Skip to main content
ShipVoice Lite is the free MIT starter for real-time voice agents. It is the whole loop, not a single script: microphone to speech recognition to model to speech, over WebRTC, with a console in front of it. “ShipVoice” names the product line. The free starter is ShipVoice Lite. The paid product is ShipVoice Pro, a separate repository that shares this one’s stack and lineage.

​
The three services

Each runs on its own. The backend and the agent share one set of LiveKit credentials, so a backend-minted token is valid for the room the worker joins. Works against LiveKit Cloud or a self-hosted LiveKit.

​
What is in the console

Overview, Calls, Call detail, Agents, Agent detail, Test call and Deployment are built and read this deployment. Campaigns, Channels, Customers and Evaluations are greyed out and inert: they show the shape of the full product without pretending to have data behind them. On the Overview, figures this repo cannot measure carry the reference console’s own numbers behind a small red dot meaning sample data. The cost, kept and config columns of a real call row are dashes, never samples.

​
What it is not

There is no authentication on the backend. Every route is open. There is no metering, no billing, no campaign runner and no evaluation harness. Those belong to ShipVoice Pro.

Quickstart

One command to Postgres, the backend, the worker and the console.

Console mode

Prove the agent talks before anything else is up.

​
License

MIT for the code written here. Some UI components are vendored from public component registries and are licensed by their upstream authors. See LICENSE and THIRD_PARTY_NOTICES.md in the repository.