API

What is an API?

Think of an API as a shop counter. You don’t walk into the stockroom: you ask at the counter, following clear rules, and you’re handed what you asked for. An API does the same between programs. It publishes a list of operations, such as “transcribe this audio” or “create this contact”, and spells out what data to send and what you’ll get back.

In AI, almost every model is used this way. Your app doesn’t run the model itself: it sends text or audio to an API and receives the result. Each of those calls is an inference that someone pays for, often in tokens.

How it works

Most web APIs follow the same pattern:

1. Request. Your program sends an HTTP call to an address (the endpoint), with a method (GET to read, POST to create) and a key that identifies you.
2. Processing. The service checks the key, does the work and prepares a reply.
3. Response. You get a status code (200 if it worked, 4xx or 5xx if not) and the data, nearly always as JSON.

A typical response body looks like this:

{
  "id": "job_123",
  "status": "completed",
  "text": "Hi, I'd like to book an appointment on Thursday."
}

It helps to keep neighbouring terms apart. With an API, you ask. With a webhook, you get told when something happens. MCP is a standard that lets an AI model discover and use tools, which are often APIs underneath. And function calling is how the model decides which function to call and with what arguments.

Example

A physiotherapy clinic in Bristol keeps the voice notes its physios record after each session. Their developer writes a short script: every night it sends the new recordings to an automatic transcription API, gets the text back and adds it to the patient record. Nobody types up notes any more, and records are complete by the next morning.

The UTIO AI API

UTIO offers an AI API for developers and agencies: transcription, text to speech, text, image, OCR and embeddings, plus UTIO’s CRM and channels, all with one key. They are the same endpoints UTIO’s own products run on in production.

Each access token is created in the panel with the client it belongs to, a spending cap and a rate limit, and can be revoked instantly. Calls that send through a channel (WhatsApp, voice or email) require an Idempotency-Key header, so a retry never duplicates a message. The full reference for parameters and models lives in the panel.