The job: customers write to your WhatsApp number, an AI agent drafts each answer, and a person approves it before it goes. The number stays yours, on your phone; nothing is sent that nobody read.
1. Link the number
curl -X POST https://makehybrid.com/whatsapp/v1/connection/link \ -H "X-API-KEY: $MAKEHYBRID_KEY"
Open the url it returns on a computer and scan the code with the phone: WhatsApp, Settings, Linked devices, Link a device. GET /whatsapp/v1/connection then shows the number as connected.
2. Hear what customers write
curl -X POST https://makehybrid.com/whatsapp/v1/webhooks/endpoints \
-H "X-API-KEY: $MAKEHYBRID_KEY" -H "content-type: application/json" \
-d '{"url":"https://acme.com/hooks/whatsapp","events":["whatsapp.message_received"]}'
3. Draft, approve, send
In a project room, give an AI agent of your own the WhatsApp agent under Can use, and write its instructions: draft an answer to each new message, post the draft in the room, and send it only when a person replies go. Everyone in the room sees the draft and the approval.
curl -X POST https://makehybrid.com/whatsapp/v1/messages \
-H "X-API-KEY: $MAKEHYBRID_KEY" -H "content-type: application/json" \
-d '{"to":"+15551234567","text":"Your order is on its way."}'
Messages leave 10 to 20 seconds apart, so a reply can take that long to arrive. Write to people who wrote to you or expect you, and never the same text to many.
Questions people ask
Can the agent send without approval?
It can if its instructions say so; the API does not stop it. Asking for a go in the room is the safer default for anything a customer reads.
Can it read pictures customers send?
It sees that a message carried media, not the file. Ask the customer to describe it, or answer from the phone.