Privacy Policy

Overview

Algoright AI ("we", "us") builds tools to help people reduce digital distraction. This policy explains what data we collect, why, and how you can control it.

What we collect

• Account data — your email address and name, provided via Clerk when you sign up.
• Usage statistics — number of videos filtered, time saved, and focus streaks. Stored in Supabase and tied to your account.
• Video titles and channel names, for the videos our on-device model sends for AI assistance — most classification happens entirely on your device and nothing about those videos is sent anywhere. For the smaller number our on-device model can't confidently categorise on its own, the title and channel name are sent to our server for AI assistance (see below). These are logged — title, channel name, the assigned category, and your account ID if you are signed in (no ID at all if you are not) — as training data to keep improving the on-device model. On TikTok, Instagram and X, the "title" is the start of the post's text or caption and the "channel name" is the account name. These logs are deleted automatically after 30 days.
• Category corrections you submit — if you tap ✓/✗ to correct how a video was categorised, the video's title, channel name, your correction, and the platform are stored on our server, and words from the title/channel are aggregated (without the full title) into a shared classifier database used to improve categorisation for everyone. This happens regardless of whether AI assistance is on or off for you — it's part of the correction feature itself, not the AI-assist feature. The stored correction (title, channel name and your correction) is deleted automatically after 30 days; only the aggregated words described below remain.
• Rate limit counters — a daily count of how many AI-assist requests you have made (for an extension without an account, counted against its random installation ID). Stored in Supabase.
• IP address — used to rate-limit abuse. For the AI-assist endpoint this is an in-memory counter only (60 requests per minute per IP), never stored. For the public feedback form, a one-way hash of your IP address — never the address itself — may be written to a small rate-limit table that is automatically deleted after two days.

How classification works, and third-party data sharing

By default, classification runs entirely on your device, for every plan — no video data leaves your device for the videos our on-device model can confidently place. Only for the smaller subset it's genuinely unsure about does a video's title and channel name get sent for AI assistance:

1. Sent from the browser extension to our secure Next.js server (hosted on Vercel).
2. Forwarded to one of several AI providers for classification, tried in order until one responds — Groq, Inc. (api.groq.com), Cerebras Systems, Inc. (api.cerebras.ai), Google Gemini (generativelanguage.googleapis.com), or OpenRouter, Inc. (openrouter.ai, which itself may route the request to one of several underlying model providers). Using more than one provider lets AI assistance keep working smoothly if any single provider is temporarily unavailable or rate-limited.
3. The category result is returned to your browser. We send the AI provider only the title and channel name — nothing that identifies you — and each provider handles that data under its own privacy policy (linked below); we do not control their retention. Algoright's own server logs the title, channel name, and assigned category for these videos — see "What we collect" above — since that's the training data we use to keep improving the on-device model these videos are sent to us to help with in the first place.

Groq's privacy policy: groq.com/privacy-policy · Cerebras's privacy policy: cerebras.ai/privacy-policy · Google's Gemini API terms: ai.google.dev/gemini-api/terms · OpenRouter's privacy policy: openrouter.ai/privacy

To opt out of AI assistance: In the extension popup, open Privacy and turn off AI assistance, or go to Dashboard → Settings → Privacy & Data → turn off AI Classification. When off, classification runs entirely on-device, all the time — no video data leaves your device, and nothing is logged as training data. Note this only affects AI assistance: submitting a category correction (✓/✗) still stores that video's title/channel and contributes to the shared classifier, as described above, since that's a separate feature from AI assistance.

What we do NOT collect

• The content of videos you watch
• Your browsing history outside of YouTube, TikTok, Instagram, and X
• Your private messages. The extension can hide a site's messages button, but it never opens or reads your messages
• Passwords (handled entirely by Clerk)
• Payment card details (handled entirely by Stripe)

Data storage and retention

Your account data and usage statistics are stored in Supabase (PostgreSQL), hosted in the EU. You can request full deletion of your account and data — including AI-assist training logs and category-correction records — by contacting us at the email below. Daily classification counters reset automatically every 24 hours.

Post titles and channel names: the AI-assist training logs and stored category corrections described above are deleted automatically 30 days after they are created. We keep them only that long because they come from other people's posts on YouTube, TikTok, Instagram and X, and we don't want to build up a lasting copy of that content.

Shared classifier data: words extracted from a category correction (not the full title) are aggregated into a shared database used to improve categorisation for every user, and decay automatically over time if not reinforced by further corrections — see the "Category corrections you submit" item above. Because this data is aggregated across users rather than tied to a single account, an account-deletion request removes your own correction records but does not retroactively remove already-aggregated shared signal.

Trial-abuse prevention: to stop the same person from repeatedly deleting an account to claim a new free trial, we keep a one-way, normalized record tied to the email used for a trial (for example, ignoring capitalization, dots, and "+" aliases on Gmail addresses). This record is not deleted when you delete your account. It cannot be used to identify you, contains no other personal information, and is used only to check trial eligibility on a future signup.

Cookies

We use only essential session cookies set by Clerk for authentication. We do not use advertising or tracking cookies.

Children

Algoright AI is not directed at children under 13, and accounts are not available to them. We ask for a date of birth before an account can be created, and again on the dashboard for any account that was created through a sign-in provider such as Google. The date of birth is used only to check eligibility — it is never stored or sent to our servers; we keep only a yes/no marker that the check was passed.

If we learn that an account belongs to someone under 13, the account and the data associated with it are deleted. Parents or guardians who believe we hold information about a child under 13 can email us at algorightai@gmail.com and we will delete it and confirm.

The browser extension can filter YouTube without an account. In that mode it sends us no browsing history and no account details. Once a day it sends an anonymous random installation ID with the extension version, browser name and simple counts (how many videos it has filtered, on how many days and on which supported sites, and whether it could find and read the posts on the pages it checked) so we can count installations and notice when a site change breaks filtering. From version 5.18, AI assistance also works without an account for a small number of videos a day: for a video the on-device model is unsure about, its title and channel name are sent for AI assistance as described above, with no account ID. Turning off AI assistance in the extension's settings stops this.

What the extension does on other sites

The extension only changes how YouTube, TikTok, Instagram and X look in your own browser: it hides or labels content, and it can take you to a site's own Following feed. It never posts, likes, follows, comments or sends messages for you. Hiding sponsored posts is off by default and only turns on if you switch it on. Those sites have their own terms, which still apply to how you use them.

Changes

We may update this policy. Significant changes will be communicated via the dashboard or email. Continued use after changes constitutes acceptance.

Contact

For privacy questions or data deletion requests, email us at algorightai@gmail.com.