Tabitha is a browser extension that helps you manage, save, and search your tabs and windows, with optional AI features. This policy explains exactly what data Tabitha handles, where it is stored, and the limited cases in which data leaves your device.
Short version: Tabitha stores everything locally on your own device. It has no servers, no account system, and no analytics or tracking. AI features are off until you turn them on and choose where they run:
- On this device: open AI models run in your browser. Your pages, questions, and summaries never leave your computer. The models are downloaded once from Hugging Face.
- Google Gemini: with your own Gemini API key, page text and your questions go directly from your browser to Google, never to us.
MCP access is off too. If you turn it on and pair Tabitha with an AI app on your computer, such as Claude, that app can read your Saved Windows' names and your Saved Tabs' titles and URLs. Tabitha sends them only to that app, on your computer, never to us; the app may send what it reads to its own AI provider.
Who we are
Tabitha is developed by Frisson Labs. We do not operate any backend service that receives, stores, or processes your data. There is no “Tabitha account.”
Data stored on your device
All of Tabitha's durable data is stored locally in your browser (via IndexedDB, extension storage, and the browser's cache). It never syncs to us or any third party. This includes:
- Saved Windows and Saved Tabs — page titles, URLs, and favicons of the windows and tabs you choose to save.
- Captured page content — when AI features are on, Tabitha reads the text of your Saved Tabs (and, transiently, their HTML) so it can summarize and search them.
- AI-generated data — page summaries, text passages, and embedding vectors used for semantic search and chat. These are created only when AI features are on.
- On-device AI models — if you choose on-device AI, the downloaded models (about 3.4 GB) are kept in the browser's cache so they are downloaded only once.
- Session Snapshots — a rolling history (up to 20) of your open windows' titles and URLs, so you can reopen them after a crash or accidental close.
- Settings — your preferences (theme, backup settings, whether AI is on and where it runs, etc.), the result of the on-device AI check described below, your Google Gemini API key if you provide one, and the MCP access pairing code if you turn MCP access on.
You can delete this data at any time by removing Saved Windows, clearing the extension's data, or uninstalling the extension.
Choosing where AI runs
AI features (summaries, semantic search, auto-organize, and chat) are optional and off by default. When you turn them on, you choose where they run; nothing is chosen for you.
On this device
Tabitha runs open models in your browser: Google's Gemma 4 E2B for summaries, grouping, and chat, and EmbeddingGemma 2 for search. Your pages, summaries, search queries, and chat questions are processed on your computer and are not sent anywhere.
Tabitha offers this option only on computers that can run the models. To decide, it checks, locally, your graphics card's capabilities (WebGPU support and memory limits), approximately how much memory your computer has, and how much free storage the browser has. After the download, it runs a short test on a built-in sample page and saves whether it passed. None of this information leaves your device.
Google Gemini
Tabitha sends page text and your questions to Google's Gemini API, authenticated with your own API key (see below).
Your Google Gemini API key
Only needed if you choose Google Gemini:
- It is stored only on your device, in the extension's local storage.
- It is never transmitted to us.
- It is excluded from any Backup files Tabitha creates, so it won't leak if you share a Backup. Restoring a Backup keeps the key already on your device.
- It is used solely to authenticate your requests directly to Google's Gemini API.
MCP access (AI apps on your computer)
MCP access lets AI apps on your computer that support MCP, such as Claude
Code, read your Saved Windows through Tabitha's MCP server,
tabitha-mcp. The setup guide explains how
to set it up.
It is off by default. It works only after you turn on MCP Access in Tabitha's Settings and enter the pairing code your AI app gives you.
What AI apps can read
tabitha-mcp gives AI apps three tools. All of them only read:
- List Saved Windows: each Saved Window's id, name, number of Saved Tabs, and when it was saved.
- Search Saved Tabs: the Saved Tabs whose title or URL matches the app's query or, when AI features are on, whose page summary does. For each match: its title and URL, the id and name of its Saved Window, and whether the title, URL, or summary matched. At most 50 matches per search (20 unless the app asks for more).
- Get Saved Window: one Saved Window's id, name, when it was saved, and number of Saved Tabs, and the title and URL of each of its Saved Tabs, in order, up to the first 100.
Titles longer than 300 characters and URLs longer than 1,000 are cut short. The tools never return page text, summaries, favicons, your open tabs, Session Snapshots, your settings, or your Gemini API key, and they can't open, change, or delete anything.
Once Tabitha is paired, every AI app on your computer that runs
tabitha-mcp under your user account can use these tools, not
only the one that gave you the code.
When AI features are on, Search Saved Tabs also matches the query against your page summaries, the same search as Tabitha's search box. With on-device AI, the query stays on your computer. With Google Gemini, it is sent to Google's Gemini API, as your own searches are (see below). Summaries are used only for matching and are never sent to the AI app.
Where it goes
Tabitha sends the tools' answers only to tabitha-mcp on your
computer, which passes them to the AI app that asked.
Nothing goes to us or to any Tabitha server. The AI app
may send what it reads to its own AI provider (Claude, for example, sends
it to Anthropic as part of your conversation), under that provider's terms
and privacy policy. Tabitha has no say in what the AI app does with it.
How it travels
Only over localhost (127.0.0.1), inside your computer.
tabitha-mcp listens on 127.0.0.1 port 47391,
which other computers can't reach, and Tabitha connects to it from Chrome.
Before anything is sent, both sides prove they know the pairing code,
without sending the code itself, and connections from web pages are
refused. tabitha-mcp makes no internet requests of its own.
The pairing code
tabitha-mcp creates the code and keeps it in
~/.tabitha/pairing.json, readable only by your user account,
with the time Tabitha last connected. Tabitha keeps the code in its
settings, in the extension's local storage. It is not included in Backups.
How to stop it
-
Turn off MCP Access in Tabitha's Settings. Tabitha
disconnects straight away and stops looking for
tabitha-mcp. - Or remove Tabitha's server or plugin from your AI app.
To forget the pairing as well, delete ~/.tabitha/pairing.json;
tabitha-mcp makes a new code the next time it needs one.
When data leaves your device
Tabitha's core tab-management features work fully offline. Data leaves your device only in these specific cases:
1. Reading the pages you save (only if you turn on AI features)
To summarize and index a Saved Tab, Tabitha downloads that page again from its own website, as your browser would when you open it, and reads its text. No other site is contacted.
2. Downloading the on-device AI models (only if you choose on-device AI)
Tabitha downloads the models once from Hugging Face
(huggingface.co and its download servers). These requests
name only the model files; like any download, they reveal your IP address
and browser to Hugging Face. No page content, summaries, queries,
or questions are sent. After the download, on-device AI works
without a network connection. All of the code that runs the models ships
inside the extension; only the models' data is downloaded.
3. Google Gemini API (only if you choose Google Gemini)
The following is sent
directly from your browser to Google's Gemini API
(generativelanguage.googleapis.com) to power the
corresponding feature:
- Summaries: the URL, title, and extracted text of a page you save.
- Semantic search & indexing: page summaries and text passages (to generate embeddings), and your search queries, including those an AI app sends through MCP access.
- Chat: your questions plus the relevant passages and summaries retrieved from your own saved content. If you turn on cross-tab query expansion, your question is also sent to find the names and topics it mentions.
- Auto-organize: the summaries of the pages being grouped and, when you organize selected windows, the names of your other Saved Windows so their topics can be reused.
This data is processed by Google under Google's Privacy Policy and the Gemini API terms. Tabitha never receives a copy. If you don't choose Google Gemini, none of this data is ever sent.
4. YouTube transcripts
When AI features analyze a YouTube video page, Tabitha requests that video's transcript from YouTube's own endpoint (in the context of the YouTube page you already have open). This happens only for YouTube pages and only during analysis.
5. Favicons for chat citations
When the AI chat shows a cited source, Tabitha loads that site's favicon
from Google's public favicon service (google.com/s2/favicons).
Only the site's domain is included in that image request. This happens in
both AI modes. (The main tab list uses the browser's own favicons and
makes no such request.)
6. AI apps you pair through MCP access (only if you turn it on)
Tabitha sends Saved Window names and Saved Tab titles and URLs only to the AI app, over localhost, so they don't leave your computer through Tabitha. The AI app may send them on to its AI provider. See MCP access above.
What we do NOT do
- We do not collect analytics or telemetry of any kind.
- We do not use trackers, advertising, or fingerprinting.
- We do not run any server that receives your data.
- We do not sell, rent, or share your data with anyone.
- We do not transmit your browsing history anywhere (your saved data stays on your device; only content you explicitly save is used, and only as described above).
Permissions
Tabitha requests only the permissions needed for the features above. A plain-language explanation of each permission is provided in the store listing.
Children's privacy
Tabitha is a productivity tool not directed at children and does not knowingly collect any personal information from anyone.
Changes to this policy
If this policy changes materially, we will update the “Last updated” date above and the version published with the extension.
Contact
Questions about this policy? Contact: [email protected]