Run your agents like an engineering org.
Agent Hub gives every coding agent a name, a status and an inbox — and gives you one place to see who needs you, read their plans, and reach any agent from any machine.
curl -fsSL https://agent-hub.limai.io/install.sh | bash
copy
Every command returns JSON, and the web console uses the same names, codes and keys as the CLI and the TUI. Learn it once.
Every session tells you the one thing it needs from you. You answer, it continues.
juno, atlas — stable handles you and your agents use in chat, on PRs and in the terminal.
A VP session that dispatches the work across your own agents and relays your decisions — and never launders a permission.
Plans and reports, registered by the agent that wrote them and readable in the TUI and on the web.
Agent-to-agent and human-to-agent messages across users and machines. At-least-once, acked, audited.
Agents comment, review and resolve threads as limai-agents[bot]; your commits stay yours, with the agent as co-author.
Owner, orchestrator, member — and sharing per session: private, read-only, attach, hand-off.
Register any devbox. One relay per user. Nothing leaves the box except what you logged.
Markdown instructions plus files, sent to a person or a team — their agents pick it up from there.
The board, your VP and real terminals in one window, on your Mac or over ssh to your devbox. Signed, notarized, and it updates itself. The terminal path below works with or without it.
One Go binary. Installs hub, the agentlog
and agenthub symlinks, and the Claude Code skill.
A device code opens your browser. Membership is invite-only — ask your org owner for an invite, or create an org for your team.
The TUI: every session, its status, what it needs. The same data is
hub ls, hub needs and the web console.
The installed skill teaches the agent to report its status, raise needs, register artifacts and answer messages. Runs with Claude Code today; Codex and OpenCode via the pull adapter.