Voice and WhatsApp agents that answer the customer, capture the essentials and hand the job to the right technician in minutes. Underneath, Sable Hub: an operations platform built to be used by agents, with an API, MCP and per-agent permissions.
In production, running our own home-service operations in Spain.
My boiler stopped and I have no hot water.
The customer calls or writes. The agent works out what is wrong, where, and how to reach them, and turns the conversation into a job ready to assign. When a person is needed, the conversation moves to them with full context.
Phone
Answer inbound calls 24/7 for home repairs and services: boilers, plumbing, locksmiths, blinds. Programmable telephony (SIP), speech-to-text and large language models.
Messaging
The same intake and follow-up, over WhatsApp. When needed, the conversation moves to a person on the team without losing the thread.
Every module is exposed through a REST API with an OpenAPI spec, an MCP server and llms.txt. Agents work on the same data the team sees in the interface.
| Module | REST API | MCP | Interface |
|---|---|---|---|
| Lead intakeWith channel attribution | |||
| Ad spendCost per job | |||
| Technician and supplier networkWith coverage map | |||
| WhatsApp sessionsHealth of each session | |||
| Earnings and invoicing | |||
| Bank reconciliation |
Each agent gets its own token, scoped per module. The voice agent can create jobs without seeing invoicing; the reconciliation agent reads transactions without touching the technician network.
OpenAPI spec, llms.txt and an MCP server: a new agent finds out what it can do without anyone hand-writing an integration.
The interface is not a separate layer: it shows the same data agents read and write.
The platform runs on Cloudflare: compute, database, queues and storage at the edge, with no servers to maintain. On top, programmable telephony, speech-to-text and large language models.
Calls and messages arrive through programmable telephony and WhatsApp.
Every event is received and queued at the edge.
Speech-to-text and language models understand and decide the next step.
Jobs, technician network, earnings and reconciliation in one store.
The technician gets the job; the team sees it in the interface.
Rules we learned running our own home-service business on these agents.
The job is offered without customer details. The contact only goes to the technician who accepts, never to two at once.
Area, problem and contact are enough to hand off the job. The technician sorts out the rest when they call.
Anything that needs human judgement goes to a person on the team, with the full conversation in front of them.
We built it for our own operations and use it every day. If you run home repairs or services and want to try the agents or Sable Hub, write to hola@sablestudio.es.