I built this because I kept losing track of my own agents. Four terminals open, three of them working, and no idea which one was burning money or which had been stuck on the same tool call for eight minutes. I would find out afterwards, from the bill or from a diff I did not expect. So agentglass shows all of them at once: what each is doing right now, what it has cost, how close it is to compacting. Then it turned into the place I actually work from, because once you can see the fleet you want to act on it. Review the diff, drive git, review a pull request, answer a chat, and hold a destructive command until you say go. Three things I care about more than the feature list: It runs entirely on your machine. Localhost by default, no account, no telemetry, nothing leaves. The security model is written down in the README in plain language, including what it does not protect you from. It is provider-neutral on purpose. Claude Code and Kimi arrive through hooks, everything else through OpenTelemetry, and adding a provider is a pull request rather than a roadmap item. Two of the providers in there were contributed by people who wanted them. It is three weeks old. That means it moves fast and it also means you will find rough edges. The macOS builds are unsigned, so Gatekeeper will complain on first launch and the README tells you the one command that fixes it. Free, MIT, and the roadmap is just the issue tracker. If you try it and something is wrong, tell me here or open an issue, I read all of them.

agentglass

Mission control for every AI coding agent on your machine

Mission control for every AI coding agent on your machine