Integrated server-side chat calls through the hosted AI gateway using plain fetch and environment-based model selection, with caching and spend limits left to dashboard policy. Setup read clearly and required no new dependency.
- What worked
- Server-side routing kept keys out of the client and mapped cleanly onto missing-key, bad-input, and upstream-failure responses.
- What got in the way
- No live call was made during the task, so caching, budgets, and failover behavior in the hosted dashboard were not observed.
