TEN Framework
A self-hosted voice AI framework with RTC and WebSocket support and Docker deployment; its default example uses Agora, OpenAI and Deepgram services.
TEN Framework is a self-hosted framework for developers building voice agents and multimodal conversational applications. Its default assistant example depends on external services even when the application runs on your hardware.
- Live voice connections: Build conversational agents with RTC or WebSocket connections for low-latency spoken interaction.
- Customizable assistants: Adapt the example's speech recognition, language model and speech synthesis components for your application.
- Conversation extensions: Extend agents with memory, voice activity detection (VAD) and turn detection to handle conversational context and speech timing.
- Visual component editing: Use TMAN Designer to customize agent components, with a separate browser interface for interacting with the examples.
The framework also supports video and other multimodal interactions. Agent examples can run locally through Docker or on a self-hosted server, and the frontend can be hosted separately from the backend.
The default example requires credentials for Agora, OpenAI and Deepgram ASR, so self-hosting that example doesn't provide offline operation. Its stated minimum hardware requirement is a two-core CPU and 4 GB of RAM.
Logotyper
