We are using the Build plan, which advertises included inference credits.
Billing shows only modest usage (approximately 46k input tokens, 10k output tokens for GPT-4.1 and similar low usage across STT/TTS), and the next invoice is $0.
However, every new Agent session immediately fails with:
HTTP 429
inference_quota_exceeded
LLM token credit quota exhausted
The agent successfully joins the room, receives the configuration, and then fails on the first inference request.
Is there an account issue or another quota that is not reflected on the Billing page?
This was working fine earlier but just failed.
Yes Inference credits are a separate allowance from the invoice. Build includes $2.50 in credits (~50 min), and the Billing page shows billed usage, not credit burn, so a $0 next invoice is expected right up until the credits run out and the 429s start.
Your LLM spend isn’t what drained it: 46k input × $2.00/1M + 10k output × $8.00/1M on GPT-4.1 is about $0.17. TTS is the usual culprit at ElevenLabs rates ($300/1M chars) roughly 7,700 characters, about 13 minutes of speech, eats the entire $2.50. Deepgram Aura-2 is $30/1M, about 10× cheaper for the same minutes.
Do you also have other projects under your account? The collective usage of all projects under your build tier account will count towards the quota - for build, it is not per-project.
