atlasAI

Every machine waits for you on its own.

Work runs on several machines at once and each of them stops now and then to ask for permission. You are the only one who can answer, so you check them one by one. The moment you leave your desk, everything stalls.

Machines find each other, the panel is one.

Machines find each other over your Tailscale network. One is picked as the server, the rest become clients. That is the whole setup — no ports opened to the outside, no central service, no account to create.

The question lands on the server, the answer runs on its own machine.

When Claude Code asks for permission on a client, the question shows up in the panel on the server. You answer once, and the answer is applied on the machine that asked.

The question is not lost, the work waits.

If you are away from your desk, nothing is lost. The work stops at a clean point and waits; when you answer hours later, the conversation carries on from where it left off. It does not time out and it does not start over.

Decisions live in files in the repo, not in a chat log.

Work is split into separate conversations per module and phase: analysis, coding, testing. Each phase writes its decisions to a file inside the repository and the next phase picks them up. Context ends up in files that travel with git, not in a chat history.