A Sautikit AI agent phones each new lead for you, asks your qualifying questions in a natural conversation, and saves the answers to your CRM so your reps only call the leads worth their time. A 90-second answered call costs about KES 10.50, and a call that is not answered costs nothing.
It suits sales teams in insurance, property, lending, schools and software that want to reach every enquiry within minutes, and research teams that run phone surveys over a list.
Someone fills in your enquiry form or a list of leads arrives from a campaign.
Your system tells Sautikit to call them. The call comes from your own Sautikit number.
The agent introduces your business, checks it is a good time, and asks three or four questions: what they need, their budget, their timeline, who decides.
The lead can ask questions back and interrupt. The agent answers from what you told it in the prompt.
At the end of the call the agent saves the answers and a score to your CRM.
Your rep sees a qualified lead with the answers already filled in and phones them while the interest is fresh.
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: "Lead qualifier", base_prompt: "You call people who asked about solar installations from a company in " + "Nairobi. Be brief and polite. Ask: is the property a home or a business, " + "what is the monthly electricity bill, do they own the property, and when " + "do they want to install. Save the answers before you end the call. If they " + "ask not to be called again, save that and say goodbye.", model: "sautikit-flash", max_call_minutes: 5, }),}).then((r) => r.json());await fetch(`${api}/agents/${agent.id}/tools/${crypto.randomUUID()}`, { method: "PUT", headers, body: JSON.stringify({ name: "save_lead_result", description: "Save the lead's answers and a qualification score at the end of the call.", url: "https://crm.example.com/agent/leads/result", method: "POST", parameters: { type: "object", properties: { lead_id: { type: "string" }, property_type: { type: "string", enum: ["home", "business"] }, monthly_bill_kes: { type: "number" }, owns_property: { type: "boolean" }, timeline: { type: "string" }, score: { type: "string", enum: ["hot", "warm", "cold", "do_not_call"] }, }, required: ["lead_id", "score"], }, }),});await fetch(`${api}/agents/${agent.id}/publish`, { method: "POST", headers });
POST /v1/agents/{id}/send places one call. Run it from your form handler:
app.post("/enquiry", async (req, res) => { const lead = await crm.createLead(req.body); await fetch(`${api}/agents/${process.env.AGENT_ID}/send`, { method: "POST", headers, body: JSON.stringify({ to: lead.phone, // "+254720000000" prompt_template: "You are calling {{name}}, lead id {{lead_id}}, who asked about solar for a {{interest}}.", variables: { name: lead.name, lead_id: lead.id, interest: lead.interest }, }), }); res.sendStatus(202);});
For a list from an event, an old database or a survey sample, create a broadcast. It calls each contact inside the hours you set and retries the ones who did not answer.
A survey is the same setup with a different prompt and a different tool. Tell the agent the questions and the answer choices, give it a save_survey_answers tool, and run a broadcast over your sample. Because the agent holds a conversation, it can clarify a question the respondent did not understand, which a recorded message cannot do.
Call leads who gave you their number and asked to hear from you, keep to reasonable hours, say who is calling in the first sentence, and record a "do not call" answer so the lead is not phoned again.
If you run your own voice model, place the call with POST /v1/calls and return a stream action from your number's voice callback. Sautikit sends the live audio to your WebSocket:
Your server must accept the audio.drachtio.org WebSocket subprotocol. Streaming costs KES 0.50 a minute on top of the KES 3.00 call rate, plus your model provider's charges. See How to build an AI voice agent.