How yasinyilmaz/ai-stegcloak works
AI StegCloak adds invisible, response-specific provenance marks to generated text, then checks copied text for surviving evidence. The main workflow runs from the browser interface through the HTTP API to the watermark service: minting embeds marks and records provenance data; detection compares extracted marks and content fingerprints against those records. The README also describes provider-side integration and partial-copy detection. The graph reflects the sampled server and watermark modules; unsampled UI markup is represented only by its exact tree path.
Main entry points: service.js Watermark service · app.js Browser app · server.js HTTP server · index.html Web page
Components 11
Browser interface
- Browser app
public/app.js - Web page
public/index.html
HTTP API
- HTTP server
src/server.js
Watermark processing
- Watermark service
src/watermark/service.js - Text tagging
src/watermark/text.js - Symbol encoding
src/watermark/alphabet.js - Content fingerprint
src/watermark/fingerprint.js - Evidence scoring
src/watermark/evidence.js
Watermark registry
- Memory registry
src/store/memory-store.js
Other components
- User
- AI provider
Connections 15
- User → Browser app: uses
- AI provider → Browser app: provides text
- HTTP server → Web page: serves
- Web page → Browser app: loads
- Browser app → HTTP server: POSTs mint/detect
- HTTP server → Watermark service: invokes
- HTTP server → Memory registry: reads health
- Watermark service → Text tagging: tags / extracts
- Watermark service → Symbol encoding: counts marks
- Watermark service → Content fingerprint: fingerprints text
- Watermark service → Evidence scoring: scores sequences
- Text tagging → Symbol encoding: encodes / decodes
- Watermark service → Memory registry: saves / lists records
- Watermark service → Content fingerprint: checks containment
- Browser app → User: shows results