Pretype 0.1.0

Pretype 0.1.0 — the first public release. 🎉

System-wide AI autocomplete for macOS: Copilot-style ghost-text suggestions in any text field — fully offline, private, and on-device.

✨ Highlights

• 🌐 Works everywhere — native AppKit/SwiftUI apps, Electron apps (VS Code, Slack, Claude Desktop), and web views.
• 👻 Pixel-matched ghost text at the caret — Tab accepts the next word, ⇧Tab the rest, or just type over it to reject.
• 🔤 Inline typo fixes (English + Russian) via the macOS system spell-checker.
• 🪄 Selection rewrites (⌥Tab) — the local LLM fixes grammar/phrasing in place while preserving your tone.
• 🤖 Local inference — MLX + Gemma 4, or the Apple Intelligence system model on macOS 26+.
• ⚡ Low latency — KV-cache reuse delivers hot completions in 0.05–0.2 s.
• 🛡️ 100% on-device — no cloud, no subscriptions, no telemetry. Your keystrokes never leave your Mac.

💻 Requirements

• macOS 14+ on Apple Silicon (M1 or newer) — Intel Macs are not supported (MLX requires Apple Silicon).
• 3.4–8.6 GB free for the local MLX model (downloaded from Hugging Face on first launch; variant auto-selected by your RAM).

📦 Install

Pretype.app.zip is ad-hoc signed (not notarized — that needs a paid Apple Developer ID), so Gatekeeper will block it once. To open it:

xattr -dr com.apple.quarantine /path/to/Pretype.app && open /path/to/Pretype.app

…or without Terminal: double-click → "blocked" → System Settings → Privacy & Security → scroll down → Open Anyway.

Then grant Accessibility when prompted — this is how Pretype reads the active text field, catches Tab, and types suggestions back. (Optionally grant Screen Recording to enable OCR window context.) If you grant it after launch, restart the app.

📝 Notes

Early-stage first release — bug reports and feedback are very welcome via Issues. Prefer building from source? See the README.