Bounded autonomy · governance-cute

Agents that don't go rogue.

Every action is a proposal. Tier-gated, blast-radius scored, fully reversible. When the operator is new or stuck, autopilot takes over and runs the lifecycle for them — and learns who they are so next time it doesn't have to ask.

STATUS · ARMEDTIERS · 4REVERSIBLE · 100%
See the loop ↓Configure thresholds
scope · bounded autonomypowered by ESAP AI
The loop · 5 stages · always reversible

Every action takes the same path.

Propose → tier-gate → review → execute → reverse. Watch one full cycle play through — or click any stage to jump in.

STAGE · 01 · PROPOSE

A proposal materializes.

APO drafts a candidate action: which entity, what change, why now. The reasoning trace types in below it line by line — no hidden steps.

PROPOSAL · DRAFTlive · 3.2s

Lower reorder threshold by 12% on 3 SKUs

  • ▸ demand softened · Riyadh South
  • ▸ 30-day trend stable (±0.4%)
  • ▸ reversal cost capped at $1,840
  • ▸ 4 sources cited POSERPFCAST
BLAST32
$ COST$1,840
Policy · live preview

Your thresholds, your rules.

Move the sliders. APO will use these gates from the next proposal forward. Blast radius is what gets touched; financial impact is what gets risked.

01AUTO
02APPROVAL
03MANAGER
04ADMIN
Live preview
What happens if blast radius =?Routes to APPROVAL.
0% of mutations escape audit·100% reversible·∞ thresholds configurable

Plug APO in read-only. Promote it later, on your terms.

Every pilot starts with zero write scopes. You watch APO propose, you approve manually, and you upgrade specific lifecycles to auto-tier only when you're comfortable.

Start a pilot →Read the architecture brief