Detect Provenance
Drop an image to look it up in the Aegis provenance index. The image is fingerprinted on-device (PDQ + tiled Merkle) and only the 256-bit hashes are sent — your image never leaves the device. Matches are ranked by tile overlap across registered assets.
The hashes that cross the wire are one-way perceptual digests — the original image cannot be reconstituted from them. It's mathematically not possible. Aegis only ever learns whether something matches an already-registered fingerprint.
Rate Limits
Limits are per-IP (anonymous) or per-DID (authenticated). Rate limit headers are included in every response.
Query Image
Drop or pick an image
PNG · JPEG · WebP — processed on-device
How Detect works
Zero-image search. The query image is decoded and fingerprinted entirely in your browser using WASM. Only PDQ perceptual hashes (one global 256-bit + 64 tiled regional hashes) and an optional DINOv2/SSCD semantic embedding cross the wire — your original image bytes never leave the device.
Index. Cloudflare Vectorize ANN over registered tile PDQs. Scoring combines tile overlap, semantic cosine similarity (DINOv2), spatial coherence, and confidence. Results are tiered: Exact → Strong → Likely → Possible → Weak.
Privacy guarantee. PDQ hashes are one-way perceptual digests — the original image cannot be reconstituted from them. The search only determines whether an image matches an already-registered fingerprint in the Aegis index.
Endpoint: POST /api/search/fragment on the Aegis worker.
Anonymous requests are rate-limited per-IP; authenticated requests (via MaataraAssertion header) are rate-limited per-DID at a
higher tier.