Install once
Download the .dmg, drag Cotabby into Applications, and launch it like any other Mac app.
macOS install
Open-source AI autocomplete for the apps you already use, powered by Apple Intelligence or local models and kept entirely on your Mac.
Cotabby is free and 100% open source. Support development & buy us a coffee
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.
Install once. It lives in your menu bar and listens quietly in every text field on your Mac.
Download the .dmg, drag Cotabby into Applications, and launch it like any other Mac app.
macOS install
Cotabby watches your cursor. When you pause mid-sentence, it suggests the next thought inline as ghost text.
The suggestion snaps in. Keep typing to adjust it, or press Escape and Cotabby steps out of the way.
A few examples of Cotabby quietly finishing thoughts (and dropping in the right emoji) in the apps you already use - clips from a real session.
Tune the suggestions so they feel helpful, not intrusive.
Four built-in models ship with Cotabby - tabby-1-nano for instant suggestions, to tabby-1-pro for the sharpest output. Also drop in your own GGUF.
tabby-1-nano
~0.1 GBSmolLM2-135M-Instruct-q8_0.gguf
Mungert/SmolLM2-135M-Instruct-GGUF
tabby-1-mini
~0.4 GBQwen3-0.6B-Q4_K_M.gguf
unsloth/Qwen3-0.6B-GGUF
tabby-1-base
~3.1 GBgemma-4-E2B-it-Q4_K_M.gguf
unsloth/gemma-4-E2B-it-GGUF
tabby-1-pro
~5.0 GBgemma-4-E4B-it-Q4_K_M.gguf
unsloth/gemma-4-E4B-it-GGUF
Three presets control how many words Cotabby suggests at a time. Default is 7-12 - enough to finish your thought, not enough to take over.
Suggestions stay concise by default so they still feel like an extension of your own sentence.
Coming soon: Cotabby will adapt to how you naturally write, keep lightweight memory over time, and make suggestions feel more like you.
Over time, Cotabby will learn the way you phrase things, remember what matters to you, and keep suggestions sounding like you.
Cotabby needs three macOS permissions to work. Here's exactly what each one does and why nothing ever leaves your machine.
Reads the focused field, caret position, and nearby text through the macOS Accessibility API to place suggestions.
Processed in memory, never written to disk or sent anywhere.
Watches keyboard events to detect typing and Tab presses for accepting suggestions.
Categorized as you type, never logged, stored, or transmitted.
Captures a small region around the focused field via ScreenCaptureKit so the model can read layout and context.
Read locally in real time, discarded the instant it's used.
The basics, without the enterprise brochure voice.
Cotabby is a local 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. It also includes inline emoji autocomplete with the same :shortcut: style.
Type a colon followed by a couple of letters anywhere on your Mac (:smi, :tada, :+1) and Cotabby pops up a small list of matching emojis. Press Tab to insert the highlighted one without leaving your current app. The emoji match is fully local - it uses the gemoji shortcode set, no model required.
Yes. Cotabby is free and open source under AGPL. No account, no subscription, no usage limits.
No. All inference happens on your Mac. Cotabby uses either Apple Intelligence or a local GGUF model - there is no cloud API in the current product. Your text never leaves your machine.
Cotabby reads the focused text field through macOS Accessibility APIs, feeds the text around your caret to a local model, and renders the continuation as ghost text near your cursor. Press Tab to accept in chunks, or just keep typing to ignore it.
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 four built-in models (tabby-1-nano, tabby-1-mini, tabby-1-base, and tabby-1-pro), and lets you bring your own. If Apple Intelligence isn't available on your Mac, the Open Source engine still works.
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.
Accessibility, Input Monitoring, and Screen Recording. Accessibility lets Cotabby detect the focused field, read nearby text, and place suggestions. Input Monitoring lets Cotabby see your keystrokes and handle Tab acceptance. Screen Recording lets Cotabby capture a screenshot around the focused field for visual context (OCR).
macOS 15.0 or later for the Open Source engine. Apple Intelligence requires macOS 26 and compatible hardware.
They range from about 0.1 GB for tabby-1-nano up to about 5.0 GB for tabby-1-pro, with tabby-1-mini around 0.4 GB and tabby-1-base around 3.1 GB in between. Apple Intelligence uses the system runtime, so no separate download. You can also add your own GGUF models if you prefer something different.
Yes. Once a model is downloaded, Cotabby runs entirely offline. No network connection needed for suggestions.
Usually it's a missing permission, Cotabby being paused, or no model installed yet. Check that Accessibility, Input Monitoring, and Screen Recording are enabled in System Settings, make sure Cotabby is active in the menu bar, and confirm you have a model selected. Some apps also don't expose enough accessibility data for Cotabby to work.
Cotabby only runs inference when you pause typing, and the smaller models are designed to be lightweight. tabby-1-nano and tabby-1-mini use minimal resources. You might notice slightly more CPU usage with tabby-1-base and tabby-1-pro, 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.
macOS 15 or later · Apple Silicon · DMG install
Cozy AI autocomplete for the everyday notes, emails, and messages you were going to write anyway.