Your agents are working. This is where you watch them.

What it shows you

• What the harness already records

  The tasks and the leases on them, the doubts and the reminders waiting for you, the fleets of runners, what each subscription’s quota has left, and the closeout report of every task that finished — per repository, as the machine last reported it.

• Nothing it did not measure

  A value nothing measured reads UNKNOWN — in the database, in the API and on the page. Never a zero, which a reader would believe.

• What needs you, on your phone

  A doubt only you can settle, or a branch that has gone red, is pushed to the devices you registered, and the notification opens the page that shows it.

• What you asked for, turned into tasks

  Write what you want in Cody. A machine pulls it, a session turns it into tasks, and Cody carries back how far each one got.

• What it will cost, before anybody builds it

  Cody estimates a piece of work from the completed tasks of the repository it is for — that repository’s own history, not an average of somebody else’s.

• The keys your agents run on

  Each credential is sealed under a key of its own and handed only to the egress proxy that spends it. No read path gives one back, masked or not — and every call that proxy handled is on the record.

Whose records are whose

Every row Cody stores belongs to one project, and the database is what enforces it: row-level security, forced on every table, so a query that failed to say which project it is for returns nothing rather than another project’s records. A namespace — yours, or an organization’s — owns the projects.

Cody is a Go service on PostgreSQL. Its web UI is static files with no server of its own, so the pages you read are files and the only thing answering a request is the API.