Available models
Every demo downloads a real model into your browser and runs it on your device with WebGPU. No inference servers, no API keys. The download size shown is the entire model.
Live now
Deep Research
● liveAn agent plans searches, reads live sources, and writes a cited brief in front of you.
970 MB · desktop · WebGPULive Object Detection
● liveEdgeSight-1 draws labelled boxes on your camera feed in real time.
27 MB · WebGPUPrivate Document Q&A
● liveDrop a contract or report and ask questions. Indexed and answered on your device with cited passages.
990 MB · desktop · WebGPUPII Redaction
● liveStrip names, cards, and identifiers on-device before a document leaves your perimeter.
970 MB · desktop · WebGPUMeeting Notes
● liveRecord a memo; two chained models turn it into minutes with action items.
1.1 GB · desktop · WebGPUSwim Timing
● liveUpload swimming footage; a detector we trained ourselves finds and tracks swimmers on-device.
9 MB · WebGPUScanned Document OCR
● liveUpload a scan or photo of a page; two models find and read the text on-device.
96 MB · WebGPUSensor Anomaly Detection
● liveA 12 KB model we trained ourselves flags spikes, outages, and stuck sensors in meter data.
12 KB · WebGPUTable Extraction
● liveUpload a statement or invoice PDF; two models turn its tables into copyable CSV on-device.
230 MB · WebGPUVoice Fraud Detection
● liveRecord or upload a voice clip; EdgeVoice-1 estimates on-device whether it is cloned or human.
125 MB · WebGPUAI Guardrails
● liveScreen text for prompt-injection attacks and abusive content before it reaches a model.
370 MB · WebGPUAI Image Detection
● liveUpload an image; EdgeVerify-2 estimates on-device whether it was generated or captured.
785 MB · desktop · WebGPUIn the lab
Depth From a Photo
in the labA flat photo becomes a 3D scene you can tilt.
19 MB · WebGPUBackground Removal
in the labClean transparent cut-outs, fully offline.
44 MB · WebGPULive Captions
in the labWhisper transcribes your voice as you speak.
41 MB · WebGPUVoice Studio
in the labType anything; Kokoro speaks it in 56 voices across 8 languages.
86 MB · WebGPUOffline Chat
in the labA streaming chatbot that keeps working in airplane mode.
290 MB · WebGPUVision Q&A
in the labUpload an image and ask anything about it.
189 MB · WebGPUAnything Recognizer
in the labType your own labels; the camera classifies against them.
153 MB · WebGPUCode Assistant
in the labA coding copilot with no telemetry.
555 MB · WebGPUPhoto Upscaler
in the labDouble the resolution of a blurry photo.
32 MB · WebGPUSemantic Search
in the labSearch any text by meaning as you type.
23 MB · WebGPUAll demo models are Apache-2.0 licensed. Demos ship one at a time; the rest of the lab is on its way.