Ai Overlay

Open-source desktop AI assistant that overlays your screen. Capture screenshots or text and send them to cloud or local AI models via global shortcuts.

Built with Tauri v2 (Rust) + SvelteKit (Svelte 5) + TailwindCSS v4 + TypeScript.

Features

• Multi-provider AI — OpenAI, Anthropic, Gemini (cloud) and llama.cpp (local, GPU auto-detection)
• Screen capture — Freeze screen, select region, send to vision models
• Text capture — Copy selected text from any app directly into the prompt
• Ephemeral chat — Conversations that are never written to disk (toggleable)
• Encrypted storage — API keys via Stronghold vault, chat history via AES-256-GCM
• Auto-updater — In-app update notifications with download progress
• Glassmorphic UI — Transparent, borderless window with frosted-glass theme

Shortcuts

Shortcut	Action	Type
Alt+C	Capture selected text into prompt	Global
Alt+S	Screenshot region selector	Global
Alt+A	Show / restore window	Global
Ctrl+N	New conversation	App
Ctrl+H	Toggle chat history	App
Ctrl+,	Toggle settings	App

Global shortcuts work system-wide (even when the app is not focused).
App shortcuts only work when the app window is focused.
All shortcuts are customizable in Settings → Shortcuts.

Prerequisites

• Rust (stable)
• Node.js (v18+)

Setup

git clone https://github.com/crony-io/aioverlay.git
cd aioverlay
npm install

Development

npm run tauri dev

Build

npm run tauri build

Release binaries are output to src-tauri/target/release/bundle/.

Lint & Check

npm run lint        # ESLint
npm run check       # svelte-check (TypeScript)
npm run format      # Prettier

Project Structure

src/                        # SvelteKit frontend
├── routes/                 # Single SPA page (+page.svelte)
├── lib/
│   ├── components/         # Reusable Svelte 5 components
│   ├── services/
│   │   ├── ai/             # Provider abstraction (OpenAI, Anthropic, Gemini, Local)
│   │   ├── local/          # llama.cpp server & model management
│   │   ├── shortcuts/      # Global shortcut registration
│   │   ├── updater/        # Auto-update service
│   │   ├── markdown/       # Marked + Shiki renderer
│   │   └── crypto/         # AES-256-GCM chat encryption
│   ├── stores/             # Chat state & AI streaming
│   └── utils/              # Shared utilities
src-tauri/                  # Rust backend
├── src/
│   ├── lib.rs              # Plugin registration & command handlers
│   ├── ai_proxy.rs         # Secure HTTP proxy (keys never leave Rust)
│   ├── capture.rs          # Screen capture (xcap)
│   ├── input.rs            # Key simulation (enigo)
│   └── llama/              # Local inference (download, process, models, platform)
├── capabilities/           # Tauri permission configs
└── Cargo.toml

Security

• API keys are stored in a Stronghold encrypted vault — never exposed to the webview or network inspector
• Chat history is encrypted at rest with AES-256-GCM using a per-install derived key
• AI requests route through a Rust HTTP proxy; keys are injected server-side
• CSP restricts the webview to self and local llama-server addresses only

License

MIT — see LICENSE.