Govern · The Console
Govern how software gets built — every change on one trail.
Your agents and operators make changes all day. The Console is where you draft them, hold them at a gate, ship them, and roll them back — each one recorded with who authorized it and when. One audit trail for everything humans and agents do.
Already running agents? The ceo command line feeds the hub. No new dashboard to babysit.
The Loop
Draft, gate, ship, roll back. In that order, on the record.
Every change moves through the same four postures, whether a person or an agent proposes it.
- 01
Draft.
A change is proposed — a config edit, a flag flip, a deploy. It's staged, not live. Nothing has happened to production yet.
- 02
Gate.
The change is held at a gate until its conditions are met. A manual gate waits for a named operator; a policy gate waits for the rule. Gates hold — they don't catch anyone, they just don't let the change through until it's allowed to pass.
- 03
Ship.
Once the gate clears, the change applies — and the trail records what shipped, when, and on whose authority.
- 04
Roll back.
If a shipped change goes wrong, you reverse it from the same surface that shipped it. The rollback is itself a recorded change, not a panic button that erases history.
draft → gate → ship → roll back · one posture per node · every transition timestamped
LOADING EPISODE ···
recorded straight off the run · agent = channel · run = episode · every beat evidence-linked
The Evidence Chain
The trail isn't a claim. It's the evidence.
Here is the question a head of platform in a regulated org actually has to answer: "We have AI agents making changes, and I can't prove what they did, when, why, or on whose authority."
The Console answers it with evidence, not assertion. Every change carries its intent binding (what it was for), its posture (where it sits in the loop — drafted, held, shipped, rolled back), and its evidence chain (the linked record of what actually happened). Nothing is "approved" in the abstract. A change is authorized by a named operator at a specific time, and that line is part of the record you can read back.
This deploy was authorized by Priya at 09:03.
Here is the evidence chain.
When you need to reconstruct what happened — for an incident, an audit, or just to understand a Tuesday — you don't reassemble it from memory and Slack scrollback. You read the trail. The evidence was recorded as the work happened, not written up afterward to flatter it.
Evidence over assertion. If it isn't on the trail, we don't claim it.
For teams roadmap · not shipped
When your house style needs to hold across machines.
Today, the Console governs the changes your operators and agents make. Next, it learns the way your team builds — so the conventions your senior engineers carry in their heads govern every machine, not just the ones they're sitting at. We call that Team DNA: shared, cross-developer build style, synced and governed through the hub. New hires start where your seniors left off, and "why does every team's code look different" stops being a question you have to ask.
Team DNA cloud sync is on the roadmap, not in the product yet. We'll say so here the day it's live.
The Horizon later · the category being mandated into existence
Where this is going: provable build provenance.
As autonomous agents take on more of the work, "what did it do, and who let it?" stops being a nice-to-have and starts being something regulators ask for. The same trail that governs your changes today becomes build provenance — the verifiable record of what an autonomous agent did, when, why, and on whose authority. That's a longer conversation, and a later one. We're not selling it today. We're building the substrate it stands on, in the open, change by change.
The Console is where you drive. Start by pointing ceo at your builds.