Developers

An API that reads
like the dashboard.

Everything you can click, you can call. Assistants, calls, leads, campaigns, voices and models over plain REST with JSON. Keys carry scopes, webhooks carry signatures, and the docs are written for people who would rather copy a working example than read a spec. The spec is there too.

Full API guide Webhooks Tools and MCP

Quickstart

Three requests to your first call.

Create a key under Developers → API keys, tick the scopes it needs
Send it as the X-API-Key header; the key acts in the workspace it was created in
Create an assistant, or use one you built in the dashboard, and place a call
Poll the call, or let a webhook tell you when it completes

Base URL: https://nordtell.com/api/v1. Sixty requests a minute per key; above that you get a 429 and wait. Errors say which field was wrong.

curl

# 1. an assistant, from a prompt in plain words
curl -X POST https://nordtell.com/api/v1/assistants \
  -H "X-API-Key: $NORDTELL_KEY" -H "Content-Type: application/json" \
  -d '{"name": "Reception", "direction": "inbound", "engine": "hybrid",
       "config": {"prompt": "You are Freja, the receptionist at Nørrebro Clinic…"}}'

# 2. a call
curl -X POST https://nordtell.com/api/v1/calls/make \
  -H "X-API-Key: $NORDTELL_KEY" -H "Content-Type: application/json" \
  -d '{"agent_id": "ASSISTANT_ID", "to_number": "+4531183990",
       "variables": {"name": "Sara"}}'

# 3. what happened
curl https://nordtell.com/api/v1/calls/CALL_ID -H "X-API-Key: $NORDTELL_KEY"

Webhooks

We tap you on the shoulder.

Register a URL and a secret. When a call starts, is answered, ends or completes, you get a signed JSON body with the transcript, the summary, the variables the agent collected and the minutes it cost. One endpoint per system, so you can revoke the CRM's without touching Slack's.

Events: call.started, call.answered, call.ended, call.completed, plus agent and lead events
HMAC signature in the header; verify it, then trust the body
Delivery log on every call, with the status your server answered

POST https://your-server/webhooks/nordtell

{
  "event": "call.completed",
  "data": {
    "id": "b0c1…", "direction": "inbound", "duration_seconds": 134,
    "from_number": "+45…", "summary": "Booked a check-up Tue 14:40.",
    "variables": { "appointment": "2026-09-15T14:40" },
    "transcript": [ … ]
  }
}

Tools and MCP

Your API, in the agent's hands.

A tool is a function with a name, a description and typed parameters. Point it at your HTTP endpoint and the model calls it mid-conversation, waits for the answer and speaks it. Already have an MCP server? Add its URL and every tool it exposes appears in the agent editor.

Tool names are function names: check_availability, not "Check availability"
Per-tool timeouts and a spoken filler while the tool runs
Built-ins: transfer, end call, send SMS, current time

tool definition

{
  "name": "check_availability",
  "description": "Free slots for a practitioner on a date",
  "parameters": {
    "practitioner": { "type": "string" },
    "date": { "type": "string", "format": "date" }
  },
  "url": "https://clinic.example/api/slots",
  "method": "GET"
}

Endpoints

What is on the menu.

Each key gets only the scopes you tick. The full reference, with request and response bodies, lives in the docs and in the OpenAPI spec.

Endpoint	Scope	What it does
GET / POST /assistants	agents:read / agents:write	List and create assistants; PATCH and DELETE by id
GET /calls, GET /calls/{id}	calls:read	Call list; transcript, summary, variables and deliveries
POST /calls/make	calls:make	Place an outbound call from an assistant's number
GET / POST /leads	leads:read / leads:write	Leads with phone, name, variables and status
POST /campaigns, …/leads, …/start, …/stop	campaigns:manage	Outbound campaigns: create, add leads, run, pause
GET /languages, /models, /voices	any key	The lookups the editor uses, so your UI can match ours