ultra-streamer-rs

Stream any Rust/wgpu application to a web browser with GPU hardware encoding and sub-30ms LAN latency.

Overview

Capture GPU frames from your wgpu render loop, encode with platform-native hardware encoders, and stream to Chrome/Chromium via WebTransport or WebSocket. The browser decodes with WebCodecs and forwards user input back to the host.

Target Hardware

Workspace

crates/
├── ustreamer-proto      # Wire protocol types
├── ustreamer-capture    # GPU frame capture
├── ustreamer-encode     # HW video encoding (VideoToolbox, NVENC, AMF)
├── ustreamer-transport  # WebTransport + WebSocket server
├── ustreamer-input      # Browser input mapping
├── ustreamer-quality    # Adaptive quality controller
├── ustreamer-app        # Integration traits for external apps
├── ustreamer-demo       # Headless demo server
└── ustreamer-nvenc-probe # NVENC validation harness (Windows)

client/                  # Browser client

Quick Start

# macOS (VideoToolbox)
cargo run -p ustreamer-demo

# NVIDIA (direct NVENC)
cargo run -p ustreamer-demo --features nvenc-direct

# AMD (direct AMF)
cargo run -p ustreamer-demo --features amf-direct

# Auto-detect GPU vendor (recommended for mixed hardware)
cargo run -p ustreamer-demo --features nvenc-direct,amf-direct

Then open http://127.0.0.1:8090/ in Chrome/Chromium.

The demo auto-detects your GPU and selects the best encoder. Build with nvenc-direct,amf-direct when you want one Windows/Linux binary that can choose NVIDIA or AMD at startup. Override with --codec hevc|av1 or --nvenc-device <n> if needed.

Building

cargo build
cargo test

Publishing

scripts/publish-crates.sh list       # show release order
scripts/publish-crates.sh stage1 --dry-run

Stages: proto/capture → input/quality/transport/encode → app

Architecture

See PLAN.md for the full design document.

License

MIT OR Apache-2.0