[!WARNING] Major upgrades underway. We're moving fast, so expect breaking changes and some instability for a while. We're pausing contributions for now, but bug reports are welcome: report a bug.

The AI that answers the phone.

Fonoster is an open-source platform for building, testing, and running voice agents on real phone lines. Use Fonoster Cloud, or run it on your own servers.

Features

The most notable features of Fonoster are:

• Multitenancy
• Easy deployment of PBX functionalities
• Programmable Voice Applications
• NodeJS SDK
• Support for Amazon Simple Storage Service (S3)
• Secure API endpoints with Let's Encrypt
• Authentication with OAuth2
• Authentication with JWT
• Role-Based Access Control (RBAC)
• Plugins-based Command-line Tool
• Support for Google Speech APIs

Code Examples

Autopilot

With Autopilot, your agent is a YAML file. Pick the speech-to-text, the voice, and the language model, then write the prompt:

# agent.yaml
name: front-desk
type: AUTOPILOT
speechToText:
  productRef: stt.deepgram
  config:
    model: nova-3
    languageCode: multi
textToSpeech:
  productRef: tts.deepgram
  config:
    voice: aura-asteria-en
intelligence:
  productRef: llm.openai
  config:
    conversationSettings:
      firstMessage: Hi! Thanks for calling. How can I help?
      systemPrompt: You book appointments for a dental clinic. Keep your answers short.
      goodbyeMessage: Thanks for calling. Goodbye!
      systemErrorMessage: Sorry, I'm having trouble right now. Please call again later.
      idleOptions:
        message: Are you still there?
    languageModel:
      provider: openai
      model: gpt-4o
      temperature: 0.7
      maxTokens: 250

Deploy it with the command-line tool:

# Log in once with an API key from your workspace
npx @fonoster/ctl workspaces:login

# Create the agent
npx @fonoster/ctl applications:create -f agent.yaml

Then point a phone number at the new application (in the dashboard, or with npx @fonoster/ctl sipnet:numbers:create) and give it a call. Autopilot also supports tools, a knowledge base, and call transfers. See the Autopilot docs.

Voice Applications

For full control over the call, write a Voice Application instead. Fonoster opens one signed WebSocket session to your app for every call (session protocol); the SDKs for TypeScript (@fonoster/voice) and Python (fonoster-voice) wrap it. A session can:

• say - Speak text (or a stream of model tokens) and report what the caller heard
• gather - Wait for speech or keypresses
• play / playDtmf - Play a media file or DTMF tones
• transfer - Cold, warm or SIP REFER transfer to a number or SIP URI
• hold / resume, handoff, hangup
• audio - Raw call audio in and out, for your own speech stack
• Events: turns, partial transcripts, keypresses

Voice Application Example:

const { VoiceServer } = require("@fonoster/voice");

// The gateway signs every connection; set APP_SIGNING_SECRET to the app's signing secret.
new VoiceServer({ port: 3000 }).listen(async (session) => {
  await session.say("Hi there! What's your name?");

  const { speech: name } = await session.gather();

  await session.say("Nice to meet you " + name + "!");

  await session.say("Please enter your 4 digit pin.");

  const { digits } = await session.gather({
    maxDigits: 4,
    finishOnKey: "#"
  });

  await session.say("Your pin is " + digits);

  await session.hangup();
});

// Your app lives at ws://127.0.0.1:3000
// and you can publish it to the Internet with:
// ngrok http 3000

Coming from the gRPC SDK? See What changed.

Making calls

Everything in Fonoster is an API first, and initiating a call is no exception. You can use the SDK to start a call with a few lines of code.

Example of originating a call with the SDK:

const SDK = require("@fonoster/sdk");

async function main(request) {
  const apiKey = "your-api-key";
  const apiSecret = "your-api-secret"
  const accessKeyId = "WO00000000000000000000000000000000";

  const client = new SDK.Client({ accessKeyId });
  await client.loginWithApiKey(apiKey, apiSecret);

  const calls = new SDK.Calls(client);
  const response = await calls.createCall(request);

  console.log(response); // successful response
}

const request = {
  from: "+18287854037",
  to: "+17853178070",
  appRef: "3e61ecb7-a1b6-4a93-84c3-4f1979165bca",
  // Optional metadata to be sent to the Voice Application
  metadata: {
    name: "John Doe",
    message: "Please call me back."
  }
};

main(request).catch(console.error);

Getting Started

To get started with Fonoster, use the following resources:

• Deploying Fonoster with Docker
• Guide for Early Access User
• Getting started with Fonoster
• How we created an open-source alternative to Twilio and why it matters

Guides in this repository

• Try the new platform on your machine
• Session protocol and What changed
• Workspace webhooks: events, signatures, retries, migrating from eventsHook
• Call records: contents, retention and erasure, redaction, verification
• Providers: catalog ids, your own keys, deprecation warnings
• Operators: deployment, regions, provider registry

Give a Star! ⭐

Please give it a star if you like this project or plan to use it. Thanks 🙏

Bugs and Feedback

For bugs, questions, and discussions, please use the Github Issues

Contributing

For contributing, please see the following links:

• Contribution Documents
• Contributors

Sponsors

We're glad to be supported by respected companies and individuals from several industries.

Find all our supporters here

Become a Github Sponsor

Authors

• Pedro Sanders

License

Copyright (C) 2026 by Fonoster Inc. MIT License (see LICENSE for details).