Claudible

Claude Code, multiplayer.
Share your live session with a link. Your team watches, talks, and co-drives: any browser, any device, even a phone.

---

Claude Code is one person, one terminal. Claudible makes it a team sport:

• 🤝 Live multiplayer. Invite anyone into your running session: they watch, chat, or take the keyboard and co-drive the same Claude.
• 🎙️ Voice, 100% local. Talk to Claude, hear it answer. Whisper + Kokoro run on your machine, not in a cloud.
• 📊 Live telemetry. Context, cost, and tokens at a glance, with a one-tap /compact before you run out of room.
• 🛰️ Agents cockpit. Watch subagents and workflow swarms work in real time: task, tools, token burn, which model runs each agent.
• 🕘 Per-prompt history + revert. Every prompt gets a git-backed snapshot. Roll the code back to any of the last 10, and undo the revert.
• 🔒 Nothing hosted. Invite links tunnel out from your machine, every guest waits in a lobby until you approve them by name, and the real Claude Code TUI stays untouched (xterm.js, full scrollback).

Claudible is in public beta: expect fast releases and the occasional rough edge; issues are very welcome.

Get started

🪟 Windows: download and run

1. Grab the latest Claudible-Setup-<version>-x64.exe from Releases.
2. Run it. It's unsigned during beta, so SmartScreen shows "Windows protected your PC": click More info → Run anyway once.
3. Launch Claudible from the Start menu. First run sets up local voice by itself (a few hundred MB of models; resumable if interrupted).

Already needs to be on the PC: Claude Code for Windows (run claude once in any terminal and sign in), Node.js 22.12+ (nodejs.org), and Git for Windows (git-scm.com). Native-Windows voice is the newest path; if it snags, logs are at %USERPROFILE%\.claudible\logs\ and the WSL build below is the proven one.

🛠️ Build from source

Maturity, honestly: Windows + WSL2 is the battle-tested path, Linux is verified, native Windows and macOS are code-complete with runtime smoke tests in progress.

🪟 Windows 11 + WSL2 · recommended

You need WSL2 with a signed-in Claude Code inside it (claude on your PATH), plus Node.js 22.12+ on Windows. Run this in Windows PowerShell, not inside WSL (Claudible is a Windows Electron app that embeds Claude Code running in WSL):

$dir = (Read-Host "Install folder for Claudible (Enter for default) [$HOME\claudible]").Trim().Trim('"'); if (-not $dir) { $dir = "$HOME\claudible" }; if (!(Get-Command git -ErrorAction SilentlyContinue)) { winget install --id Git.Git -e --source winget --accept-source-agreements --accept-package-agreements; $env:Path = [Environment]::GetEnvironmentVariable('Path','Machine') + ';' + [Environment]::GetEnvironmentVariable('Path','User'); if (Test-Path "$env:ProgramFiles\Git\cmd\git.exe") { $env:Path = "$env:ProgramFiles\Git\cmd;$env:Path" } }; if (!(Get-Command git -ErrorAction SilentlyContinue)) { Write-Host "`n[!] Git for Windows is required and couldn't be auto-installed (no winget on this system?). Install it from https://git-scm.com/download/win, reopen PowerShell, then run this line again." -ForegroundColor Yellow } else { git clone https://github.com/synthetiks/claudible "$dir"; if ($LASTEXITCODE -ne 0) { Write-Host "`n[!] git clone failed -- see git's error above. Likely causes: no internet/proxy blocking github.com, or `"$dir`" already exists and isn't empty. Fix that, then run this line again." -ForegroundColor Yellow } elseif (Test-Path "$dir\install.ps1") { powershell -NoProfile -ExecutionPolicy Bypass -File "$dir\install.ps1" } else { Write-Host "`n[!] install.ps1 is missing after clone -- your antivirus most likely quarantined it (a false positive). To fix: allow/exclude the folder `"$dir`" in your antivirus, then run these two lines:`n    git -C `"$dir`" restore install.ps1`n    powershell -NoProfile -ExecutionPolicy Bypass -File `"$dir\install.ps1`"`nPer-antivirus steps are in `"$dir\SETUP.md`" (the Antivirus section)." -ForegroundColor Yellow } }

One line: asks where to install, auto-installs git via winget if missing, clones, installs everything, and launches. First run only: the voice build compiles whisper.cpp and downloads ~480 MB of models (10–20 min; rerunning the line resumes). Voice build deps inside WSL (git cmake build-essential ffmpeg espeak-ng python3 curl + uv): npm run setup checks and prints the exact apt line. No winget? Install Git for Windows manually, reopen PowerShell, rerun. Full troubleshooting: SETUP.md.

Prefer to run the steps by hand?

git clone https://github.com/synthetiks/claudible
cd claudible
npm install        # deps + node-pty's N-API prebuilt + the small JS patch (no compiler needed)
npm run setup      # builds the WSL voice services (installs their deps; ~480 MB models on first run)
npm start          # opens the cockpit

Optional Desktop shortcut: powershell -NoProfile -ExecutionPolicy Bypass -File launch\make-shortcut.ps1

🪟 Windows native (no WSL) · new

Needs Windows Node 22.12+ and Git for Windows (bash.exe). Installs Claude Code for Windows if it's missing.

git clone https://github.com/synthetiks/claudible
cd claudible
powershell -NoProfile -ExecutionPolicy Bypass -File .\install.ps1 -Native

Downloads the prebuilt whisper-server.exe + speech model and sets up Kokoro via uv (no Visual Studio, Python, or cmake), then pins CLAUDIBLE_RUNNER=win (remove that user env var to fall back to WSL). Run it through powershell -ExecutionPolicy Bypass -File exactly as shown; Windows' default policy blocks .\install.ps1 directly. Clone into your home or projects folder with a normal, non-admin PowerShell.

🐧 Linux · verified backend

Needs Node 22.12+, a C toolchain (build-essential + python3; node-pty compiles on install), and Claude Code on your PATH (claude, signed in).

git clone https://github.com/synthetiks/claudible
cd claudible
npm install            # builds node-pty for Linux (needs build-essential + python3)
bash setup/setup.sh    # whisper.cpp + Kokoro (~480 MB models on first run)  ·  NOT `npm run setup` (that's the Windows→WSL wrapper)
npm start              # auto-selects the native Linux backend (no WSL involved)

Packaged installers build with npm run dist:linux (AppImage + .deb).

🍎 macOS · new

Needs Node 22.12+, Claude Code on your PATH, Xcode Command Line Tools (xcode-select --install), and brew install cmake ffmpeg espeak-ng.

git clone https://github.com/synthetiks/claudible
cd claudible
npm install
bash setup/setup.sh    # detects Homebrew for the build deps  ·  NOT `npm run setup` (Windows→WSL only)
npm start

A signed .dmg is the planned release artifact (npm run dist:mac). Runtime smoke pending (needs a Mac).

Live sharing, in 20 seconds

Click Share: Claudible starts a tiny server on your own machine and hands you a private link. With cloudflared installed it's a free https://<random>.trycloudflare.com tunnel that works across networks; without it, localhost/LAN only. Up to 8 guests open the link, wait in a lobby, and see nothing until you approve each by name. Per share you pick read-only (watch + chat) or interactive (type into the same session). Guests get a chat, a voice room, and a mobile-friendly viewer. The link is ephemeral: it dies the moment you stop sharing, and New link locks out anyone holding the old one. A guest who can't reach the link is almost always a network or VPN issue; a fresh link or a phone hotspot sorts it.

Two things worth knowing before you share:

• Share live in a synced project doesn't just mint a link; it also advertises it on the project's claudible/sessions branch so teammates see a ● LIVE badge and can join without you pasting anything. Anyone with git access to that repo can read the link (they still wait in your lobby). Want exactly one invitee? Use plain Share and send the link yourself.
• Turning on session sync for a repo-backed project commits full transcripts to that private repo: prompts, code, tool output, anything Claude read, plus your name, gh login, git email and machine. Every collaborator with git access can read them. Disclosed in-app before you enable it, and off until you do.

Security & privacy

• Permissions are yours. The embedded Claude Code prompts before running tools, exactly like the normal CLI; Settings can remember Accept edits or Bypass permissions for a one-click/voice flow. A session synced from a collaborator is always sandboxed (normal prompting, never auto-resumed), regardless of your setting.
• Voice is local; Claude Code is not. Whisper and Kokoro never leave your machine and Claudible sends no telemetry. The embedded Claude Code talks to Anthropic exactly as the normal CLI does.
• Interactive guests type into your session, which may run with skipped permissions. Grant it only to people you trust, prefer read-only otherwise, and rotate or stop the share when you're done.
• The voice services bind 0.0.0.0 so the Windows app can reach them over the WSL2 NIC; on WSL2's default NAT networking they aren't exposed to your LAN. Full threat model and disclosures: SECURITY.md.

Platform support

Every OS-coupled call goes through one Runner seam (runners/{wsl,posix,win}.js); the script fleet is Python-free, so the only per-OS runtime is Node + a shell.

Platform	Status
Windows + WSL2	Battle-tested; the recommended path
Linux	Native backend live-tested; AppImage + .deb build in CI
Windows native (no WSL)	Code-complete; runtime smoke in progress
macOS	Code-complete; needs a Mac for smoke + signing

Deep dive: docs/ARCHITECTURE.md.

Staying up to date

• Clone installs update themselves. An Update & restart chip appears when a newer commit lands; one click pulls, reinstalls deps if they changed, and relaunches (install.ps1 -NoUpdate opts out). Packaged .exe installs get a notice-only "newer release available" toast.
• In a live session, a peer on a different commit shows a build-skew note on their ● LIVE badge, so "did you restart onto the update?" answers itself.
• Optional sub-second presence: by default "went live" reaches collaborators in a few seconds over the shared git branch; a team that wants faster can self-host the tiny Cloudflare Worker in relay/. Off unless configured. See relay/README.md.

Configuration (env vars)

Env var	Default	Meaning
CLAUDIBLE_WHISPER	http://localhost:2022	STT endpoint (any OpenAI /v1/audio/transcriptions)
CLAUDIBLE_KOKORO	http://localhost:8880	TTS endpoint (any OpenAI /v1/audio/speech)
CLAUDIBLE_VOICE	~/.claudible/voice	where the local voice services are installed
CLAUDIBLE_CLOUDFLARED	(auto-detect)	path to a cloudflared binary for cross-network co-work tunnels
CLAUDIBLE_WHISPER_PORT	2022	port services.sh binds Whisper on
CLAUDIBLE_KOKORO_PORT	8880	port services.sh binds Kokoro on

Claudible speaks the OpenAI audio API, so any compatible STT/TTS works (LM Studio, your own server). If you change a port, set the matching URL too.

Uninstalling

Windows: uninstall from Settings → Apps. Left behind on purpose so a reinstall keeps your work: ~/.claudible/ (projects, settings, session history) and ~/.claudible/voice/ on the WSL/Linux side (models + build, several hundred MB); delete them by hand to reclaim the space. The Whisper (:2022) and Kokoro (:8880) helpers can keep running after quit: pkill -f whisper-server; pkill -f kokoro (or close the WSL session).

Linux/macOS: delete the cloned repo and ~/.claudible/, then stop the voice services the same way.

License

MIT. Third-party components bundled in the installer are listed in THIRD-PARTY-LICENSES.md.

---

Built by the_crazydev with Claude.