Four layers: connectors feed a live company state; a policy engine gates a graph of specialist agents; and an immutable ledger records every step. This page is the map.
Connectors pull from the systems you already run and compile one live company state. The GitHub connector is live today — try it on the demo page against any public org and watch real data flow in.
Six verdicts — allow, allow-with-limits, one approval, dual approval/multisig, simulate-only, deny. Applied by code, not by model judgment. Autonomy tiers are assigned per decision class based on risk and reversibility.
Decomposes goals, routes work between agents, escalates exceptions to humans.
Maintains company state, runs scenarios, proposes allocation. Never decides alone.
PR triage, test runs, staged deploys. Prod requires policy clearance.
Roadmap drafts, PRDs, experiment design. 24h human veto window.
Runway tracking, payment drafts. Any payment above cap → dual approval.
Content and campaigns under a hard spending cap. Fully audited.
Pipeline hygiene, proposal drafts, follow-up sequencing.
Risk flags and contract redlines. Never signs anything.
Permission audits, anomaly flags, kill-switch authority.
Capacity planning, hiring recommendations. Humans hire and fire.
Append-only, immutable. Every recommendation, approval, execution and outcome. Rejections with reasons become training data — the evidence that unlocks the next tier of autonomy.