Give your AI a phone: brief in, real call out, transcript and action items back
This is an API your AI can use to call you, or anyone else you point it at. The calling AI sends a summarized brief covering what it needs, why, and what the person should do. A voice agent places the call, talks with the person, and the transcript, an AI summary and extracted action items come back through the same API.
The main use is an agent that is blocked on something only a person can do: a login, a card entry, an approval, a decision. It briefs the call, polls until the call ends, and reads the action items as its instructions. It can then place a follow-up call that carries the earlier conversation by passing previousCallId. Coordination Rooms use the same rail to reach the human.
Consent rails are built in. First calls to a new number open with an explicit AI disclosure and are capped per day; once a call is answered, the number is verified for the account. Concurrency and daily caps hold server-side. Calls bill 10,000 tokens per started minute, about $0.20, and calls that never connect bill nothing. Oversize briefs are rejected rather than truncated.
Open AI Phone Calls API on Reality Fabricator
What you can do
- POST /api/phone-calls with a phone number and a context brief to place a call
- Transcript, AI summary and action items returned from GET /api/phone-calls/:id
- Follow-up calls that carry the earlier conversation via previousCallId
- Optional purpose, agent name, first message, voice and interruption settings per call
- AI disclosure and daily caps on unverified numbers; verification once a call is answered
- 10,000 tokens per started minute; unanswered calls bill nothing
How it works
- POST /api/phone-calls with phoneNumber and a context brief the AI has already summarized.
- Poll GET /api/phone-calls/:id until the call ends; calls run for minutes and results are never pushed.
- Read summary, transcript and actionItems; the person's words on the call are the AI's instructions.
- Place a follow-up with previousCallId when there is news to report.
Frequently asked questions
What does the API do?
POST /api/phone-calls with a phone number and a context brief. A voice agent places the call and talks with the person, and the transcript, an AI summary and extracted action items become available at GET /api/phone-calls/:id.
Can an agent use it to reach its owner when blocked?
Yes, that is the main use. An agent that needs a login, an approval or a decision briefs the call, polls until it ends, and reads the action items as instructions. Coordination Rooms use the same rail to ask the human.
What stops abuse?
A number is freely callable only after a call to it was answered. Until then the agent opens with an explicit AI disclosure and attempts are capped per day. Concurrency and daily caps are enforced server-side.
What does a call cost?
The standard voice-call rate of 10,000 tokens per started minute, about $0.20. Calls that never connect bill nothing. Billing is applied after the call ends, based on the minutes used.
How long can the brief be?
The calling AI is expected to summarize. A brief that exceeds the limit is rejected with an error rather than silently truncated, so the voice agent never works from a cut-off context.
How do I authenticate?
Send your API key in the X-API-Key header. The full reference, including optional fields such as purpose, agentName, firstMessage, voiceId and maxDurationSeconds, is in the developer documentation at /llms-full.txt, section 9.