AEGIS helps protect local AI-agent workflows with security reviews and configured action controls.
Choose a local folder or agent profile for static review of suspicious patterns in skills and settings. Compare saved snapshots. Investigate process trees, configured file activity, TCP endpoints and alerts with captured evidence.
Suspend or resume a currently observed selected process, or confirm Stop. AEGIS rechecks its identity before acting.
Explicitly configured AEGIS execution and selected-action routes block policy-denied commands before they start. The validating MCP gateway requires exact one-use approval for a tool and its arguments.
CPU/RAM charts, Claude Code token subtotals and estimated costs, audit history and exports support investigations.
MIT-licensed Windows-first public beta; macOS and Linux source support is experimental.
Features
- Static reviews of suspicious patterns in selected local folders and agent profiles
- Saved snapshots for comparing security-relevant configuration changes
- Detected-agent process trees, configured file observations and TCP endpoints
- Alerts with captured evidence, audit history and exportable reports
- Suspend/resume a currently observed selected process or confirm Stop; process identity is rechecked
- Configured execution routes block policy-denied commands; MCP gateway requires exact one-use tool/arguments approval
- CPU/RAM charts and Claude Code token subtotals with estimated costs
- MIT-licensed Windows-first desktop beta; macOS/Linux source support experimental
Categories
CybersecurityLicense
MIT LicenseFollow AEGIS
User Reviews
-
I like that AEGIS combines visibility with controls: seeing what an agent does is useful, and having a way to review or deny selected actions makes the idea even more interesting. Curious to see where the project goes.
-
Interesting approach. Reviewing agent skills before use and putting specific MCP calls behind approval seems like a practical way to make AI workflows safer. The open-source angle makes AEGIS worth following.