Overview
Short version: Backtrack runs an AI model on your device to let you search the pages you have read by meaning. The page text it indexes and the search index it builds stay in your browser. The only network request is a one-time download of the AI model. No servers, no accounts, no tracking.
What we collect
Nothing is sent anywhere. Backtrack does not transmit or sell your browsing data, page content, search queries, or usage analytics. There are no servers, no accounts, no cookies, no tracking pixels. The data Backtrack creates while indexing (extracted page text and embedding vectors) is stored only on your device.
How the extension works
When you open a readable page, Backtrack extracts the article text with Mozilla Readability, splits it into short chunks, and converts each chunk into an embedding vector using an on-device model (all-MiniLM-L6-v2) that runs in an offscreen document via ONNX and WebAssembly. The vectors and their source text are saved locally in your browser through IndexedDB.
When you search, your query is turned into a vector the same way and compared against the stored chunks to rank the closest matches. Indexing and search both happen inside your browser. Your queries are never sent to a server.
What stays out of the index
Backtrack skips a built-in blocklist of sensitive surfaces: email, banking, login and sign-in pages, and local addresses such as localhost. Pages with little readable text are skipped, and each page is indexed once (deduplicated by URL). You can wipe the entire local index at any time from the search page.
Permissions explained
storage– saves local state and settings on your device. This data stays in your browser.tabs– reads the active tab's address so Backtrack can index the current page when you ask it to, and opens its own search page.scripting– re-runs the indexing script on the current tab when you click "index this tab".offscreen– runs the AI model in an offscreen document, off the service worker, so the browser stays responsive.<all_urls>content script – reads the readable text of pages you visit so they can be added to your local index. Backtrack reads this text only to build the index on your device. It does not transmit it, and it ignores form fields, passwords, and the blocklisted surfaces above.- host access to
huggingface.coandhf.co– downloads the model files once, on first run.
Third-party services
The only external network connection Backtrack makes is a one-time download of the model from the Hugging Face CDN (huggingface.co, hf.co). That request fetches model files only. It does not include your page content, your history, or your search queries. After the model is cached, Backtrack works offline.
Backtrack does not use Google Analytics, Sentry, Mixpanel, or any other analytics or telemetry. It does not load remote scripts.
Children's privacy
Backtrack is not directed at children under 13. Because the extension does not collect or transmit any data, no specific compliance procedures are required.
Changes to this policy
If this policy changes, the updated version will be published at this URL with a new "Last updated" date.
Contact
Questions about privacy or the extension? Reach out at lobovlabs@proton.me.