permissions

Security is a primary design goal of Andromity. Because AI agents can execute commands and modify files, Andromity provides four distinct permission modes to give you full control over what happens on your machine.

graph LR
  SAFE[SAFE: Max Verification] --> TRUST[TRUST: Auto File Edits]
  TRUST --> FULL[FULL: Auto Shell Exec]
  FULL --> YOLO[YOLO: Fully Autonomous]

• Write Files: Requires explicit confirmation.
• Edit Files: Requires explicit confirmation.
• Shell Commands: Requires explicit confirmation.
• Recommended For: New codebases, untrusted models, or complex changes where you want to preview every single diff before it lands.

• Write Files: Auto-approved.
• Edit Files: Auto-approved.
• Shell Commands: Requires confirmation.
• Recommended For: Iterative development. You trust the agent's code modifications, but still want to audit what scripts or test commands it executes.

• Write Files: Auto-approved.
• Edit Files: Auto-approved.
• Shell Commands: Auto-approved.
• Destructive Shell Commands: Requires confirmation (e.g. rm -rf, git reset --hard, npm uninstall).
• Recommended For: Fast autonomous coding where you trust the model to build and run test commands in the background.

• All Tool Calls: Auto-approved.
• All Shell Commands: Auto-approved (including destructive ones).
• Recommended For: Sandboxed environments, containerized setups, or high-trust operations where you want the agent to operate completely unattended.

Switch modes dynamically inside the active session:

Type directly into the chat input bar:

/mode safe
/mode yolo

Specify the permission level when launching a headless instruction:

andromity run "refactor database.py" --mode trust
andromity run "run all clean tasks" --mode yolo