Redact Agent

Redact Agent is a privacy-first AI assistant for your Chrome side panel. It combines a Gemini-powered chatbot with a local privacy pipeline that automatically redacts sensitive data from screenshots before any image or page content is sent to the cloud. PRIVACY-FIRST BY DESIGN • The LLM never sees raw PII values — only category labels and confidence scores. • The LLM never sees the raw screenshot — only an opaque-mask version with sensitive regions painted out. • All detection and redaction runs locally in your browser. HOW IT PROTECTS YOU • DOM scan — detects form fields by attribute analysis. • OCR (Tesseract.js) — reads visible text in images and canvas elements, fully local. • YuNet face detector — on-device face detection via ONNX Runtime Web. • Pattern + context analysis — regex and label/value heuristics for emails, phones, cards, IDs. • Fusion — merges signals into high-confidence detections, then masks them. TWO MODES • Chat Mode — ask the model anything about the current page; the attached screenshot is already sanitised. • Agent Mode — give a privacy task (e.g. "redact all PII") and watch the agent observe, decide, and act step-by-step using native Gemini function calling. Bring your own Gemini API key. No account required. No telemetry.