The self-hosted voice AI engine
Bring the AI to your calls, not your calls to their cloud.
Synauson is a real-time media engine with speech AI built in. Build voice agents you own and run them on your servers, with no per-minute bill.
Free plan
Start free, and stay free.
- 4 AI sessions at once
- 10 conferences at once
- Every AI feature, in production
- Never expires
Get a free keySee how it works
No credit card required. One free key per organization, sent by email.
Who it is for
For platforms
Embed it in your platform
Add voice AI to your PBX, contact center or communications product without sending your customers’ calls to a third party. Run it behind your hosted service, or ship it inside your product under an OEM agreement.
For businesses
Build your own agent
Stop paying per minute for a voice agent you will never own. Run the speech models yourself, use the LLM you already pay for, and watch your cost per call fall as your volume grows.
How it works
You write the agent. Synauson handles the call.
-
A call comes in
From a phone line over SIP, or from a browser or app over WebRTC.
-
Synauson listens
It hears the caller speak, works out when they’ve finished and writes down what they said, all on your servers.
-
Your agent answers
Your code gets the words, asks the LLM you choose what to say and sends back text. Synauson speaks it into the call.
Call audio stays on your machines. Only text goes to the model you choose.
Scale
Start on your laptop. Grow across your servers.
It’s the same engine at every size, so what you build on your desk is what you run in production.
On your laptop
Embed the engine in a Java app with jsyn, or run the server in a Linux container and call it over gRPC from any language with a gRPC library. SDKs for more languages are planned.
Across your servers
Run as many instances, replicas and sites as you need. They all draw on your license’s pools of AI sessions and conferences, so capacity follows your traffic. No machine needs its own license.
The server runs on x86_64 Linux; the Java SDK, jsyn, on x86_64 Windows and Linux. On a Mac, run the Linux container in Docker Desktop. On Apple silicon it runs under emulation, and slower, until the ARM build ships.
For contact centers
Hold queues and supervisor tools come with it.
Hold queues
Loop hold music in a call and queue announcements over it, such as a caller’s place in line that your code works out. You get an event as each one starts and ends.
Listen, whisper and join
An audio routing matrix decides who hears whom in each conference. Change it mid-call to let a supervisor listen in, coach the human agent or join.
Pricing
Need more than the Free plan? Pay for concurrency, not minutes.
Every plan includes every AI feature. Paid plans add AI sessions and conferences, let you burst above what you buy, and include support.
Each instance sends us usage reports, renews its license and downloads its models from us; here is exactly what it sends.