Documentation | Issues | Discord

PulseBeam is an open source, general-purpose WebRTC SFU server for connecting browsers, mobile, and IoT clients. We believe real-time application development shouldn't be complicated, nor should it rely on heavy architectures with many moving parts. PulseBeam reduces this friction by adhering to these core design goals:

• Support all WebRTC clients.
• Keep the architecture simple, but not simpler.
• Natively support vertical and horizontal scaling.
• Provide client SDKs strictly for convenience, not necessity.
• Require minimal configuration.

If your client device speaks WebRTC, it can communicate with PulseBeam.

• CPU induced p99 jitter: <2ms

PulseBeam is opinionated about media handling to prioritize battery efficiency, hardware support, and predictable performance:

• Video: H.264 Baseline profile up to Level 4.1
• Audio: Opus
• Data Channel: Planned

This is a thread-per-core system that isolates the data plane from the control plane. It follows a shared-nothing design, keeping the fast path lock-free and cache-local.

The following quickstart assumes that you have a Linux machine. As a fallback, you can go to https://pulsebeam.dev/#quickstart and check the "fallback" toggle.

Docker/Podman (recommended):

docker run --rm --net=host ghcr.io/pulsebeamdev/pulsebeam:pulsebeam-v0.4.5 --dev

Open Port Requirements:

• TCP/7070: HTTP signaling
• TCP/6060: Internal CPU & Memory profilers
• UDP/3478: WebRTC traffic (Multiplexed)
• TCP/3478: WebRTC over TCP fallback (Multiplexed)

Note: The --dev flag used above configures PulseBeam to use port 3478 to avoid requiring root privileges. In production (running without --dev), WebRTC traffic defaults to standard port 443 (UDP/TCP).

Other options:

• Binary: download from Releases
• Source: cargo run --release -p pulsebeam

Run the following snippet in the browser console:

const pc = new RTCPeerConnection();
const stream = await navigator.mediaDevices.getUserMedia({ video: true });
const transceiver = pc.addTransceiver("video", {
  direction: "sendonly",
  // Define scalability layers (low, medium, high)
  sendEncodings: [
    { rid: "q", scaleResolutionDownBy: 4, maxBitrate: 150_000 },
    { rid: "h", scaleResolutionDownBy: 2, maxBitrate: 400_000 },
    { rid: "f", scaleResolutionDownBy: 1, maxBitrate: 1_250_000 },
  ],
});
transceiver.sender.replaceTrack(stream.getVideoTracks()[0]);

const offer = await pc.createOffer();
await pc.setLocalDescription(offer);

const res = await fetch("http://localhost:7070/api/v1/rooms/demo/participants", {
  method: "POST",
  headers: { "Content-Type": "application/sdp" },
  body: offer.sdp,
});

await pc.setRemoteDescription({ type: "answer", sdp: await res.text() });

Go to https://codepen.io/lherman-cs/pen/pvgVZar, then put "demo" as the room to connect to.

PulseBeam exposes an internal debug HTTP server on http://localhost:6060.

• Metrics (Prometheus): http://localhost:6060/metrics
• CPU profile (pprof): http://localhost:6060/debug/pprof/profile?seconds=30
• CPU flamegraph: http://localhost:6060/debug/pprof/profile?seconds=30&flamegraph=true
• Memory profile: http://localhost:6060/debug/pprof/allocs

CPU profiling measures CPU usage, not wall time. For meaningful results, profile while the server is under load.

View CPU profiles with:

go tool pprof -http=:8080 cpu.pprof

Or view as a flamegraph on a browser by specifying flamegraph=true to the URL query.

You can view the full roadmap here.