Security
Powerful agents need explicit boundaries
YouCode combines hard runtime safety checks, machine identity, connector isolation and approval-aware external actions.
- Authenticated machine identity
- One-use hidden enrollment
- Dangerous command blocks
- Approval inbox for high-impact actions
- Connector secrets kept off normal UI
Safe by default
Email sends, calendar writes, destructive file actions and external API writes require approval unless the user explicitly grants full autonomy.
Full autonomy is explicit
Users may grant more autonomy to selected agents, but hard destructive-system command protections remain.
You still supervise
AI output and agent actions can be wrong. Review important work and keep backups.