Specter

Most websites silently load dozens of third-party scripts the moment you open them — behavioral trackers, ad networks, fingerprinting scripts, session replay tools recording your every click. Specter makes all of it visible. And now, stoppable. **What Specter does:** Specter intercepts every network request your browser makes and classifies it in real time using a locally-running ML model — trained on 25,290 real browsing requests, 96.2% accuracy. No data ever leaves your machine. No server. No proxy. No account required. **Detect** — Behavioral trackers, fingerprinting scripts, ad networks, analytics, session replay tools (Hotjar, FullStory, LogRocket) — Confidence score and per-request explainability for every classification — Privacy score (0–100) per site, updated live as you browse — Force-directed network graph showing which third-party systems each site connects to **Block (opt-in)** — Adaptive blocking engine acting on classifier output — not a static list — Strips tracking parameters (fbclid, gclid, utm_*, and 27 others) from URLs automatically — Per-site allow-list so you can restore sites that break **Defend** — Fingerprint defense hooks that add calibrated noise to canvas, WebGL, AudioContext, and navigator APIs before page scripts run — Session-scoped randomness — consistent within a session, different across sessions, so sites stay functional **Analyze** — Live request feed with category badges, confidence scores, and URL paths — Session history with per-site breakdowns, sortable by date, tracker count, or privacy score — Full JSON export or plain-text report for any session — Optional VirusTotal domain reputation lookups (opt-in, your API key) **Privacy by design:** Specter does not collect, transmit, or store any data externally. Everything lives in your browser's local storage, deleted on your schedule. The ML model runs entirely on-device using ONNX Runtime. Open source — github.com/divij404/specter