Cogumi AI Shield

Protects users from unsafe AI agent actions with deterministic policy enforcement.

Cogumi AI Shield is a security-first browser extension for “agentic” workflows. It helps you control what AI agents can **read** and **do** in the browser with deterministic enforcement (policy + time‑bound grants) and local auditability. **Key features** - Decision prompts for sensitive actions: paste, send/submit, file upload, downloads, clipboard write, OAuth consent. - Sensitive-read protection (advanced): gates reads from form fields/contenteditable, storage, cookies, and clipboard read (agent context only). - Network egress allowlists (advanced): optional allow/deny lists for outbound destinations (agent context only). - Time-bound approvals: allow once or allow for 1–10 minutes, with quick revoke + optional auto-revoke on tab switch/idle. - Detections + Threat Hunting UI: local incident clustering, timeline, and a “threat path” view (no backend required). - Policy matrix: Global + per-agent controls for ChatGPT / Claude / Gemini / Other, plus “use global for all”. **Privacy by design** - Local-first: decisions, detectors, grants, and audit logs are stored locally. - Telemetry is OFF by default and opt-in only (no prompts/files/form values). - Clear, user-visible “why” for prompts/blocks + transparent Settings.