Manage tasks,
not agents.
Tell your AI agents what to build. Go do your own work. Come back to PRs ready for review.
Captain
The backlog that
works itself
Describe a task. Captain spins up a Claude Code worker in its own worktree and watches it. When the PR is ready, you review and merge. Or send it back with a note.
Digest
Your reading list,
wired to your backlog
Paste a URL. Mando fetches it, extracts the content, and summarizes it. See something worth acting on? It becomes a backlog task. Captain takes it from there.
Sessions
Full history.
Self-improving agents.
Every Claude Code session is logged with full transcripts and tool calls. Captain keeps a decision journal recording why it nudged, restarted, or escalated each worker. When the same failure keeps showing up, a distiller extracts the pattern and writes it into the project guardrails.
FAQ
What AI agents does Mando work with?
Claude Code. Each task gets its own headless Claude Code session running in an isolated git worktree. You don't talk to the agent. Captain does.
Do I need to keep Mando open for agents to work?
No. The Rust daemon runs under launchd in the background. Workers keep going whether the app window is open or closed. Open Mando when you want to check in, not to keep things running.
How much does it cost to run?
Mando is free. You pay Anthropic for the Claude API usage. Bring your own API key.
Can I use it with multiple repos?
Yes. Add as many repos as you want. Each project gets isolated worktrees and its own context preamble. The backlog shows everything in one list, filterable by project.
What happens when an agent gets stuck?
Captain watches for it. No output for 8 minutes? Nudge. Tool hanging for 15 minutes? Restart. Still stuck after that? It flags the task for you. The thresholds are configurable.
Is it open source?
Not yet. We plan to open source it. The repo is on GitHub already, just private for now.