Write at the speed of thought.In any app.

Open-source AI autocomplete for the apps you already use, powered by Apple Intelligence or local models and kept entirely on your Mac.

Star on GitHub
Open Source
AGPL-3.0 licensed. Free forever - read every line, fork it, contribute back.
On-Device
No cloud, no analytics, no accounts. Your writing never leaves your Mac.
Model Customization
Apple Intelligence, open-source GGUF models, or bring your own.

See how the suggestions feel inside a real writing flow, instead of floating around like a checklist.

Mail, notes, docs, messages, wherever you happen to be typing.

Apple Mail icon
Apple Mail
Apple Notes icon
Apple Notes
Google Chrome icon
Google Chrome
Microsoft Outlook icon
Microsoft Outlook
Gmail icon
Gmail
iMessage icon
iMessage
Notion icon
Notion
Discord icon
Discord
Slack icon
Slack
Apple Mail icon
Apple Mail
Apple Notes icon
Apple Notes
Google Chrome icon
Google Chrome
Microsoft Outlook icon
Microsoft Outlook
Gmail icon
Gmail
iMessage icon
iMessage
Notion icon
Notion
Discord icon
Discord
Slack icon
Slack

how tabby works

Install once. It lives in your menu bar and listens quietly in every text field on your Mac.

01

Install once

Download the .dmg, drag tabby into Applications, and launch it like any other Mac app.

macOS install

Download the .dmg

drag to Applications
02

Type anywhere

tabby watches your cursor. When you pause mid-sentence, it suggests the next thought inline as ghost text.

MailNotesSlackNotionDocs+ anywhere
03

Press Tab

The suggestion snaps in. Keep typing to adjust it, or press Escape and tabby steps out of the way.

Tabghost text becomes your words
I folded your feedback into the deck and tightened the closing CTA.

A few examples of tabby quietly finishing thoughts in the apps you already use - clips from a real session.

write your emails faster

write your notes faster

write your Slack messages faster

write your messages faster

Tune the suggestions so they feel helpful, not intrusive.

models

choose your model

Two built-in models ship with tabby. Pick fast for speed, or balanced when you want sharper suggestions. You can also drop in your own GGUF.

tabby-fast-1
~0.4 GB · speed-optimized
tabby-balanced-1
~0.8 GB · higher quality

length

short or long

Three presets control how many words tabby suggests at a time. Default is 7-12 - enough to finish your thought, not enough to take over.

3-7 words7-12 words12-20 words

Suggestions stay concise by default so they still feel like an extension of your own sentence.

ai personalization

learns your voice

Coming soon: tabby will adapt to how you naturally write, keep lightweight memory over time, and make suggestions feel more like you.

coming soonwriting stylememoryadapts over time

Over time, tabby will learn the way you phrase things, remember what matters to you, and keep suggestions sounding like you.

tabby needs three macOS permissions to work. Here's exactly what each one does and why nothing ever leaves your machine.

accessibility

reads text fields

Reads the focused text field, caret position, and surrounding text via the macOS Accessibility API to place suggestions.

Text content is processed in-memory and never written to disk or sent anywhere.

input monitoring

detects your typing

Monitors keyboard events to detect typing activity and Tab presses for accepting suggestions. Keys are categorized, never logged.

Keystrokes are categorized but never logged, stored, or transmitted.

screen recording

captures visual context

Captures a small region around the focused text field via ScreenCaptureKit so the model can read formatting and context.

Screenshots are processed locally in real-time and immediately discarded.

The basics, without the enterprise brochure voice.

What is tabby?+

tabby is an on-device AI autocomplete for macOS. It sits in your menu bar, watches the text field you're typing in, and suggests the next few words as ghost text. Everything runs locally on your Mac.

Is tabby free?+

Yes. tabby is free and open source under AGPL. No account, no subscription, no usage limits.

Does tabby send my writing to the cloud?+

No. All inference happens on your Mac. tabby uses either Apple Intelligence or a local GGUF model - there is no cloud API in the current product. Your text never leaves your machine.

How does tabby actually work?+

tabby reads the focused text field through macOS Accessibility APIs, feeds the text around your caret to an on-device model, and renders the continuation as ghost text near your cursor. Press Tab to accept in chunks, or just keep typing to ignore it.

What is the difference between the Apple Intelligence and Open Source engines?+

Apple Intelligence uses Apple's built-in FoundationModels runtime - no download required, but it needs macOS 26 and supported hardware. The Open Source engine runs local GGUF models through llama.cpp, ships with two built-in models (tabby-fast-1 and tabby-balanced-1), and lets you bring your own. If Apple Intelligence isn't available on your Mac, the Open Source engine still works.

What apps does tabby work with?+

Most macOS text fields - Mail, Slack, Notes, iMessage, browser text areas, and more. Compatibility depends on what each app exposes through Accessibility APIs, so placement and reliability can vary. Some apps work great, others are hit or miss.

What permissions does tabby need?+

Accessibility, Input Monitoring, and Screen Recording. Accessibility lets tabby detect the focused field, read nearby text, and place suggestions. Input Monitoring lets tabby see your keystrokes and handle Tab acceptance. Screen Recording lets tabby capture a screenshot around the focused field for visual context (OCR).

What macOS version do I need?+

macOS 15.0 or later for the Open Source engine. Apple Intelligence requires macOS 26 and compatible hardware.

How much space do the models take?+

tabby-fast-1 is about 0.4 GB and tabby-balanced-1 is about 0.8 GB. Apple Intelligence uses the system runtime, so no separate download. You can also add your own GGUF models if you prefer something different.

Does tabby work offline?+

Yes. Once a model is downloaded, tabby runs entirely offline. No network connection needed for suggestions.

Why am I not seeing suggestions?+

Usually it's a missing permission, tabby being paused, or no model installed yet. Check that Accessibility, Input Monitoring, and Screen Recording are enabled in System Settings, make sure tabby is active in the menu bar, and confirm you have a model selected. Some apps also don't expose enough accessibility data for tabby to work.

Does tabby slow down my Mac?+

tabby only runs inference when you pause typing, and the models are designed to be lightweight. tabby-fast-1 uses minimal resources. You might notice slightly more CPU usage with tabby-balanced-1 since it's a larger model, but it shouldn't affect normal use.

It is built to feel like a small cozy helper that lives on your Mac, not another dashboard asking for attention.

Star on GitHub

macOS 15 or later · Apple Silicon · DMG install

type a lot faster.

Cozy AI autocomplete for the everyday notes, emails, and messages you were going to write anyway.

Star on GitHub