TheoB should know what can act, what needs review, and what must stop.
An autonomy layer for auto-allowed actions, review-required actions, blocked actions, human override, risk tiering, permission matching, autonomy memory, and escalation routing.
Every TheoB pathway can move through Past, Present, and Future without losing context.
Read current signals, conditions, and live context.
Voice ready
The smartest action is sometimes no action.
Selective Autonomy OS decides which actions TheoB may execute automatically, which require review, and which must be blocked. It protects the system from overreach while allowing safe, useful, reversible work to keep moving.
Low-risk actions that may run automatically, such as dashboard enrichment, non-sensitive memory notes, status refreshes, and harmless classification.
Must remain reversible, low-cost, and non-public-facing.Medium-risk actions that need human, founder, legal, evidence, or operational review before execution.
Route through workflow execution, reasoning validation, and domain-specific gates.Actions that must not execute because they involve unsafe authority, exposed secrets, legal risk, harmful claims, coercion, or irreversible impact.
Block immediately and record reason without exposing sensitive details.Allow humans to pause, redirect, correct, approve, reject, or stop autonomous actions.
Human wellbeing and founder control override automation.Classify actions by impact, reversibility, cost, sensitivity, public exposure, and authority level.
Higher risk requires stronger review and slower execution.Match action requests to agent permissions, system state, connector approval, and constitutional boundaries.
No agent may exceed its assigned scope.Remember approvals, denials, blocked actions, override events, and autonomy outcomes.
Use memory retention, privacy rules, and audit trails.Route uncertain or high-impact actions to the right gate, human, legal, recovery, or verification pathway.
When in doubt, escalate instead of improvising.