For in-house teams
Give your agent the account. Keep the controls.
Your growth team wants the agent moving budgets and fixing tracking. Your finance and security teams want proof it can't go wrong. Ducto is how both get a yes.
Need → mechanism
What you need, and the thing that enforces it
Finance wants a hard ceiling on spend changes
Caps from the policy row
Budget and bid ceilings live in Postgres, server-side. The model never sees them and cannot raise them — a hijacked prompt changes nothing about what executes.
Security asks what happens when the model is wrong
Preview-before-apply
Every mutation dry-runs first, and apply requires a live preview whose fingerprint matches the exact change. Wrong or mutated arguments simply don't execute.
Some changes should get human eyes
Approval on the classes you flag
Route big budget moves and structural changes to an approval inbox. Everything else executes instantly within caps — approval is a scalpel, not a toll booth.
The CMO asks “what changed last week?”
Hash-chained audit
An append-only ledger records every tool call — who, what, which policy allowed it. No role can edit it, including us. Filter, export, answer in minutes.
Rollout
The path from zero trust to real autonomy
No big-bang grant. Each stage is a policy setting, not a leap of faith.
Week 1
Read-only
Connect accounts by OAuth. The agent reports, audits, and answers questions. Zero write risk.
When ready
Writes within caps
Set budget and bid ceilings in the policy row. Routine changes execute instantly inside them.
Ongoing
Approval on what matters
Flag the tool classes that deserve human eyes. They wait in the approval inbox; the ledger records everything.
FAQ
The questions this page raises
Can we roll this out gradually?
What does the security review need from us?
Start read-only. Expand when the veto-holders nod.
Free forever, read-only, every platform. Plus 14 days of the full product. No card required.