Every request has a timeline.

Ostrid runs agents on warm workers and records each step. Read the page left to right, the way a request travels.

T + 0 MS

Edge intake

Requests land on the nearest of three regions. TLS ends at the edge, so the first byte travels the shortest path.

T + 120 MS

The agent decides on its next tool. Plans are logged with their timing, so a slow decision is never a mystery.

T + 190 MS

Tool results stream back as they arrive. The model reads the first rows while the rest are still on the way.

T + 310–340 MS

The answer starts reaching your user. The whole path is on one timeline in the console.

ALSO ON EVERY RUN

BEFORE T + 0

A worker that already holds the model picks the run up. No cold start, no container boot, no wait for weights.

AFTER T + 340

Every step is stored with its inputs and timings. Replay any run from the last 30 days, one step at a time.

Start a run