.pi/agent/models.json:

{
  "providers": {
    "llama-swap": {
      "baseUrl": "http://192.168.1.222:1080/v1",
      "api": "openai-completions",
      "apiKey": "local",
      "models": []
    }
  }
}

Your server must expose an OpenAI-style model endpoint at <baseUrl>/models, returning a response like:

{
  "data": [
    { "id": "qwen3.6", "name": "Qwen3.6-27B" }
  ]
}

If the server needs authentication, configure its API key and headers normally in models.json. The extension uses those credentials for the catalog request and never displays them.

What happens when you sync

1. Choose a provider, if necessary.
2. Pi fetches its model catalog.
3. Pi shows model IDs it can add and asks for confirmation.
4. Confirm to add them to that provider's models list.

The extension never removes models, never overwrites existing model entries, and makes no change if the request fails, the server response is invalid, or you cancel. New models contain only the server's id and optional display name; add model-specific limits and capabilities yourself afterwards.

New models are available in /model immediately after a successful sync.

Troubleshooting

• Provider is not listed: add it under providers in Pi's models.json first.
• Provider has no baseUrl or api: add an OpenAI-compatible base URL (normally ending in /v1) and an API such as openai-completions.
• Request fails or response is invalid: confirm that curl <baseUrl>/models returns an object with a data array.
• A model needs vision, reasoning, or custom limits: edit that model's entry in models.json after it is added; later syncs preserve those fields.