Documentation

StreamCore AI Docs

StreamCore AI is an open-source, real-time voice agent framework. A high-performance Go server connects browsers or phones via WebRTC, streams through STT, reasons with an LLM, speaks back with TTS, and extends everything with Python & TypeScript plugins.

Audio

48kHz Opus

Latency

20ms frames

Client SDKs

4 languages

License

Apache 2.0

Quickstart

Three commands to a working voice agent. Clone, configure, run.

Architecture

Four goroutines, bounded channels, one voice loop. How the pipeline works end-to-end.

SDKs

Four client SDKs (TypeScript, Go, Python, Rust) and two plugin SDKs. One protocol, any language.

Plugins

Extend the agent with external tools in Python, TypeScript, or JavaScript via JSON-RPC.

Skills

Shape agent behavior with Markdown files. Tune personality, domain expertise, and response style.

Configuration

Complete config.toml reference for server, providers, plugins, and pipeline settings.

SIP & Telephony

Bridge phone calls into the voice pipeline via SIP trunks. Twilio, RingCentral, and Asterisk.

How it works

Each voice session runs an isolated 4-goroutine pipeline connected by bounded channels:

Browser Mic→Reader (RTP → Opus decode)→Inbound (STT + VAD)→Agent (LLM + Tools + TTS)→Sender (Opus encode → RTP)→Browser Speaker

WHIP signaling (RFC 9725) creates sessions with a single HTTP POST. A DataChannel carries transcripts and events bidirectionally. No WebSocket required.