pi-sync-models

A Pi package that discovers model IDs from OpenAI-compatible providers configured in <agent-dir>/models.json, applies intelligent context window & capability inference, upgrades existing metadata, and prunes deprecated models safely.

Install

pi install npm:pi-sync-models

If you already have ~/.pi/agent/extensions/sync-models.ts, remove/disable that standalone copy before installing, otherwise two copies of /sync-models will collide and both may run at startup. Restart Pi after installation.

Usage

• At startup, check providers once in the background (except when PI_OFFLINE=1).
• /sync-models checks all configured providers.
• /sync-models provider-id checks only that provider.
• Open /model after a successful sync to reload Pi's model list.

Only providers with api: "openai-responses" or "openai-completions" and baseUrl are queried at <baseUrl>/models. Keys and headers may use Pi-style literal, $ENV, ${ENV}, or !command values. This package does not read auth.json or discover built-in providers. Each request times out after 12 seconds. Providers that cannot be queried are left unchanged. The file's comments and unrelated configuration are preserved; writes are atomic and skipped if the file changed during the request.

Metadata & Capabilities:

• When discovering or syncing models, the package matches against Pi's built-in official model catalog (and heuristic rules) to assign accurate contextWindow, maxTokens, vision (image) modalities, and reasoning flags.
• Existing models with outdated or provisional metadata are upgraded.
• Models removed from the upstream endpoint are pruned safely (guarded against empty responses).
• Custom fields like cost on existing models are preserved.

The Pi agent directory defaults to ~/.pi/agent and honors PI_CODING_AGENT_DIR. The extension executes !command credentials from your own models.json in your local shell, as Pi does. Only install and run with trusted configuration.

Develop

npm install
npm run verify
pi -e .

License: MIT.