openagent

Least-privilege permissions

• Restrict system access: Run openagent on a dedicated user account with minimal file permissions—not as administrator. It should only access folders it genuinely needs (e.g., documents folder, not your entire hard drive).
• Separate credential storage: Keep API keys and passwords in a locked vault (like 1Password or similar) rather than in files openagent can read. Grant access only when actively needed.
• Limit browser scope: If using browser-use features, configure it to access only approved websites. Disable access to banking, email, and internal systems unless absolutely required.
• Isolate from production systems: Never connect openagent to production databases, customer data repositories, or critical infrastructure—use test environments only.

Network isolation

• Air-gapped or sandboxed network: Run openagent on a separate network segment or virtual machine isolated from your main corporate network to contain any breach.
• Outbound firewall rules: Whitelist only the LLM provider's API endpoints (e.g., OpenAI, Claude) and essential services. Block all other outbound traffic by default.
• No direct internet for sensitive data: If RAG is enabled, ensure it cannot access or download files from untrusted sources; pre-screen all documents it ingests.
• VPN or proxy logging: Route all openagent traffic through a monitored VPN or proxy to track what data leaves your network.

Monitoring

• Log all agent actions: Enable comprehensive logging of every command, file accessed, and website visited—store logs separately from openagent itself where they cannot be tampered with.
• Real-time alerts on anomalies: Set up alerts for unusual behavior: large data transfers, access to unexpected folders, or attempts to modify system files.
• Credential usage tracking: Monitor when and how API keys and passwords are being used; flag any use outside normal patterns.
• Regular audit reviews: Weekly review of logs to spot suspicious activity, failed access attempts, or unintended data exposure.

Red flags to watch for

• Unusual network traffic: Large uploads, connections to unfamiliar IP addresses, or repeated failed login attempts to external services.
• File or folder access you didn't authorize: Openagent reading files outside its configured scope, especially credential files or backup directories.
• Unexpected LLM billing or API usage: Sudden spikes in API calls or costs that don't match your expected usage.
• Inconsistent or missing logs: Gaps in activity logs or inability to explain what openagent did during specific timeframes—may indicate tampering or unaudited execution.