Rivesa R3 is ready for an API key
Add an AI Gateway key (or an Anthropic key) to the Vercel project and the agent switches to Claude on the next deploy. Nothing else to change.
Until now the deployment's own OIDC token always came first, so a gateway key added to the project would have been ignored. Now a key set on the project wins: AI_GATEWAY_API_KEY runs Claude through Vercel AI Gateway, and ANTHROPIC_API_KEY runs Claude directly with adaptive thinking. With neither, the agent keeps running on the deployment's token, as it does today.
When the team's gateway plan has no Claude, the agent falls back to GPT-5 mini and says so. That fallback now lasts 10 minutes instead of the life of the server, so credits added on the team bring Claude back without a redeploy.
GET /api/agent now says which credential is in use. npm run test:agent-provider checks every case offline, without calling any model.
- Improved
- A project AI Gateway key takes priority over the deployment token
- Free-tier fallback retries Claude every 10 minutes
- /api/agent reports the credential in use
- Shipped in
89c9013