On-Device AI Privacy & Perception
VEIL-X
VEIL-X (Visual Edge Intelligence Layer — X) is a privacy-preserving visual perception engine and autonomous browser agent pipeline engineered for defense, space, finance, and confidential enterprise environments.
Traditional AI web agents transmit raw screenshots and complete DOM trees directly to cloud-hosted Vision-Language Models (VLMs), creating high-severity data leakage risks. VEIL-X eliminates this risk by establishing a zero-leakage on-device perception pipeline: sensitive credentials, tokens, and PII are detected and redacted locally inside volatile browser memory before any telemetry or image leaves the client.
Powered by ONNX Runtime Web with WebGPU and WASM SIMD acceleration, VEIL-X executes lightweight edge vision models directly within the browser, passing only sanitized visual contexts and redacted scenes to reasoning backends for verified autonomous action execution.
Core Capabilities
- 01Zero-leakage client-side DOM inspection detecting credentials, IDs, and tokens in volatile memory
- 02In-browser HTML5 canvas redaction supporting Solid Blackout, Gaussian Blur, and Scrambled Placeholders
- 03Hardware-accelerated edge vision inference running ONNX models via WebGPU and WASM SIMD in the browser
- 04Privacy guardrail gateway ensuring unredacted screen buffers strictly never leave the local environment
- 05Autonomous browser action execution engine translating structured VLM action plans into verified DOM events
Engineering & Architecture
- In-browser content script perception engine with zero-dependency client-side ONNX Runtime Web integration
- Privacy guardrail middleware enforcing fail-closed sanitization prior to external API dispatch
- Python/FastAPI reasoning and telemetry service orchestrating multimodal VLM inference