We've rebuilt several monitoring views in the AIProxy dashboard to make production questions easier to answer: Which provider is failing? Which client is responsible for a spike? How many reasoning or cached tokens did a request consume? What happened during a Realtime session?

Charts you can explore

Live Charts now support timeframes from ten minutes through six months. You can drag to zoom into an incident, move between preset ranges, and filter the charts to one service or view the entire project. The charts cover incoming requests, DeviceCheck outcomes, rate-limited requests, and provider errors.

Top Client Usage now links into charts scoped to that client. From there, you can switch between the client's trends and its individual request log, with the same provider filter available in both places.

Request history with useful context

Request History is now organized as scannable cards and can be filtered by provider or to failed requests only. Each entry can show the HTTP method and path, status, client and IP, rewritten paths, request details, and relevant headers. Sensitive and noisy infrastructure headers are excluded from the display.

Usage reporting now separates input, output, cached input, and reasoning tokens. Audio requests can include duration, file size, and model information. Provider and AIProxy error bodies are shown independently, making it much faster to see which layer rejected a request.

WebSocket sessions have their own history

Realtime and other WebSocket traffic now has a dedicated dashboard view. Sessions include duration, bytes sent and received, client ID, path, IP address, and safe request headers. This gives long-lived connections the same kind of visibility that ordinary HTTP requests already had.

These updates are live in the developer dashboard now.