__NAME__

Run several Claude Code sessions in parallel from any browser, phone included, and see at a glance which one is working, which one is waiting for your permission, and who is touching what.

Staged demo: synthetic conversations for a made-up acme-api project and real slog commands; the state events were fired by hand.

The idea: give Claude Code its own machine. Run it on a VPS or a sandbox box, not on your laptop, let it work with broad permissions there, and drive it from anywhere: your desk, your phone, a train. The worst a session can break is that box, which is what makes long unattended runs and auto mode reasonable. The control room is how you watch and steer it.

Three layers, one idea: several agents working on the same project without stepping on each other.

What you get:

• A chat view on every tile, with the terminal still underneath. Each session reads like a chat app (markdown, tables, formulas, diagrams) with every tool call folded into one line showing the real command. A permission request is a card with exactly what Claude wants to run and one button per option. Paste or drop any file into the message box; what Claude hands you back shows up as one-tap Download and Copy cards. One tap on Term and the real terminal is there, never restarted. See The Chat view.
• Live state per session: working, idle, or waiting for your permission. It comes from Claude Code's own hook events, not from screen-scraping, so it doesn't break when the CLI's UI changes. A red badge in the top bar takes you straight to the session that needs you, and it clears whether you answer yes, no or Esc (no hook fires on a "no", so the server confirms it on screen; see below).
• slog, a shared board with hierarchical locks. Every session can see what the others are doing and claim a resource (api) or just part of it (api:auth). Each tile shows the locks its session holds. Locks are advisory by default; tie one to files with --paths and a hook enforces it: other sessions get their edits to those files denied.
• Every session starts informed: a SessionStart hook injects the live locks, the recent feed, the files other sessions just changed, the project's notes (done / next / don't redo) and how to hand you files and clips in the Chat view into each new session, also after /clear and compaction. Nobody has to remember to run slog status.
• History by session: every file a session writes is committed on its own, labeled with the session and its transcript id, in a shadow git repo that never touches your project's .git. cr-hist changed answers "what did the other sessions change?", cr-hist who <file> leads to the conversation that made a change.
• Mode and plan usage at a glance: each tile shows Claude Code's permission mode (auto / accept edits / plan / manual; tap it to switch, like Shift+Tab) and how much context the conversation uses. The top bar shows your plan's 5-hour session and weekly usage with their reset times, read from a silent status line.
• Built for the phone: the Chat view is plain web text, so you select and scroll with your finger; in Term, an on-screen key bar (Esc, arrows, Tab, Ctrl-C), swipe to scroll history, tap-to-jump between sessions. See On the phone.
• 📋 One-tap clip: the agent puts exact text (a command, a token, a draft) in the clip with cr-clip and you copy it with one tap. The ⌄ next to 📋 keeps the last four, since every session shares one clip.
• 🎤 Dictation (optional): speech → Whisper → typed into the prompt without pressing Enter, with a hallucination filter based on Whisper's per-segment metrics.
• Skills that survive the session: a 3-layer structure (router / current state / decision log), a skill-sync skill that consolidates each session's findings before it closes, and skill-lint to catch what parallel sessions break (duplicate changelog ids, dead pointers, bloated routers).

In action

Staged like the demo above: synthetic conversations, rendered by the real UI.

No frameworks, no build step, no npm install: Python standard library, bash, and one HTML page with its script and stylesheet.

Community project, not affiliated with or endorsed by Anthropic. "Claude" and "Claude Code" are trademarks of Anthropic.

---

Try it in one click

The Try it button above opens it in GitHub Codespaces: GitHub builds a container with tmux, ttyd and Claude Code, starts the panel and opens it in your browser. The forwarded port is private, so you get authentication for free, and the free quota covers dozens of hours a month.

Open it in a browser tab (🌐 in the Ports tab): VS Code's built-in preview can't show it, because the panel refuses to be embedded in other pages (that's what keeps ot