Per-call cost attributed to each client from the Vapi end-of-call breakdown. Unattributed = calls on a Vapi number not yet mapped in call routing.
Provider budgets and alarms live
Postpaid providers (Anthropic, Google, Deepgram, ElevenLabs) alarm on month-to-date spend and error rate; Vapi is prepaid and alarms on remaining credit. Thresholds live in cost_provider_budgets. Never a kill switch, a live call is never cut.
Cost by the job the AI is doing real
Names from vertical_personas (voice personas). Untagged is biggest until the AI stamps the job on each event. Job x channel is the dream cut, once a channel tag exists.
Budget spend real
used this month
·cap $50
Budget + alert thresholds live in Settings (client_financial_config). Alerts at 80% and 100%. Never a kill-switch, a live call is never cut.
Every cost event real
When
What
Ran on
Model
Persona
Cost
loading…
Cost to serve, per call real
When
Client
Channel
Top model
Parts
Cost
loading…
Platform overhead actual · monthly
infraFixed monthly platform costs the per-call cost rides on. Dev tooling is shown separately and is not cost-to-serve.