WebClaw Launcher macOS Β· Windows Β· Linux

No setup, no config β€”
an AI coding OS in your browser

WebClaw packs an entire development machine into one browser window: chat with an AI assistant, write code, use a full graphical desktop. It all runs in a Docker sandbox, so you can break anything without breaking your computer.

Docker sandbox Β· one-click reset Β· your real files stay untouched

Prefer to try before installing?Open the live demoShared environment, resets every 5 min β€” don't enter anything sensitive

WebClaw β€” My WebClaw
Assistant Β· My WebClawGLM-4.6WebClaw routesandboxed
land007/webclaw:latestdesktopup 2h 14m

β€Ί Scrape the pricing tables from these 3 sites and put them in one spreadsheet

β”œβ”€ reading ~/work/sites.txt
└─ βœ“ installed playwright, chromium

! site 2 blocks headless β€” retrying with the desktop browser
βœ“ wrote pricing.xlsx Β· 3 sites Β· 47 rows

β€Ί Done. The file is in the workspace β€” open it in the editor or the desktop.β–Œ

AssistantWorkspace Β· Desktop Β· Terminal
Why a sandbox

An AI agent needs a whole computer.
It doesn't have to be yours.

Coding agents are useful precisely because they can install things, run commands and edit files. That is also exactly why you don't want one loose on your laptop.

Γ—Running the agent directly on your machine

  • β€”You install the runtimes, the CLIs and the dependencies yourself β€” before you get to do any actual work.
  • β€”The agent can read and write every file your user account can: documents, photos, SSH keys.
  • β€”A bad install leaves your system in a state that's hard to unwind.
  • β€”New laptop, new machine, new colleague β€” you set the whole thing up again.

βœ“Running it inside WebClaw

  • βœ“The launcher installs Docker (and WSL 2 on Windows) for you, then brings the whole environment up in one click.
  • βœ“The agent only sees what's inside the container. Your real files are mounted only if you choose to mount them.
  • βœ“Broke something? Reset in seconds, or roll back to a snapshot.
  • βœ“Same image everywhere β€” your machine, a colleague's, or an SSH server across the country.
Three steps

From nothing installed to writing code

01

Install the launcher

One native app for macOS, Windows and Linux. If Docker or WSL 2 is missing, it downloads and walks you through them.

WebClaw.appDockerWSL 2installed for you
02

Pick an edition and start

Lite, Desktop or Pro. Images come over a firewall-resistant CDN with resumable downloads, and ports are allocated automatically β€” no conflicts to debug.

R2 mirrordocker compose up:20000 ready
03

Open the browser and work

Three entry points behind one port: chat with the assistant, write code in the workspace, or use the full graphical desktop.

:20000AssistantWorkspaceDesktop
Under the hood

What sits behind that one port

Everything is bound to localhost inside the container and reached through a single authenticated gateway on port 20000. Nothing is exposed to your network unless you turn it on.

Processes in the container
AI assistant
Workspace (web IDE)
Desktop session
Desktop web forwarder
Web terminal
Management panel
SSH server
Container engine
Public tunnel
Clipboard bridge
Audio forwarder
Task board

Dimmed items are optional β€” enable them per instance to save memory and battery.

PortWhat it serves
20000Dashboard β€” the single authenticated entry point
20001Workspace (code-server)
20003Assistant gateway
20004Desktop (noVNC)
Capabilities

One machine, or twenty

A complete Linux, in a browser tab

VS Code-style editor with a file tree and a real terminal, a GNOME desktop for anything graphical, and a web terminal when you just need a shell. No X server, no remote desktop client, no VPN.

code-serverGNOMEweb terminal

Sandboxed by default

The agent sees the container, not your disk. Mount a folder only when you actually want it in there.

Swap the model route, not the workflow

Use the built-in WebClaw GLM route, or bring your own GLM, OpenAI or Ollama key. Each assistant can be pointed at a different route independently.

WebClaw GLMyour keyOllama

Run several at once

One instance per project, each with its own volumes, ports and AI settings. Ports are allocated from 20000 in steps of 10 and checked against everything already running, locally and remotely.

Snapshots and one-click reset

Roll a volume back to a known-good state, or wipe the instance and start clean in seconds.

SSH remote nodesPro

Point the launcher at a server and manage its containers the same way β€” start, stop, logs, image import.

Browser extension

Summon the assistant in a side panel on any page. Selected text, page content or your voice β€” straight into the context.

Mount folders on demand

Desktop, Documents, Downloads or any path you pick. Nothing is mounted unless you say so.

Offline images that get through

Multi-threaded resumable downloads from an R2 mirror, with digest comparison so you only pull what changed.

It can rebuild itself

The container carries its own source at /home/ubuntu/dna. The agent can edit it, build a new image and spawn a child container.

Expert mode when you need it

Docker-in-Docker or host socket, bridge / macvlan / external networks, named volumes or bind mounts, and an OpenWrt isolation mode that forces all container traffic through a side router.

It installs the prerequisites

Missing Docker Desktop or WSL 2? The launcher downloads the installer and walks you through it.

Editions

Take as much machine as you need

Every edition includes the assistant and the workspace. The difference is how much of a desktop and toolchain comes with it β€” pick by disk space and by what you actually do.

Lite

Basics only

Assistant and workspace, no graphical desktop. Best when disk space is tight or you only need the AI coding loop.

  • βœ“AI assistant
  • βœ“Workspace (web IDE)
  • βœ“Web terminal + SSH
  • Γ—Graphical desktop
~1–1.5 GB deployed

Pro

Full toolchain

Adds the task board and four AI CLIs. Choose it when you genuinely want the whole toolchain, not just to have the biggest one.

  • βœ“Everything in Desktop
  • βœ“Vibe Kanban task board
  • βœ“Claude / Gemini / Codex / Cursor CLIs
  • βœ“Docker-in-Docker, tunnels
largest image β€” pull only if needed
Pricing

Free to run. Pay only for the extras.

The container, the launcher and every edition are free β€” they run on your own hardware. A membership adds the things that only matter once you have more than one machine to look after.

Free
Β₯0forever
  • βœ“One instance
  • βœ“All three editions
  • βœ“Assistant, workspace, desktop
  • βœ“Browser extension
  • βœ“Restore existing snapshots
  • Γ—Creating snapshots
  • Γ—Live monitoring
  • Γ—SSH remote nodes
Download free
Most popular
Personal
Β₯79/ year

Cancel any time

  • βœ“Unlimited instances
  • βœ“Snapshots, backup and restore
  • βœ“SSH remote nodes
  • βœ“Live monitoring
  • βœ“Scheduled tasks
Do I need to understand Docker?

No. The launcher installs Docker for you if it isn't there, generates the compose file, allocates ports and starts the container. Docker is how it works underneath, not something you have to operate.

Will it slow my computer down?

It uses real CPU and memory while it runs, like any other app β€” Lite is the lightest, Pro the heaviest. When an instance is stopped it consumes nothing. You can stop and start instances from the launcher at any time.

Where does my data live?

In Docker volumes on your own machine, or on your own SSH server for remote nodes. Nothing inside your workspace is uploaded anywhere β€” the only data stored on our side is the website account data described in the privacy policy. Folders on your real disk are visible to the container only if you explicitly mount them.

Does it work offline?

The environment does β€” the editor, desktop, terminal and your files are all local. The AI assistant needs a model to talk to, so it needs either a network connection or a local model through Ollama.

What does a membership actually add?

Running many instances, creating snapshots, backup and restore, live monitoring, scheduled tasks, and managing remote SSH nodes. A single free instance with all three editions never expires.

What is the relationship with OpenClaw?

OpenClaw is one of the assistants preinstalled in the container, alongside Claude Code, Hermes and cc-switch. WebClaw is an independent third-party product and is not affiliated with OpenClaw.

Get started

Your next machine
is one download away

Install the launcher, pick an edition, and you have a full Linux workstation with an AI assistant in it β€” without touching the machine you're reading this on.

macOS (Apple Silicon / Intel)Windows 10/11Linux (deb / AppImage)