My voice agent stopped responding

We restarted the existing
agent and tested a single session, but it still produces no speech.

Plan: Build
Project ID: p_464r1yhgp42
Agent ID: CA_vsdrJXbdWb4q
Agent name: Ava
Affected session/room ID: RM_ysYVrHU8BTwV
Job ID: AJ_a869SJDAQLkb
Date/time: October 10, 2026, 21:19:59–21:20:17 UTC

The session connects successfully through WebRTC and receives
an audio track, but the audio is effectively silent.

STT repeatedly returns HTTP 429 Too Many Requests.
The same job also reports an LLM error:

type: inference_quota_exceeded
quota_type: llm
category: MaxGatewayCredits
remaining_limit: 0
current_usage: 0

The Billing page displays:
“Credit balances are temporarily unavailable.”

Based on the rates and quantities visible in the current month's
billing table, estimated inference usage is approximately $0.783.
This is only an estimate from the displayed rows, not a verified
account-wide balance.

The dashboard shows peak Nova-3 STT concurrency of 2/5.

Could you please check the account's shared inference allowance
and the effective quota enforced by the gateway?

Have we exhausted the aggregate credit allowance, or is there
an account/project restriction or quota configuration issue?

I have attached screenshots of Billing, inference limits,
and the STT errors.

Thank you.

Hi Joe, I’m a little confused what the root cause could be to be honest. It reminds me of a recent issue:

but that was fixed (plus we got lots of reports since lots of folks were impacted). It’s been a couple of days now since you started hitting quota limits (10th Oct) and if it was an issue, I would expect other folks to be reporting.

I don’t have full insight into billing, but as far as I can see, I thought you should be under your limits - quite possibly I’m missing something however. I have confirmed you have the $2.50 limit on your account correctly, and you only have a single project that could be drawing this down.

I think the concurrency might be a red herring, usually the root cause of this sort of issue is exceeding the Inference quota on build tier.