Safe AI provider errors in Pro Prisma and Pro Drizzle
AI chat now explains provider quota, credential, model and rate-limit failures without exposing raw upstream details in the browser.
Both starter kits now use the AI SDK UI message protocol for chat responses. This lets the application send a useful failure message when OpenAI rejects a request after streaming has started.
Safe failure messages
The browser can distinguish provider quota, invalid credentials, unavailable models, provider rate limiting and an unknown temporary failure. The original provider response is logged only on the server and is never forwarded to the customer.
Application credit errors continue to use their existing purchase guidance. The client transport, route response and documentation were updated together so the protocol remains consistent.
Verification
The implementation was checked with the actual provider quota failure, focused unit coverage, authenticated browser tests and production builds in both kits.