Complete, local-first AI for your Mac — in the notch. Point it at any surface you can show it: your screen, any app window, or your camera. Press ⌘⇧P and a local vision model streams a short answer into the notch. The whole stack is yours to customize — bring your own model and swap it anytime, no code changes.
Private by default. Capture only fires when you press the hotkey. Inference runs on local models on your machine via Ollama (or any OpenAI-compatible server). Saved conversations are opt-in and encrypted. No account, no cloud round-trip.
Highlights
- Every surface: screen, window-under-cursor, full display, and camera capture
- Local-first inference: Ollama by default (Gemma 4), plus optional remote Ollama,
:cloudtags, and OpenAI-compatible backends (LM Studio, vLLM) — all HTTPS-gated - Bring your own model from the in-app library, with built-in vision-capability checks
- Opt-in, encrypted conversation archive with a History switcher
- On-device speech (opt-in): voice input + read-aloud, never leaves your Mac
- Opt-in web lookup (DuckDuckGo), secret-aware so API keys/passwords never become a query
- Signed + notarized direct download — opens cleanly through Gatekeeper
Install
Download Peeknook.dmg below (or from peeknook.com), open it, and drag Peeknook to Applications.
Requirements: macOS 15+, plus a local model for inference (an Ollama vision model such as Gemma 4). Full setup in INSTALL.md.
Free and fully open-source under Apache-2.0.