Pato OS
A cockpit for keeping my agents in the loop. One local-first surface for my brain, my fleet of agents, and my work, with plain Markdown and YAML as the source of truth.
Why I built it
I run a lot of agents. Different repos, different tasks, all moving at once. The hard part isn't starting them, it is staying on top of them: what each one is working on, what it decided, what it learned, where it got stuck. That state was scattered across terminals and files. Pato OS is the single surface where it all lives.
Markdown is the source of truth
Everything is plain Markdown and YAML on disk: ideas, decisions, projects, agent tasks, runs, evals, learnings. No database to babysit, no lock-in. Pato OS reads those files and renders them; when you move a card, it writes the file back atomically, preserving the original formatting. The files are the truth, the UI is just a fast lens on them.
Live, not a snapshot
A file watcher tails the directory and streams changes to the browser over server-sent events, so when an agent (or I) edits a file, the board updates without a refresh. There's a small detail that matters: the writer marks its own saves so the watcher doesn't echo them back as phantom changes. The result feels live, like watching the work happen.
Three domains, one surface
- Brain — ideas, decisions, learnings, and a map of how they connect over time.
- Agent fleet — what each agent is doing, its runs, its evals, the anti-patterns it hit.
- Work — projects and tasks as a kanban I can drag, backed by the same files.
Why it matters to me
The more capable agents get, the more the bottleneck moves from doing the work to steeringit: knowing what's in flight, what decided what, what each agent learned. Pato OS is my attempt at that steering surface, personal and local-first by design, where my notes, my code, and my fleet of agents all show up in one place I actually trust.