AI Gateway inference is returning HTTP 403 before provider routing for my Vercel Hobby workspace.
Verified configuration:
- Project: neverdie-os
- Team ID: team_RdP2Ass08g9fVQJ7ReMUWAuE
- Project ID: prj_jVO9eccGYmfI7uMOqwiVPQTzYdeh
- Production deployment: dpl_EeJUKvc35jHxxeAaDP2kUE4d6r8D
- Git commit: 2ffc93c8284f85a5b3a00933ff0d2ae1ae9d2fd3
- Model: openai/gpt-5.6-luna
- AI_ASSISTANT_PROVIDER is set to gateway
- AI_GATEWAY_API_KEY is server-only and marked Sensitive
- A newly generated API key was saved to Production and Preview
- The new key’s “Last used” timestamp confirms the deployment used it
- AI Credit balance is $5.00
- A valid payment method is configured
- Production deployment is READY
Observed result:
- HTTP status: 403
- Error class: GatewayInternalServerError
- Error type: internal_server_error
- Credential mode: server-key
- Zero input tokens
- Zero output tokens
- $0 spend
- The request fails before provider inference
Multiple freshly generated Gateway keys have produced the same result. The application safely falls back locally, but live inference remains unavailable.
Please verify whether AI Gateway inference entitlement or authorization is incorrectly disabled for this team/workspace.
No API keys or other credentials are included in this report.