agentmash
The multiplayer pitch is what grabs me. Most coding agents today are built around one developer staring at one assistant, which means the AI only ever sees a single person's context. Agentmash flips that to let a whole team's agents share the sandbox, and for codebases where work moves across people, that could be the difference between an AI that actually helps and one that just gets in the way.
The open-source part matters just as much. Teams get to see what their agents are really doing, patch the knobs, and host it themselves if they want. That's a real advantage over locked-down commercial tools, especially for teams with compliance constraints or a healthy distrust of black boxes.
I'd want to see how the agents coordinate without stepping on each other, because concurrent edits are where multiplayer tools usually break. But the direction feels right, and the fact that it's open means we'll find out fast.