On-device AI autocomplete for macOS text fields.
Tabby is a menu bar app that brings inline autocomplete to the text field you're already using. Keep typing in your host app — Tabby watches the focused field, generates a continuation, and renders it as ghost text next to your caret. Press Tab to accept a chunk, keep pressing to advance, or just keep typing to diverge.
Everything runs on-device. No hosted API, no cloud round-trip.
Apple Intelligence — uses Apple's on-device FoundationModels runtime. No download required. Availability depends on your Mac; Tabby checks at runtime and explains when this engine is unavailable.
Open Source — runs local GGUF models in-process through llama.cpp via llama.swift. Built-in downloadable models:
| Model | File | Size |
|---|---|---|
tabby-fast-1 |
Qwen3-0.6B-Q4_K_M.gguf |
~0.4 GB |
tabby-balanced-1 |
gemma-3-1b-it-Q4_K_M.gguf |
~0.8 GB |
tabby-depth-1 |
gemma-3n-E4B-it-Q4_K_M.gguf |
~3.5 GB |
You can also drop your own .gguf files into Tabby's models folder and refresh the model list.
- Download the latest
Tabby.dmgfrom GitHub Releases. - Drag
Tabby.appintoApplicationsand launch it. - Grant Accessibility and Input Monitoring when prompted.
- Pick an engine — Apple Intelligence if available, otherwise Open Source + a model.
- Start typing in any supported editable field.
If macOS blocks first launch, right-click Tabby.app → Open, or allow it in System Settings → Privacy & Security.
- Accessibility — read the focused text field's value and caret position.
- Input Monitoring — detect global
Tabpresses for acceptance.
- Ghost text rendered live next to your caret
- Partial
Tabacceptance — take a chunk, keep the tail alive, press again to continue - Menu bar quick controls: enable, engine, model, indicator mode, completion length
- Settings for launch at login, ghost text color, model downloads, and updates
- Activity indicators that can be hidden, anchored to the caret, or shown as a field-edge icon
- Accepted-word counter
Requires macOS 26.0 or later. Behavior depends on what each host app exposes through the Accessibility APIs — some fields only provide coarse caret geometry, so Tabby falls back to more conservative placement.
Requires Xcode and Command Line Tools. Apple Silicon is strongly recommended for local model performance. For setup, build, test, and contribution workflow details, start with CONTRIBUTING.md.
git clone https://github.com/FuJacob/tabby.git
cd tabby
open tabby.xcodeprojIf you want to understand the runtime and suggestion pipeline before contributing, read ARCHITECTURE.md.
Tabby is licensed under the GNU Affero General Public License v3.0. The AGPL's network-use clause means any modified version made available to users over a network must also be source-available under the same terms.