Summary

• add a central model policy for self-hosted OpenAI defaults and their bare, gateway, LangChain, and display-label forms
• add pnpm models:sync to update managed references and pnpm models:check to detect drift without writing files
• add unit coverage and a lightweight GitHub Actions consistency check
• synchronize three existing outliers in the Vercel AI guide and self-hosted CLI template

Why

Model defaults are currently repeated across runnable examples, documentation, environment samples, and CLI templates. Updating the recommended model can leave some surfaces behind.

This introduces role-based synchronization rather than requiring every model in the repository to be identical. The initial policy covers self-hosted OpenAI examples and integration guides. OpenUI Cloud is explicitly excluded because its provider catalog and default-model policy are different.

Developer workflow

1. Update config/model-policy.json.
2. Run pnpm models:sync.
3. Commit the generated changes.
4. CI runs the equivalent of pnpm models:check and reports any files that are out of sync.

The script currently verifies 41 managed references and supports bare (gpt-5.5), gateway (openai/gpt-5.5), LangChain (openai:gpt-5.5), and display-label (GPT-5.5) forms.

Checks

• node --test scripts/models/model-policy.test.mjs
• pnpm models:check
• Prettier check for all changed files
• git diff --check