Privacy

Overview

tabby is an on-device AI autocomplete app for macOS. It is designed around a local-first architecture - your writing, keystrokes, and suggestions never leave your Mac. There are no user accounts, no hosted dashboards, and no cloud-based processing.

What the app accesses locally

To provide inline suggestions, tabby reads data from your Mac through macOS system APIs. All of this data stays on your device and is never transmitted.

• Accessibility API data. tabby reads the content of the currently focused text field, caret position, and element bounds through the macOS Accessibility API. This is how it knows what you are typing and where to place suggestions.
• Input Monitoring. tabby monitors global keyboard events to detect typing activity and Tab key presses for accepting suggestions. It categorizes key events (text input, navigation, shortcuts) but does not log or store individual keystrokes.
• Clipboard contents. tabby reads the system clipboard at the time of generating a suggestion to provide additional context to the local language model. Clipboard data is not cached, stored, or transmitted.
• Screen capture. tabby can capture a region around the focused text field using ScreenCaptureKit to provide visual context. These captures are processed locally and are not stored or transmitted.

What the app does not collect

• No usage analytics or telemetry of any kind
• No personal information, email addresses, or identifiers
• No remote logging of text, keystrokes, or suggestions
• No user accounts or authentication
• No cookies or tracking pixels

Network activity

tabby makes two types of network requests, both optional and user-visible:

• Update checks. tabby uses Sparkle to check for app updates from updates.tabbyapp.dev. Automatic update checks are disabled by default.
• Model downloads. When you choose to download an open-source GGUF model, tabby fetches it from Hugging Face. This only happens when you explicitly initiate a download.

No other network requests are made by the app.

Local storage

• Preferences. Settings like ghost text color, model selection, and suggestion length are stored in macOS UserDefaults on your device.
• Downloaded models.GGUF model files you download are stored on disk in the app's local directory. You can delete them at any time from within the app.

macOS permissions

tabby requests the following macOS permissions:

• Accessibility. Required to detect focused text fields, read their content, and position ghost text suggestions near the caret.
• Input Monitoring. Required to detect typing activity and handle Tab key acceptance of suggestions.
• Screen Recording. Required to capture a screenshot around the focused field for visual context (OCR).

Third-party components

tabby integrates the following third-party components. None of them transmit your data off-device:

• llama.cpp / llama.swift. Local inference engine for running GGUF models on your hardware.
• Apple Intelligence.Apple's on-device FoundationModels runtime (macOS 26+). Governed by Apple's own privacy policy.
• Sparkle. Open-source macOS update framework.
• Hugging Face.Model hosting platform used for GGUF downloads. Subject to Hugging Face's privacy policy during downloads only.

Open source

tabby is licensed under AGPL-3.0. The full source code is available on GitHub. You can audit exactly what the app does at any time.

Children's privacy

tabby is not directed at children under 13. We do not knowingly collect any data from children. Since tabby does not collect personal information from anyone, there is no data to identify or remove.

Changes to this policy

If this policy changes, we will update this page and the "last updated" date above. For significant changes, we will note them in the release notes.

Contact

Questions about this policy? Email hi@tabbyapp.dev.