A Sautikit AI receptionist answers every call on your business number, greets the caller, answers common questions, takes a name and a message, and passes the call to your team when a person is needed. It costs KES 4.00 a minute, metered per second, and incoming calls are free.
It suits law firms, property agencies, clinics, schools and any office that loses business when the phone rings out at lunch or after 5pm.
The caller dials your number and the agent answers with your greeting.
The caller says what they want. The agent answers questions about your hours, location, services and prices from what you wrote in its prompt.
For anything else, the agent takes the caller's name, number and reason for calling.
If the caller asks for a person, or the matter is outside what you told the agent to handle, it hands the call to your team with a summary of the conversation so far.
Callers can interrupt the agent mid-sentence and it stops to listen.
The base_prompt is the receptionist's job description. GET /v1/models lists the models you can choose.
const api = "https://api.sautikit.com/v1";const headers = { Authorization: `Bearer ${process.env.SAUTIKIT_API_KEY}`, "Content-Type": "application/json",};const agent = await fetch(`${api}/agents`, { method: "POST", headers, body: JSON.stringify({ name: "Front desk", base_prompt: "You are the receptionist for a law firm in Nairobi. Greet callers, answer " + "questions about office hours (Monday to Friday, 8:00 to 17:30), location and " + "services, and take a name and callback number for anything that needs an " + "advocate. Hand over to a person whenever the caller asks for one.", model: "sautikit-flash", max_call_minutes: 10, }),}).then((r) => r.json());
mode: "always" puts the agent on every call. To have your team answer during the day and the agent cover nights and weekends, use after_hours with your working hours:
When the agent hands over, Sautikit POSTs a request to your handover_callback_url with the reason and a summary:
{ "sessionId": "HD_a1b2c3d4-e5f6-7890-abcd-ef1234567890", "callerNumber": "+254720000000", "status": "CallHandover", "handover": { "agent_id": "9d2b1f53-8c0e-4f1d-9a6b-5d3a8c47e9f0", "agent_name": "Front desk", "reason": "caller asked to speak to a person", "summary": "Caller wants to book a consultation about a land matter. Available Thursday afternoon. Callback number confirmed." }}
Reply with voice actions, usually a dial to your team:
{ "actions": [ { "say": { "text": "Connecting you to the front desk now." } }, { "dial": { "number": "+254720000001", "timeout": 30 } } ]}
Whoever picks up already knows who is calling and why.
Use your systems. Attach HTTP tools and the agent can check a booking or look up a client during the call. Requests are signed so your server can verify them.
Pass the caller to another agent. A front-desk agent can move a billing question to a billing agent on the same call.
End the call when the caller is done.
Stay within a limit.max_call_minutes caps every call, 15 minutes by default.
Fall back safely. If no AI seat is free, the call follows the number's normal routing and the caller never hears an error.
Every call is recorded, and GET /v1/calls/{id}?transcript=1 returns the transcript: who said what, and which tools the agent used. Read the first week's calls, adjust the prompt and publish again.
If you want your own model to run the conversation, return a stream action from your number's voice callback. Sautikit sends the live call audio to your WebSocket and plays back the audio you send.
Your WebSocket server must accept the audio.drachtio.org subprotocol:
import { WebSocketServer } from "ws";import { connectToModel } from "./model.js"; // your own speech model wrapperconst wss = new WebSocketServer({ port: 8080, handleProtocols: () => "audio.drachtio.org",});wss.on("connection", async (call) => { const model = await connectToModel({ sampleRate: 16000 }); // Caller audio (16-bit PCM) to your model call.on("message", (frame, isBinary) => { if (isBinary) model.sendAudio(frame); }); // Model audio back into the call model.on("audio", (pcm) => call.send(pcm, { binary: true }));});
Set connect: true, or the call ends about a second after the stream starts. Streaming costs KES 0.50 a minute on top of the call rate, and your model provider bills you for the model. The full build is in How to build an AI voice agent.