OpenClaw, Grokbot, OpenDot, Dot, OpenMaus or Claude get a voice: they call from your WhatsApp number when they need to, talk in real time — and if you ask them to do something on the call, they do it. One MCP tool, one API call or one skill.
Reminders, alerts and briefings — the agent picks the time and the topic, calls from your number and talks.
Call your number and say “book a meeting with John tomorrow at 10” or “cancel my 3pm”. The agent understands and does it on its side. Only allow-listed numbers reach the agent.
Through the place_agent_call tool (MCP), the API or the skill — with the goal of the call and what to say.
A number connected to Hasos calls the destination. They see your name and photo — and pick up.
A natural voice that listens and replies in under a second and can be interrupted, following the agent's instructions.
Answered? How long? What did they say? Transcript and summary for the agent to keep in memory.
“Call me at 8am and remind me about the supplier meeting.” You swipe away notifications; you answer calls.
Server down, payment declined, VIP customer waiting: the agent calls right away and explains what's going on.
Calendar, important emails and to-dos — summed up in a 2-minute call while you drive.
Confirm an appointment, notify a supplier, chase a reply — the agent calls from your number and brings back the result.
On the call, say “post on Slack that I'm running late” or “book John tomorrow at 10”: the voice hands it to your agent, it runs it with its own tools and confirms right away.
MCP (CLAUDE CODE, CURSOR, OPENCLAW…)
claude mcp add --transport http hasos https://api.hasos.com.br/mcp \ --header "x-api-key: hs_live_your_key"
REST API
curl -X POST https://api.hasos.com.br/v1/calls/agent \
-H "x-api-key: $HASOS_API_KEY" -H "Content-Type: application/json" \
-d '{
"deviceId": "YOUR_DEVICE",
"to": "+5511999998888",
"instructions": "Remind Augusto about the 3pm meeting and ask if he wants to reschedule. Speak English.",
"firstMessage": "Hi Augusto! This is your assistant."
}'Any agent that uses MCP, skills or HTTP calls — like OpenClaw, Grokbot, OpenDot, Dot, OpenMaus, Hermes, Claude and Cursor. Hasos exposes the place_agent_call tool on its MCP server, a REST endpoint and a ready-made skill (SKILL.md).
No. The voice is real-time speech-to-speech from OpenAI (Realtime) or xAI (Grok Voice), with your own API key, set up once in the dashboard. Voice minutes are billed by the provider on your account.
Yes, with live actions: set an HTTPS endpoint for your agent in the dashboard (or per call). When you ask for something — post on Slack, book a meeting, look something up — the voice sends the request to your agent, it runs it with its own tools and replies; the voice speaks the answer right away. Requests are HMAC-signed.
Yes. During the call the voice listens and replies. Afterwards, the get_call tool (or GET /v1/calls/{id}) returns whether it was answered, the duration and, with the AI Pack, the transcript and summary — for the agent to save and act on.
Yes, people who expect your call: you, or contacts you asked the agent to call. It's not for mass dialing — calls from each number are paced to protect your WhatsApp from being restricted.
It works on the Pro plans (from $49/mo, with the browser dialer and mass calling; transcripts via the AI Pack, $15 per number) and on the Agent plan ($19/mo, 1 number, agent calls only; transcripts and summaries with your own OpenAI key). Both with a 5-day free trial, billed in USD. The voice is billed by OpenAI or xAI on your key.
Your own WhatsApp number, connected to Hasos with a QR code (it keeps working on your phone). The person sees your name and photo, like a regular WhatsApp call.
$0 today: 5-day free trial. Connect your WhatsApp and ask for the first call.