If you're running multiple AI agents across several machines, you know the pain. Fourteen terminal tabs. SSH into each box. Alt-tabbing to check if Claude finished its task. You know the pain of deploying and managing AI agents across several machines, but 49Agents, a new open-source project, consolidates that mess by putting terminals, agents, and files onto a single zoomable 2D canvas.
The setup is straightforward. You run a `49-agent` process on each machine. It connects via WebSocket to a relay server, and you interact with everything through a browser. No SSH required. The canvas shows live CPU, RAM, and Claude API usage across all connected machines as a HUD overlay. You can drag panes around, zoom in to focus or zoom out for the big picture, and the layout sticks between sessions. It even works from a phone.
Broadcast input lets you type a command once and send it to multiple terminals at the same time, similar to how Imbue runs 100+ Claude agents in parallel to achieve distributed execution. The Monaco editor (the same one powering VS Code) lives right on the canvas so you can edit files without switching contexts. There's also Beads, an interactive issue tracker built on Dolt, a version-controlled SQL database. Terminal I/O is relayed through their server but never stored, which matters if you're working with sensitive codebases.
The project uses BSL 1.1 licensing, free for individuals and small teams, converting to MIT in 2030. You can self-host the entire stack with a quick `git clone` and `./49ctl setup`, or use their hosted version at 49agents.com. That 2030 MIT conversion is the interesting part: in four years, the code becomes fully open source whether the company still exists or not.