I started AI Maestro with a simple frustration. I was running several AI coding agents at once — Claude Code, Codex, and others — each in its own terminal, and I had no single place to see what they were doing. tmux is powerful, but it makes you memorize shortcuts and manage sessions by hand. AI Maestro is the dashboard on top of that reality: it auto-discovers every tmux session, so there is nothing to configure, and it shows every agent as a card with its photo, status, and role.

That sounds cosmetic until you have ten agents working at once. Then the views are the product. Zoom shows every agent at a glance. The Matrix puts every open card into four Eisenhower squares, across agents and hosts, with due dates and co-owners. The Attention page is the one I check first: overdue cards, work waiting on me, agents asking a question. Chat is not a free-form thread; questions and approvals arrive as cards I answer with one click. Team meetings put agent tiles beside a shared Kanban board with drag and drop, shared tasks, and real-time chat.

The agents also remember. Persistent memory, powered by CozoDB, means an agent learns the codebase instead of re-reading it from zero every session. A code graph shows the structure of the whole project. And the plugin ships 14 skills — memory, code graph, messaging, files, canvases, secrets, safe SQL, support cases, SEO audits — that an agent loads when the task matches.

Two design decisions matter more than any feature. First, everything is local. AI Maestro runs on your machine and stores all data locally. There are no servers, no data collection, no phone home. Your code stays yours. It is free and MIT licensed, and that is not a pricing tier — there are no premium tiers, no feature gates, no usage limits. Second, the protocols are open and usable on their own: AMP for agent messaging, AID for agent identity, AAP for agent actions, AFP for agent files. If you only need one of them, take it.

The newest piece is the one I am most excited about: a world view. Instead of a list of cards, you see your agents as characters in a 3D place, grouped by organization and host — the way a strategy game shows a map. When you run 160 agents across several hosts, a list stops working. A place does not.

One command installs it, and the dashboard opens on localhost:23000. Or install just the Claude Code plugin. And if you want a working chief of staff on day one, deploy Lola — that is the next piece.

Links: ai-maestro.23blocks.com, github.com/23blocks-OS/ai-maestro