Evalgent — voice agent test lab

Capacity

How the backend queues and runs live test calls: a global concurrency cap, a durable queue, and worker autoscaling. Changes apply to new calls immediately.

Live usage

Loading…

Cap & queue

Global in-flight cap across all workers.

Requests beyond this are rejected with 429.

A call queued longer than this is expired (failed).

Worker autoscaling
off

0 = scale-to-zero when idle (cold start on first call).

Hard ceiling on workers — your cost cap.

How many calls one worker handles at once.

Wait this long at low demand before scaling down.

Max replicas is your hard spend ceiling; min 0 lets workers scale to zero when idle. In dry-runthe controller computes and shows what it would do but doesn't touch Railway — the live railway scale call is wired at deploy.