Voice AI API for developers
One API for agents, calls and call data
A voice AI API lets your software create AI voice agents and run phone conversations with them from code. With Retell, you create and publish agents, start outbound calls, answer inbound ones on your numbers, and receive call events and transcripts at your own endpoint. Retell handles the telephony, speech recognition, language model and voice, so you don't connect those providers yourself.
.svg)
Bring your own LLM over WebSocket
With a custom LLM, Retell still handles telephony, transcription, turn-taking and speech, and opens one WebSocket to your server per call. It sends the live transcript and asks for a response whenever the agent should speak, and your server streams back what the agent says. Use it when prompts and transcripts must stay in your own infrastructure, or when you run a model Retell doesn't offer. For most agents, Retell recommends its own single prompt or conversation flow agents.
A Retell é a plataforma de voz com IA sobre a qual os contact centers corporativos operam
How to go from API key to live call
Node.js: npm i retell-sdk
Python: pip install retell-sdk
Or call the REST API at https://api.retellai.com with the key as a bearer token
- Hubspot
- Twilio
- Vonage
- Go High Level
- n8n
- Zapier
- Salesforce
- Hubspot
- Twillio
- Vonage
- Go High Level
- 8n8
- Zapier
- Salesforce
- Avaya
- Genesys
- Five9
- Amazon Connect
- Telnyx
- Make
- Cal.com
.svg)
- Avaya
- Genesys
- Five9
- Amazon Connect
- Telnyx
- Make
- Cal.com
.avif)
.avif)
.avif)
.svg)
What the Retell voice AI API covers
A voice AI API is an interface that lets your code create AI voice agents and run calls with them. Retell's API covers calls and chats, voice and chat agents, phone numbers, voices, knowledge bases and tests, plus custom telephony and custom LLM integrations. Every REST request goes to one base URL, https://api.retellai.com, with your API key as a bearer token, and Retell runs the telephony, speech and language model layers of each call.
The building blocks and where they are documented
| What you want to do | How it works in the API | Docs |
|---|---|---|
| Create an agent | Create a Retell LLM response engine with your prompt and tools, then an agent that attaches it to a voice, language and call settings. Publish a version when it is ready. | Create an agent with the SDK |
| Place outbound calls | POST /v2/create-phone-call with a from number you bought from or imported to Retell and a to number in E.164 format |
Create Phone Call |
| Receive inbound calls | Bind an inbound agent to the number. An optional inbound webhook can pick the agent and add dynamic variables before the call connects. | Inbound call webhook |
| Talk in the browser | Create a web call and connect it with the Web SDK, using a public key in the browser and your API key only on the server | Make a web call |
| Get call events | Account-level or agent-level webhooks POST each event with the call object. Verify the x-retell-signature header with the SDK. |
Webhooks overview |
| Run your own LLM | Retell opens one WebSocket to your server per call and asks for a response each time the agent should speak | Custom LLM overview |
| Use your own carrier | Import numbers over elastic SIP trunking, or dial calls to Retell's SIP URI | Custom telephony |
A first outbound call in Python
This sample from the Create Phone Call reference starts a call from a number on your account. The agent bound to that number for outbound calls handles the conversation.
import os
from retell import Retell
client = Retell(
api_key=os.environ.get("RETELL_API_KEY"), # This is the default and can be omitted
)
phone_call_response = client.call.create_phone_call(
from_number="+14157774444",
to_number="+12137774445",
)
print(phone_call_response.agent_id)Source: Retell API reference, Create Phone Call, read November 9 2026. The Node.js version is on the same page.
Our glossary explains what a voice API is and how a voice API differs from a voice AI API, and how webhooks push events to your systems. If you plan to bring your own model, compare the options in our guide to the best LLM for voice AI agents. Agents that run on the Retell LLM can draw on a Retell knowledge base, which the API manages too, and every agent's post-call analysis arrives in the call_analyzed webhook.
