$ cat ~/work/oliver.case
Oliver.
macOS app · In progress
// A macOS app that gives you invisible AI help during meetings—live transcription, chat, and screenshots without showing up on screen share.
// Impact at a glance
- - Invisible to screen share via native macOS window exclusion
- - No backend relay—keys and history stay on your machine
- - Combines STT, vision (screenshots), and multi-provider chat in one overlay
// summary
Oliver is a Tauri desktop app for macOS. It floats above your windows, transcribes mic and system audio, streams answers from the AI provider you choose, and stays hidden from Zoom, Teams, and screen recordings.
// problem
During interviews and meetings you often need fast answers without breaking flow—alt-tabbing, typing in another app, or showing an obvious AI window on a shared screen.
// what I built
I built Oliver as a native overlay excluded from screen capture (`NSWindowSharingNone`), with local SQLite history, encrypted API keys on-device, Whisper-based STT (local or API), and direct calls to OpenAI, Claude, Gemini, Groq, OpenRouter, or Ollama—no backend relay.
// core experience
- - Toggle the overlay with global hotkeys and get streamed AI responses in a floating panel
- - Live transcription from microphone and system audio (BlackHole) with optional screenshot context
- - Onboarding for provider keys, STT model choice, prompt presets, and usage history in a local dashboard
// architecture
- - Tauri 2 + Rust for macOS windowing, permissions, and SQLite via tauri-plugin-sql
- - Vite + React + TypeScript for the dashboard and overlay UI
- - Provider adapters with AES-GCM-SIV encrypted key storage; all traffic goes straight to the provider you configure
// ai involvement
Whisper handles speech-to-text; LLM providers handle chat. The product work is making that pipeline feel instant, private, and invisible to everyone else on the call.
// challenges
- - macOS screen-recording and audio routing (BlackHole multi-output setup)
- - Keeping latency low while streaming tokens and transcripts
- - Shipping a notarized desktop app with sensible defaults for hotkeys and permissions
// outcome
Active development with a public marketing site and open-source app repo. Oliver is my flagship native AI product.
// why this matters
It shows I can ship past the browser: OS APIs, privacy-sensitive data handling, and real-time AI UX in a tool people would actually run during a live meeting.
// reflection
The hardest part is not the model call—it is trust: local storage, no telemetry, and a UI that never leaks into the recording.
// capabilities
// links