Privacy Policy

Under the General Data Protection Regulation (GDPR), the controller of your personal data is:

We use your information to:

• Provide the service: Process and store your AI experts, manage your account
• Improve the platform: Analyze usage to optimize functionality and performance
• Communicate with you: Send updates, important notifications, and technical support
• Personalize your experience: Remember your preferences and settings
• Security: Prevent fraud, detect technical issues, and protect the platform
• Legal compliance: Respond to legal requirements and protect our rights

We do not sell your personal information. We only share data in these cases:

• AI model providers: Google, Cerebras, Mistral AI and OpenAI, each with the role and country detailed in section 3.1
• Other service providers: YouTube Data API, Stripe (payment processing), Resend (email), Cloudflare R2 (file storage), Google's Firebase Cloud Messaging (push notifications) and, only if you connect it and when you choose to send something, Notion (section 3.5). If you use those channels, also Meta's WhatsApp (section 3.8) and Telegram (section 3.9)
• Data processors for support and monitoring: AgenteUno (help chat, a product of Codelabs Studio S.L.; data on servers in the European Union, on our own infrastructure) and GlitchTip (self-hosted error monitoring, no personal data), as described in section 3.1
• Legal requirements: When required by law or to protect legal rights
• Business transfers: In case of merger, acquisition, or asset sale
• With your consent: When you explicitly authorize us

TubExperto does not train models of its own on your data. We send text to artificial intelligence model providers for these kinds of task. Each task has a primary provider and a fallback that only comes into play if the primary fails or runs out of quota:

• Catalogue indexing: transcripts, titles and descriptions of public YouTube videos, together with the chapters, summaries, analyses, infographics and audio we generate from them. This task involves no personal data of yours.
• Chat with your experts and research: your question or research topic, together with the fragments of indexed content retrieved to answer it.
• Voice: if you record a voice note, in voice interaction or in audio feedback, the recording is sent to a transcription provider to turn it into text.
• Semantic search: to locate the fragments that answer your question we compute vectors (embeddings) of the content and of the question itself. This task runs on an open model hosted on our own infrastructure, in the European Union, and never leaves it.

These are the providers involved today:

Swipe the table sideways to see every column.

Speech synthesis. If you use voice interaction or audio summaries, the text of the answer (not your recording) is sent to the Google Cloud text-to-speech service to be turned into audio.

Help assistant (AgenteUno). The conversational assistant on this site runs on AgenteUno. Chat messages are processed with Cerebras Systems Inc. (United States) as data processor. Cerebras Systems Inc. is not on the EU-US Data Privacy Framework (DPF) list: the transfer to this provider relies on the standard contractual clauses (Commission Implementing Decision (EU) 2021/914) incorporated into its data processing agreement (DPA). In voice conversations with the assistant, the audio and its transcript are processed by Google LLC (Gemini Live, United States), which is certified under the EU-US Data Privacy Framework (DPF): that transfer relies on the European Commission adequacy decision. AgenteUno keeps on its servers the text of your messages, the assistant's replies and the date and time of the conversation so it can help you; you can ask us to delete it. Please do not type in that chat data you do not want us to keep.

Data processors for support and monitoring. AgenteUno (the help chat) is a product of Codelabs Studio S.L., based in Málaga, Spain, and acts as our data processor: the messages and the voice conversations of the chat are stored on servers in the European Union, on our own infrastructure. GlitchTip is the error monitoring tool we host ourselves on that same infrastructure: it receives technical error reports and is configured not to collect personal data (default personal data sending is disabled and request headers, cookies and similar fields are scrubbed before they leave your browser or our server).

YouTube videos. The YouTube players embedded in TubExperto are loaded in enhanced privacy mode (youtube-nocookie.com), so YouTube does not set cookies for tracking your visit just by loading the player.

Anthropic. Earlier versions of this policy mentioned Anthropic Claude. No task is routed to that provider today: it is kept only as a contingency option, and this table will be updated before it is activated.

Important: we do not train models on your conversations and we do not sell them. Google, Cerebras and OpenAI state in their terms that they do not use the data they receive through their API for training. Mistral AI's paid-API terms let a customer opt out of training on its calls; we have not verified whether that setting is off on our account, so we do not claim that it is (see the table above). We no longer send content to Alibaba Cloud: we removed it from every fallback because we do not have safeguards for the transfer to Singapore. Except for the retention periods listed in the table, providers process the text to generate the answer at that moment and do not store it permanently.

TubExperto uses Cloudflare R2 to store the content generated from your activity on the platform: podcasts, infographics, and dubbing audio.

• Data stored: Podcasts, infographics, and dubbing audio generated from your AI experts
• Purpose: Securely host the media files you generate and make them available when you access them
• Retention: Files are kept while your account is active, per section 6 (Data Retention)
• Location: Cloudflare operates a global network of data centers; transfers outside the European Economic Area are covered by Cloudflare's contractual safeguards (Standard Contractual Clauses)

More information: Cloudflare Privacy Policy

You can connect your MemoryFirst account from Connections. It is optional and starts only with your explicit consent on MemoryFirst's own screen. Once connected, TubExperto holds a read-only permission over your memory and reads it only when you turn on "Use my memory" in a research, and only from the project you pick.

The memory fragments retrieved are sent to TubExperto's AI provider to produce the answer, the same way as the rest of the content we process (section 3.1). They are stored inside that research so you can see exactly what was taken into account, they are used for no other purpose, and they are deleted when you delete the research.

You can disconnect your memory at any time from Connections, or revoke the permission from your MemoryFirst account. On disconnection we stop reading it immediately and delete the access credentials we held.

Your conversations are private. Something is published only when you decide to and expressly confirm it: an answer (together with the question that produced it), a whole conversation, a video summary or a video's analysis. Anyone with the link can read it without signing in. The link carries a random identifier that cannot be guessed or derived from any other data.

• What is published: the title, the text of the chosen questions and answers (or the analysis text, for a video analysis), the names of the channels cited and the links to the videos and moments each citation relies on. The rest of the conversation's messages, its tags and your team are not published.
• Your name: by default the page shows as published by a TubExperto user. Your name is only shown if you tick the option to show it, and it is removed as soon as you untick it.
• It is a frozen copy: we store a copy of the content as it was when you published it, and the public page reads only that copy. Anything you write later in the same conversation does not appear. If you publish the same thing again, the existing link's copy is updated instead of creating another link.
• Search engines: by default the page may appear in search results. You can turn on the no-index option, and the page then asks search engines not to index it or follow its links.
• Revocation deletes the copy immediately, not just the link: you can revoke a link at any time, and deleting the conversation also revokes its links. From that moment the published text is deleted from our database in the same action that revokes it — not kept until you delete your account. What remains is only enough to know a link existed and is gone (so a repeated revoke attempt does nothing, rather than erroring). The page itself has stopped working for everyone before that: it is generated on every visit, with no cache, so a revoked link is already unreachable on the very next load. We cannot withdraw copies a third party made while the link was active (for example, a search engine cache or a screenshot).
• Visit metrics without cookies: so you can see how many people have read it, we count visits and clicks on the page's sign-up button. To avoid counting the same person twice we use, for six hours, an irreversible fingerprint of their IP address (a keyed hash) that expires on its own: the IP address is not stored, and for each link we keep only the number of visits, the number of clicks and the date of the last visit. Visits from known bots are not counted. If a reader later signs up, that sign-up is attributed to the shared link only when they accepted the attribution cookies (see the Cookie Policy).

You can connect your Notion workspace from Connections to send an answer, a conversation or a summary to Notion. It is optional and is authorised on Notion's own screen, where you choose which pages TubExperto may access: the integration can only see and write in the pages you share with it.

• What is sent to Notion: only the content you choose to send (title, questions, answers and citations with their links), and only when you press send. It is created as a new page inside one of the pages you shared; to pick it we query the list of shared pages and read only their identifier, not their content. Nothing Notion returns to us is ever sent to an AI model.
• What we store: the access tokens Notion issues, encrypted with AES-256-GCM (never in plain text and never returned by our API), the workspace identifier and name, the identifier of the page where we create exports, the date of last use and, if the connection fails, the reason for the error.
• Disconnecting: when you disconnect Notion from Connections we delete those tokens and the rest of the connection data. To also withdraw the permission on Notion's side, remove TubExperto in the connections section of your Notion account. Pages already created in your Notion are yours and stay there: TubExperto does not delete them or access them again.

More information: Notion Privacy Policy

If you schedule a quiz for review, we store a copy of each of its questions as a review card: the question, the options, the correct answer, the explanation, the topic and the expert or video it comes from. With each card we store when it is next due and how you rated it last time, and we log every review (the rating and the date) to compute your streak and progress. Nothing is scheduled on its own: quizzes you had already taken do not become cards unless you ask for it.

• Reminders: once a day we check whether you have cards due and, if so, send you at most one reminder that day, by email, by push notification in the app, or both. The reminder only says how many cards are due, not what they contain. To avoid repeating it we record the day and the number of cards due.
• Channels: email is sent through Resend (or, if that fails, our own mail server) and push notifications through Google's Firebase Cloud Messaging, to the devices on which you enabled the app's notifications.
• How to turn them off: in Settings, Notifications, Spaced review section, you can turn off the email reminder and the push notification separately.
• Deleting one card: from a review session (web or app) you can delete a single card outright, not just pause it — its question, options, answer and review history are removed immediately, not kept until you delete your whole account.
• Retention: cards and review history are kept while your account is active, you delete the specific card, or you delete the account, whichever comes first.

In Settings you can choose which model answers in the chat with your experts (also from the extension). We store your choice on your account and check it against your plan on every question. Choosing a model changes which provider processes your question and the context fragments retrieved to answer it:

• Automatic (default): the answer is generated by Cerebras; if that fails, by the Google Gemini API; and if that is not available either, by Google Cloud Vertex AI through our gateway.
• Gemini 3.5 Flash or Gemini 3.1 Pro: the question and context are sent to the Google Gemini API and do not go through Cerebras. If that model fails, the automatic route above is used so you are not left without an answer, so in that case the question may reach Cerebras.

The options offered are always Google models, and the model router refuses by code any choice that would send the chat to a provider not cleared for private conversations. Each provider's processing country and terms are those in the table in section 3.1.

We implement robust security measures:

• SSL/TLS encryption for all communications
• Secure storage on encrypted servers
• Restricted access only to authorized personnel
• Regular security audits
• Automatic backups and data loss protection

You have the right to:

• Access: Request a copy of your personal data
• Rectification: Correct inaccurate or incomplete information
• Erasure: Request deletion of your data (“right to be forgotten”)
• Portability: Receive your data in a structured, transferable format
• Object: Object to the processing of your data in certain cases
• Restriction: Request restriction of processing of your data
• Withdraw consent: At any time, for consent-based processing

To exercise these rights, contact us at Loading...