Autonomy Settings
Set the maximum autonomy level for AI agents, cap the money they can move, force approval on high-risk cases, and tune autonomy per action category.
Autonomy Settings (/settings/autonomy) is where you "Configure AI agent autonomy levels and
safety boundaries" — the operator controls behind the L0–L3 model explained in
the email-agent autonomy concept. Related monitoring screens
sit alongside it: the Exception Inbox and the Trust & audit
views.
Global Policy
Global Policy ("Set the maximum autonomy level for all agents in this tenant") holds the master dials:
- Max Autonomy Level — a select with four options: L0 — Observe Only, L1 — Suggest + Approve, L2 — Auto Low-Risk + Notify, and L3 — Bounded ODD Autonomy. It is a ceiling for every agent, and it saves the moment you change it.
- Auto-Execute Max Value ($) and Invoice Auto-Send Max ($) — money caps above which an action always needs a human. Edit the number and click away to save.
- Hard Override Toggles — four switches that force operator approval no matter the level: Require Approval for DG Shipments, Require Approval for Sanctions Matches, Require Approval for New Customers, and Require Approval for New Lanes. Each saves as you flip it.
Per-Action-Category Levels
Per-Action-Category Levels ("Override autonomy level for specific action types") gives each action category (classify email, extract shipment data, send customer invoice, screen sanctions, and so on) its own L0–L3 grant. The effective autonomy is always the lower of this grant and the global cap. Unlike the global level, these are staged — a Save Changes button appears with the count of pending edits; click it to commit.
Risk Band Overrides
Risk Band Overrides ("Adjust risk bands for specific customers, lanes, or cargo types") lists any configured overrides with a delete action; an unconfigured tenant shows "No risk band overrides configured."