Proprietary software by InferApps Tech PVT LTD. The source code is private. This page describes the project and what we're building.
An embeddable voice agent that does the time-critical work on the device: listening, knowing when you've finished speaking, transcribing, calling local tools and speaking back. Only the reasoning goes to a cloud language model that the host product chooses. The result is snappy conversation with less data leaving the device.
- On-device listening and turn-taking: voice activity and end-of-turn detection.
- On-device speech-to-text and text-to-speech.
- Natural interruptions: talk over the agent and it stops and listens.
- Host-defined tools and instructions, with tool calls handled locally.
- Bring your own model: any OpenAI-compatible endpoint.
- Embeds anywhere: a C interface, a WebRTC relay with a browser demo, and Android and iOS apps.
In build. Available to partners under a written agreement.
To ask about it, open an issue in this repository. Part of Shanmuga Raja Portfolio.