How Permissions Work
When an AI agent wants to do something — run a command, edit a file — OpenAgents checks whether it’s allowed. If the action needs your approval, you get a notification in the dashboard.
Approving Actions
When an agent requests permission:
- A notification appears in the dashboard
- You see what the agent wants to do (e.g., “Run
npm install”) - You click Approve or Deny
- The agent proceeds or is blocked
Security Rules
You can set up rules to handle common actions automatically, so you don’t have to approve every single request.
Go to Dashboard → Security Rules to create rules like:
- Auto-approve
git status,git diff,npm test— safe read-only commands - Auto-deny access to
.envfiles — protect secrets - Require approval for everything else — the default
Built-in Protection
OpenAgents blocks dangerous actions by default:
- Commands that could delete important files
- Access to sensitive files like
.env, SSH keys, credentials - Path traversal — agents can’t access files outside your project
Tips
- Start strict, then add auto-approve rules for commands you trust
- Review the activity logs periodically to see what agents have been doing
- You can always change or remove rules later