Shadow

Your computer was paying attention the whole time: screen/audio/input capture, OCR, and semantic search, all on-device.

Shadow is a Rust capture-and-intelligence sidecar (~23k LOC): screen/audio/input capture, OCR, a proactive engine, and semantic memory + search. It records pure-Rust JPEG keyframes (no ffmpeg required). A component of Ryu, it runs standalone on :3030.

• 📸 Capture. Screen, audio, and input, with OCR over every frame.
• 🎞️ Zero-setup keyframes. Pure-Rust JPEG keyframes at the changed frames; no ffmpeg required.
• 🔎 Semantic memory. Everything captured is embedded and searchable on-device.
• 💡 Proactive engine. Surfaces context and suggestions from what it sees.
• 🏠 On-device by design. A device-bound sensor: capture never leaves the machine.
• 🔓 Open, consent-gated, pausable. Sensitive by nature, so it's auditable and gated behind explicit consent.

Layout

Path	What
apps/shadow	the capture sidecar (HTTP server, :3030)
crates/shadow-core	capture + semantic-search engine
crates/ghost-{core,eyes,hands}	shared automation crates (vendored from Ghost)

Build

cd apps/shadow && cargo build --release

Dual-use & consent

Continuous screen + audio capture is inherently sensitive. Shadow is a device-bound sensor: it runs locally, never routes capture off-device, and gates every source behind explicit consent and a pause control. It's open-source so this is fully auditable. See SECURITY.

Credits & license

Shadow is derived from Shadow by Ghostwright (MIT). The original copyright + license notice are retained in NOTICE. Licensed under Apache-2.0 (see LICENSE) with MIT-licensed portions per NOTICE. © 2026 A Major Pte. Ltd.