Cavvy
Download

Cavvy for Claude Code

Claude Code, beyond a terminal tab.

Claude Code keeps working when you close the window, tells you when it needs you, and picks up its conversation after a restart.

  • It knows when Claude Code works.

    Cavvy reads what Claude Code shows: “esc to interrupt” means working, a permission question means it needs you. Background agents and shells it is waiting for keep the session on working.

  • The conversation survives a restart.

    Cavvy starts Claude Code with a session ID of its own. After the background service restarts, or with cavvy restart, it continues with --resume as long as the transcript is there.

  • Your limits, at a glance.

    Usage shows the five-hour and weekly limits, and those per model, with the login Claude Code already has. Turn it on in Usage; Cavvy asks Anthropic at most every five minutes and keeps the token in memory only.

  • The login just works.

    Sessions read the Claude Code login through /usr/bin/security, so no keychain prompt comes back. If it doesn’t work, Run Doctor says why: keychain entry, login session, a shadowing security on PATH or orphaned entries. Details are in the docs.

  • Its questions stand above the terminal.

    Cavvy gives Claude Code hooks that report every tool call. That is how a question or a permission request stands readable above the terminal while the task waits under Needs You; you answer in the terminal. Turn the hooks off under What Cavvy changes in agents.

  • It knows the rules of the task.

    In a task, Cavvy appends the task’s rules to Claude Code’s system prompt: commit on the branch, don’t merge, don’t push, and how to take in other repositories with cavvy task add. When its work is finished, Claude Code says so with cavvy done; if it stops before that, the task waits under Needs You.

  • It can give the second opinion.

    As the reviewer of a project, Claude Code reads a change in plan mode, without touching it, and lists findings by severity, file and line. See Second Opinion.

  • On your servers, too.

    Claude Code runs on Linux and Mac servers with the same status, questions, rules and resuming. Install, update and sign it in there from Settings → Agents; the sign-in shows a link to open. The limits shown are those of the login on this Mac.

Start it in a task of its own

cavvy new --worktree feature/x -- claude

Try Cavvy free