AI phone agent for inbound and outbound calls

An AI phone agent is a voice agent that talks to people over the phone network: it answers a ringing number, speaks, listens, and hangs up. You supply the instructions, the LLM, and any tools (checking an order, booking a slot); Cartesia supplies the speech models, the turn-taking, and the phone integration. See Managed Agents for the agent side.

Yes. Assign a phone number to an agent in the Playground or through the Phone Numbers API, and inbound calls on that number route to the agent. A number serves one agent at a time, and an agent can hold several numbers, so a support line and a sales line can share one agent or split across two.

Yes. One API request names the agent, the number to call from, and one or more destinations in E.164 format. The single-call endpoint dials immediately and rejects destinations beyond your agent-call concurrency; for larger lists, batch calling queues up to 5,000 recipients and dials them as capacity frees up. For the outbound-sales shape of this, see AI cold calling.

Yours or ours. Cartesia-provisioned numbers are US-only and carry a telephony add-on ($0.014/min per the current phone numbers guide). Numbers imported from Twilio use your Twilio rates, and SIP trunk numbers use your carrier. Agent calling rates apply to all three options; check pricing for current terms.

With Cartesia-provisioned numbers, no: they are US numbers with US outbound. Imported Twilio numbers place international calls at your Twilio cost, and SIP trunking depends on your carrier. The speech side is separate from the dialing side; Sonic's language coverage is listed on the languages page.

You are responsible for complying with the regulations that apply to your calls, including the Telephone Consumer Protection Act (TCPA), and Cartesia's Acceptable Use Policy sets the rules of the road. Get consent before you call people, honor opt-outs, and check the rules for your industry and jurisdiction; this page is not legal advice.

Three costs stack: the agent's calling rate, the telephony path, and your LLM. The telephony add-on is $0.014/min and only applies to Cartesia-provisioned numbers; Twilio calls bill at your Twilio rates and SIP at your carrier's. Current rates live on the pricing page rather than here, and your LLM usage bills per the model's listed token prices (LLMs; Cartesia manages the integration, so there is no separate provider account).

No, not for a normal deployment: connect a number, assign the agent, and Cartesia bridges the call. You only need media code if you're bridging your own telephony provider, and that path is a single two-way WebSocket (Agents WebSocket).

The agent handles turn-taking and interruptions as part of the call, not as a feature you wire up yourself. That behavior comes from the same realtime stack described on Managed Agents, running Ink for listening and Sonic for speech.

An AI receptionist is one job an AI phone agent can do: answer inbound calls, screen them, and book or route. The setup differs by job, not by plumbing. Compare the inbound case in AI receptionist and how Cartesia compares with phone-first vendors in Cartesia vs Bland.