🦌 deer
Run multiple unattended Claude Code agents safely in parallel.
deer is a TUI dashboard for launching, monitoring, and managing multiple Claude Code agents at once. Each agent runs in its own sandboxed git worktree with full filesystem and network isolation.
deerbox is the standalone CLI that runs a single sandboxed agent. deer uses deerbox under the hood, but you can also use deerbox directly in scripts or when you only need one agent at a time.
Goals
- Parallel agents -- launch as many agents as you want from a single dashboard.
- Safe
--dangerously-skip-permissions-- every agent is sandboxed (filesystem, network, credentials), so unrestricted mode is actually safe. - Uses your Claude subscription -- no separate API key needed. deer picks up your existing Claude Code login automatically.
- Feels like
claude-- attach into any agent's tmux session and interact with it directly, just like the regular CLI.
How it works
- Launch
deerinside a git repo. - Type prompts -- each one spins up a new agent in an isolated worktree.
- Monitor agents from the dashboard, or press Enter to attach into one.
- When an agent finishes, press
pto create a PR.
That's it. No config files required, no orchestration layer to learn.
Using deerbox standalone
If you only need a single agent, you can skip the dashboard entirely:
deerbox "refactor the auth module to use JWT"deerbox handles worktree creation, sandboxing, and cleanup on its own. When the agent finishes, an interactive menu lets you create a PR, keep the worktree, open a shell, or discard the work.