Documentation

Enforcement

Mode matrix per layer — adopt the secure baseline before expecting denies.

Recommended first

Default install is observation-friendly. Before expecting denies on npm or exfil, adopt the secure baseline and run the scenario labs.

Mode matrix

Layerpolicy: monitorpolicy: enforceApp: observation mode
Network Extension
Log flows; allow connectionsDrop blocked domains/IPs & behavioral gatesWorkflow drops suppressed; integrity paths may still filter
Endpoint Security (sync)
Allow + log all execution attemptsProcess governance, file access control, and supply-chain gating deny unauthorized actionsWorkflow rules simulate; platform integrity still enforces
Behavioral engine (async)
Score & graph; optional notificationsQuarantine, verdict propagation, human-in-the-loop controlsSame scoring; enforcement simulated in UI

Verdict bridge

Behavioral scores propagate to the sync enforcement layer via a zero-copy bridge. The Endpoint Security extension reads enforcement decisions on the next system call; human-in-the-loop controls suspend execution pending user approval.

Plane 1 · Synchronous

System extension

  • Process governance · file access control
  • Supply-chain gating · command-line detection
  • ~2ms budget — blocks before launch
3 governors13 patternsSupply-chain gate

Plane 2 · Asynchronous

Behavioral engine

  • Ingests telemetry from system extensions
  • Builds causal graph + runs detection rules
  • Propagates verdicts to sync layer via zero-copy bridge
25+ rulesSession replayHuman-in-the-loop
Zero-copy verdict bridge between enforcement planes

Synchronous (system extension)

  • 13 built-in command-line detection patterns
  • Intent-based policy enforcement and file access governance
  • Supply-chain gating on package manager intercept (when enabled)
  • Cross-correlation of sensitive file reads with subsequent execution

Asynchronous (behavioral engine)

  • Graph-based behavioral detection rules
  • Session quarantine and smart notifications
  • Anonymized trace sharing for SOC review

Secure baseline · Behavioral detection · Evaluation · Policy

Running AI agents on Mac at scale? We'll tune policy with you.

Design partners →