Agent Cockpit
A control room over an AI setup that outgrew its chat windows — every agent, session, decision and provider in one view.
Why I built it
I run AI agents for real work, and after a while the work was spread across chat threads with no way to see it whole. I could not answer simple questions like what is running, what did it do last week, and what did it cost.
The problem
Agent work happens in conversation. Every transcript is a scrollback, every decision is buried in a thread, and the only person who can find anything is the one who remembers where it was. That does not scale past one person with a good memory.
What I was using before
Chat windows and a memory. Which meant re-asking the same questions, losing the reasoning behind past decisions, and having no idea what anything cost until the invoice arrived.
What it does
- Live state of every agent and session in one place
- A browser for past sessions — flip through transcripts and find when a decision was made
- Read-only context from my own notes, so I can see what a session touched
- Provider health, credits and usage in the same view
- Interventions and an audit trail, so automation stays accountable
Where it is now
Built and in daily use. It is the reason several of the other things on this list exist — most of them were built by agents I can now watch working.
Who it might interest
Anyone running agents for something that matters, who has discovered that a chat window is not an operations console. It is the least copyable thing here, because it is shaped around how one specific person works.