prastik

Interesting, what is the timeline for the wrapped feature? Is it yearly? Or is the window shorter?

SwarmWatch is an activity monitor and control plane for AI coding swarms. It runs as a tiny always-on-top overlay with a Tamagotchi pet avatar that shows what your agents are doing and when they need you.

Right now it works with Cursor, Claude, Cline, GitHub Copilot and VS Code via small per-agent shims. The shims call a local runner, which streams events (tool calls, file writes, shell commands, etc.) into the desktop app over a local WebSocket. Everything is fully local; there’s no server or external telemetry.

The interesting part (at least to me) is the approvals layer. You can mark certain actions as “guarded” (for example: shell commands, destructive file operations). When an agent hits one of those, the runner pauses, the avatar flips into a “blocked/needs you” state, and an approvals pane pops up. From there you can approve or reject the action with one click, and the runner immediately continues or aborts the step. Also, you have a transparent Audit Trail of what all happened and when exactly, and other basic teleemry.

It's open-sourced, feel free to try it out or fork it and use it. Thank You!

That's a really good suggestion, honestly haven't thought about it yet. But yeah, it might evolve into being a teammate if we start adding other MCP servers. But thanks for the suggestion. Appreciate it!

Thank you. Do reach out if you need any help with the integrations!

Not being able to add “one more domain” is an issue, but how does wsgrok ensure the domain link remains consistent after each server restart? That seems like a bigger problem since you’d otherwise have to update the domain across workflows every time the application restarts.