Scoped access
Declare what an app can use
Surfaces receive only the capabilities and agents they declare. Reviews can focus on real boundaries instead of hidden imports.
Security first
AI-native systems need explicit trust boundaries.
MarkOS treats access, secrets, environments, private transport, workload identity, and policy checks as platform concerns from the beginning.
Secrets
Environment and host scoped
Secrets are encrypted, separated by environment, and materialized only where the platform needs them.
Network
Private service paths
Tailscale connects local and hosted machines so capabilities can be reached securely without exposing every service to the public internet.
Policy
Checks before runtime
Registry freshness, contract alignment, lifecycle, routes, capability use, and agent declarations are checked before changes ship.
Workload identity
Bounded machine actors
Surfaces, workers, workflows, agents, and gateways need explicit identity so calls can be authorized by principal, environment, and operation.
Provenance
Source-backed state
AI outputs should point back to source records where possible. Evidence, derived facts, memory, and agent reasoning are different records with different trust levels.
Security posture
Useful AI with bounded authority.
The practical goal is strong defaults, minimal ambient privilege, clear declarations, traceable actions, and fast feedback when a system drifts from its contract. That discipline matters most when agents can call tools, read context, send mail, update records, and act across applications.