TL;DR

Neko (m1k1o/neko) is a self-hosted virtual browser that runs in Docker and streams over WebRTC at under 300ms latency. One container, many users, full mouse/keyboard sharing. It started as a watch-party tool to replace rabb.it — but in 2026 it's become a quiet favorite for OSINT investigators, red teamers, and ops teams who need a disposable browser that leaks nothing to the endpoint. 20.7k stars, Apache-2.0, runs on a Raspberry Pi or an NVIDIA box.

What it is, in one paragraph

Neko boots Firefox, Chrome, Brave, Tor, Vivaldi, Edge, Waterfox, Opera, or Ungoogled-Chromium — or a full XFCE/KDE desktop, or VLC, or Remmina — inside a Linux container, then streams the desktop to your browser via WebRTC with synchronized audio. Multiple participants can join the same room and the host hands out control like a remote presenter. Because only video and audio leave the container, nothing the browser does ever touches the client machine: no cookies, no cache, no history, no fingerprint.

Why it matters

Most “remote browser” stories you've seen are paid SaaS (Hyperbeam, Mighty, Browserless cloud). Neko is the open-source equivalent that you run yourself — on a $40 Raspberry Pi if you want, or an NVIDIA-equipped node if you're running it for a team. The combination of WebRTC + multi-user control + real Linux apps is genuinely rare in FOSS. Apache Guacamole and noVNC stream images over WebSockets, have no real audio, and don't do native multi-participant control. Neko fixes all three at once.

Technical facts

  • Latency: <300ms WebRTC, smooth video, synchronized audio.
  • Stack: Go 1.24/1.25 backend, Pion WebRTC v3, GStreamer pipelines, custom X11 input drivers, Vue 3 + TypeScript client.
  • Latest release: v3.1.0 — base bumped to Debian Trixie, Chromium now runs no-sandbox so --cap-add=NET_ADMIN is no longer required.
  • Multi-arch: linux/amd64, linux/arm64, linux/arm/v7. Raspberry Pi works.
  • GPU acceleration: NVIDIA NVENC h264 encoding + VirtualGL 3.1 for Chromium.
  • Surface area: REST API with OpenAPI 3.0, Prometheus metrics, Go plugin system, embeddable iframe via ?embed=1, auto-join links via ?pwd=.
  • Auth (v3): multi-method auth, persistent cookie sessions, granular per-user permissions (mute chat, block uploads, etc.).
  • Stream knobs: VIDEO_BITRATE, AUDIO_BITRATE, MAX_FPS (set to 0 for real 60fps). Adaptive bitrate experimental.
  • Distribution: 36 Docker packages on ghcr.io, 500K+ pulls on Docker Hub, 57 contributors.

How it stacks up

SolutionStreamingAudioMulti-user controlSelf-hostEmbeddable API
NekoWebRTC <300msBuilt-inNativeYes (Apache-2.0)Yes
Apache GuacamoleImages / WebSocketNoneNoYesLimited
noVNC + websockifyImages / WebSocketNoneNoYesNo
HyperbeamWebRTCYesYesNo (SaaS)Yes (paid)
Kasm WorkspacesWebRTC / WebYesLimitedYes (community / paid)Limited

Where it actually shines

OSINT & threat intel

Drop a suspicious URL into a Neko-hosted Tor Browser. The container handles any drive-by exploit, any fingerprinting attempt, any tracking pixel. When you close the room, the container is destroyed and nothing made it to your laptop. To your ISP, the traffic looks like a video call. Combine with a VPN for additional anonymity.

Secure jump host without a VPN

Expose Neko in front of internal admin panels, databases, and dashboards. Engineers reach them through a browser tab. Only video and audio leave the bastion — no session tokens, no cookies, no DB credentials ever touch the endpoint. If a laptop is compromised, the attacker can't pivot. Bonus: every session can be RTMP-recorded for compliance audit trails.

Collaborative debugging & demos

Three engineers join the same room. One drives, the others watch and chime in — or the host grants control mid-call. Same flow works for product demos to clients without ever shipping the underlying code: only pixels travel.

Browser automation with a human override

Install Playwright or Puppeteer inside the container. Watch the script run from outside. When it stalls on a CAPTCHA or a UI change, just take control and unstick it manually.

The original use case still works

Watch parties, interactive presentations, classroom teaching, even running it inside VRChat — all still first-class.

Limitations & pricing

Cost: $0. Apache-2.0. You only pay for the box.

  • An active crypto-token scam on X/Twitter is impersonating the project. The real Neko project has no Twitter account and no crypto involvement — the homepage explicitly warns about this.
  • AV1 pipeline ships but doesn't work — encoding is too slow today; stick with h264 (NVENC if you have a GPU).
  • iOS audio needs iOS 15 or newer.
  • v3.0 shipped a backend rewrite ahead of the client refresh, so a few new server features aren't exposed in the UI yet (clarified in the v3.1 docs).
  • Strict firewalls that block WebRTC fall back to a WebSocket screencast — functional, but you lose the smoothness.

What's next

The roadmap signals worth watching: stabilizing webcam and microphone passthrough, adaptive bitrate going from experimental to default, and a long-discussed mode where Neko acts as a pure WebRTC relay in front of arbitrary RDP/VNC backends — turning Neko into a universal “web frontend for any remote desktop.” The sister projects (neko-rooms for multi-room orchestration with zero-knowledge HTTPS install, neko-apps for community-packaged Docker images, plus neko-vpn) make it realistic to deploy a small in-house Hyperbeam clone in an afternoon.

Sources: m1k1o/neko on GitHub, neko.m1k1o.net, v3 docs, v3.1.0 release notes.