Jargo Brings Real-Time Voice Agents to Go

Summary

Jargo is a WebRTC-native, audio-first conversational-AI framework for building real-time voice agents in Go. It implements streaming speech-to-text, LLM reasoning, and text-to-speech pipelines, with turn-taking and barge-in support; it also supports speech-to-speech services including OpenAI Realtime, Gemini Live, AWS Nova Sonic, and xAI Realtime. The project is a Go port of the Pipecat architecture, aimed at deployments where a static binary, predictable memory use, fast startup, and Go concurrency are preferable to a Python runtime. It supports WebRTC through Pion, WebSockets, LiveKit, local audio, Opus processing, optional Twilio telephony, RTVI data channels, prompt caching, memory through mem0, and interchangeable provider interfaces. Silero VAD and Smart Turn v3 run locally through ONNX for interruption and end-of-turn detection, although the bot can fall back to STT endpointing without ONNX Runtime. Provider integrations cover numerous STT, LLM, and TTS services, including local and hosted options. The repository includes runnable echo, voicebot, provider-specific, and Twilio examples, Docker deployment images, and documentation. The default build is cgo-free; ONNX Runtime is needed at runtime for turn-taking, while RNNoise is optional.