Muxiva
A real-time multimodal agent runtime with one graph and lifecycle contract across Rust, C++, Python, and TypeScript.
Muxiva owns scheduling, bounded queues, backpressure, cancellation, lifecycle, control messages, and observability. Applications provide Nodes for speech, reasoning, media, transport, and business logic.
Pre-alpha
The runtime foundation is tested, but public packages and official Node integrations are still evolving. Do not execute untrusted Node code or expose Studio directly to the internet.
Experience a real voice agent
macOS/Windows shells default to Studio; Linux defaults to Headless Runtime. Use
--studio or --headless to choose explicitly.
flowchart LR
WEB["Browser microphone"] --> AGORA_IN["Agora C++ ingress"]
AGORA_IN --> QWEN["Qwen speech intelligence"]
QWEN --> PI["Pi TypeScript Agent when cascaded"]
PI --> FORMAT["Speech Formatter"]
FORMAT --> TTS["Qwen TTS"]
TTS --> AGORA_OUT["Agora C++ egress"]
QWEN --> AGORA_OUT
AGORA_OUT --> WEB
QWEN --> DATA["Agora ordered data stream"]
PI --> DATA
DATA --> LIVE["Client transcript · barge-in state"]This credentialed flagship application captures a real microphone in the browser, transports real audio through Agora, and uses Qwen for speech understanding and generation. Studio offers a low-latency Realtime graph and an inspectable full-duplex Qwen Server VAD + ASR → tool-using Pi TypeScript Agent → Speech Formatter → cancellable Qwen TTS graph with live Node, Tool Call, Frame, and conversation state.
Run the flagship voice demo Open the Studio guide
Choose your path
| Goal | Start here |
|---|---|
| Run a real voice assistant | Flagship voice demo |
| Install and run a graph | Installation and first run |
| Build visually | Muxiva Studio |
| Understand the runtime | Runtime architecture |
| Integrate an existing Agent | Agent integration SOP |
| Build an Agent application or Node | Developer manual |
| Integrate RTC or media | Official and custom Nodes |
| Contribute safely | Contributing |