Privacy policy
Applies to befocusy.com, the campus and the WorkOS platform (labs.befocusy.com).
Last updated: 2026-11-18
1. Data controller
The WorkOS platform is operated by SAL DIGITAL SKILLS, S.L., tax ID B39855788, registered at Plaza María Blanchard 2, 3F, 39600 Maliaño (Cantabria), España. You can reach us at eduardo@befocusy.com for anything regarding this policy, and at soporte@befocusy.com for product support.
2. Our role
Our role changes depending on where you are, and the distinction matters. On befocusy.com and the campus we are the data controller: we decide what data we ask for and why. On the WorkOS platform it is different: each customer company uses it with its own accounts and databases, so with regard to the content it enters or connects we act as a data processor on its behalf, under article 28 GDPR, and the controller is the customer company. We are the controller for that company own contact and billing details.
3. What data we process
- Account data: name, email address, company and interface preferences.
- Connection credentials (OAuth): the access tokens for the third-party services the user explicitly authorizes. They are stored encrypted and never shared between customers.
- Working content: the content of the services the user connects — messages, calendar events, documents, files — displayed and organized inside the platform.
- Technical data: IP address, device identifiers and activity logs, for security and troubleshooting.
- Billing data of the customer company. Full payment card details are held by the payment provider; we do not store them.
- Newsletter subscription data: your email address and, if you give it, your name.
- Training data: if you enrol in the campus, your enrolment and your progress through the courses.
- Support chat: the messages the user writes in the in-app support chat and the replies received from the automatic assistant or the support team.
- Notifications: the alerts the application generates for the user (for example, that someone on their team has completed a task they created, or that someone has commented on one of their items), with the item title and the name of the person who acted or commented (and, for absence requests, the person who approved, rejected or cancelled it, visible only to the other people who can decide on that request), and their notification preferences.
- Browser notifications (push): if the client turns the feature on and the user allows them on their device, the delivery address assigned by the notification service of their browser (Google, Mozilla, Microsoft or Apple), that device's encryption keys and a label (browser and system). They are used to send encrypted alerts such as the end-of-day reminder. The user can turn them off at any time from the bell settings, and they are deleted when the account is deleted. A device's subscription is also removed when the alert is turned off, when the browser's notification service reports it no longer exists, or after repeated failures; the reminder counter is deleted after 14 days. The notification service of the user's own browser (Google, Mozilla, Microsoft or Apple) only receives the delivery address and technical data of the sending; the content of the alert travels encrypted and it cannot read it.
- Room booking: if the client turns the feature on, room bookings (room, day, times, optional purpose and the person who booked), visible to the client's team. They are deleted after 400 days.
- Last search: the text of the last search made in the WorkOS search (up to 150 characters), the result type chosen and whether other people items were included, so the search can reopen where it was left.
- Vacations and absences: if the customer turns the feature on, the absences requested or recorded (person, type, dates, half days, status, who decided them and an optional note), the company holidays and the annual day balance. If the customer sets up an absence type that requires a supporting document (for example, an exam certificate), also the attached file (PDF or image up to 5 MB), which is stored only in the platform database — never in Notion — and can be opened only by the person themselves and whoever can approve their absence; a sensitive type cannot require a supporting document. The customer may also allow it to be handed in later: it is then attached afterwards to the same request and, if it is still missing after 10 days, the person receives a single reminder notification. A type marked as sensitive (for example, sick leave) carries no reason or diagnosis: only the person themselves, the account Owner and their direct Supervisor see it by name; the rest of the team sees "Away". A copy of this data, without the type or note of sick leaves, is stored in the customer's Notion database. The AI tools (Kairos and the MCP connections) never receive or show the type or note of a sick leave. The birthday (day and month) of each person, if the customer stores it on their Team profile to create their yearly birthday absence, can only be seen and edited by the Owner and people with technical administration permissions; the absence it generates is treated like any other absence (it is visible to the person, their Supervisor and anyone who checks the team's absences, including through Kairos and the MCP connections, and it is copied to Notion with its type and date), and the birthday is deleted when the person is removed from the Team. On the factory kiosk, if the customer uses it and leaves that screen enabled, a person who identifies with their personal PIN sees on the shared screen only THEIR OWN absences (including sick leave, with its name and dates but no reason or diagnosis) and their day balance, and can request or cancel one; the session closes by itself after a few seconds of inactivity and the customer can turn the screen off. In addition, if the customer turns it on in their settings (it is off by default, and it is the customer, as employer, who decides to turn it on and to inform their team), any user of that same account can see a read-only view, «Team availability», with everyone's approved absences (name and dates and, depending on the option the customer chooses, the type name; types the customer marks as sensitive —by default, sick leave— appear only as «Absent», and with the «no type» option every absence appears only as «Absent»; notes, supporting documents, balances, pending or rejected requests and birthdays are never shown). Anyone who sees an absence as «Absent» without a type may infer it is of a sensitive type. Every change of this setting is recorded in the customer's audit log.
- Commercial information capture (if the customer enables this custom plugin): the assistant stores what the user writes, including the client company and the name and email of the contact people who will receive a proposal. Those contacts are linked to the document and, if the document is shared with the client through the portal, the people listed as contacts can access it. The text of the document, including that data and any name or email the user types by hand, is sent to the AI provider to draft the quote. If the quote is generated in a language other than Spanish, the fixed clauses of the proposal that the customer has configured (third-party expenses, withholdings, CPI/IPC adjustment, success fee, out of scope, in MXA Zona Franca proposals the acceptance and signature and the terms and conditions and, if any, Spanish-prevails) are sent to the same AI provider to be translated: they are generic customer texts, not data about its contacts, and a person must review the machine translation before the proposal is sent. When the client signs a proposal, the platform notifies, inside the application (and with a browser notification, if enabled), the owner of the document and the people who emailed it or copied its link; for this it stores the identifiers of those people and of the document (and deletes them when the person is removed or the document is deleted); the in-app notice and the browser notification, which travels encrypted, include the title of the proposal and the name the signer typed, and are deleted on the notification schedule (read ones after 30 days, all after 90). If the customer enables it, the signer also receives one automatic email, only once, sent on behalf of the customer company from the platform account, with replies going to the person who sent the proposal, thanking them and asking them to complete their company details in the portal. For each quote a technical record is kept (who generated it, when, the AI model, the language, whether any text was translated with AI and a cryptographic fingerprint of the fixed texts applied) and the customer keeps a history of changes to those texts (date, user identifier and text) for as long as its account exists; free text typed in the assistant is added untranslated. Contacts created automatically from an email and those marked as spam or “no AI” are not offered, not linked and not inherited by the quote (exclusion by origin depends on the customer having that data configured). The assistant's answers are kept for as long as the document exists and are deleted when it is deleted from the platform (deleting it directly in Notion does not remove them); if the right to erasure is exercised for a contact, their identifier, email and name are removed from those answers when they match what was stored, but the text already written in the body of the document or in the generated quote is not changed and must be corrected or deleted separately. If the company record states whether it is a natural person or a legal entity, that data is only used to preselect the client profile. Each link is recorded in the activity log (identifiers only, no names or emails). The proposal can also be emailed to those contacts from the platform: the message is sent from the platform mail account, with the signature and reply address of the sender, and carries the portal link (never an address typed in the browser). No AI provider is involved. The message text is not stored; a record is kept of who sent which proposal, when and to how many people (with shortened addresses).
- Kairos daily summary: if the customer turns the feature on, the summary Kairos prepares each morning for each user (a context sentence and up to eight actions, each with the title of the email, task or event it refers to, the reason and the priority) and the actions the user ticks as done. To prepare it, the subjects, senders and dates of the user's pending emails, and the titles and dates of their tasks and events, together with the company's current objectives (name, short description, term, deadline, progress figures and linked projects) used as a priority criterion, are sent to OpenAI under the user's own permissions; never the content of the emails.
- Voice-created publications with cited pages: when a publication is created by voice, the WorkOS pages the person cites with “@” (wiki, notes, documents, projects, companies and publications, read with their own permissions) are sent to OpenAI to correct the transcript and, if they are not private or from email, are linked (their content is not copied) in the publication’s research toggle, in the customer’s Notion, and the drafting reads them in full and up to date when the post is generated, with the visibility of the person who cited them, checked when they were cited (MariaDB keeps, with no text, which pages that person validated for that publication, together with their user identifier; it is valid for 7 days for drafting and is deleted after 30 days or when the person is removed). If the publication’s drafting can use an AI provider other than OpenAI, no page is linked. Contacts, meetings, emails, messages and pages generated by Cerebro IA are not linked. The link lives with the publication, can be seen by anyone with access to it or to the customer’s Notion workspace, and is removed by editing the toggle.
- Reading links in Publishing (optional): if the customer enables it, when drafting a post or using “Improve with AI”, the OpenAI model may open and read the web pages the person linked in the publication’s research (only those addresses; URL parameters that look like keys or signatures are removed). To do so, the addresses and the searches the model makes are sent to OpenAI and its search service, and the visited sites receive the request. The resulting text is a proposal that a person reviews before publishing. None of this is stored in our logs other than counts.
- Improve with AI (Publishing): if the customer uses social media, when a person presses “Improve with AI” on a post that is in review, the following are sent to OpenAI: the post text on the person’s screen, the publication’s research (with the pages linked in it and their attachments), the content of up to five WorkOS pages the person cites with “@” (wiki articles, notes, documents, projects, companies, contacts, meetings and publications, read with the person’s own permissions) and the instruction they write. Emails and messages cannot be cited with “@”. Pages linked in the research are only used if they are not private and do not come from emails, meetings or contacts created from email and, when the person uses the button or Kairos, if that person can read them; the rest are skipped (also tasks and pages from unrecognised databases). Email conversation summaries are removed from a linked project, and when the research is saved the links that are not allowed are removed. The result is only a proposal that the person reviews and saves themselves; nothing is published on its own. If the person asks for it, their instruction is saved in the research section of the publication, in the customer’s Notion. The person can also ask Kairos to do it: in that case the same text and the same research are sent, not the pages cited with “@”, but what Kairos has read and summarized in the instruction is (which may include content from emails, meetings or records if the person asked for it); Kairos only saves the text if the person confirms it, and the proposed text stays in the conversation with Kairos, with the same retention. The AI usage log records only the model, the tokens and the user, never the text. The same happens when a publication is moved to “Ready to write” or “Regenerate with AI” is pressed: the publication’s research (with its linked pages and attachments) and the customer’s style personalisation are sent to OpenAI to write the post again, and the result replaces the current text; a person reviews it before it is published.
- Library: if the customer turns the feature on, the links, texts and images (for example, screenshots) each person saves, with their title, notes, tags (the Digital Garden topics), collections, date and who saved them. Each item is private to the person who saves it unless they share it with the team. That privacy between people is enforced by WorkOS: anyone with direct access to the customer’s Notion workspace or to our database can see what has been saved. It is stored in the customer’s Notion (a page with the text or the file) and as a copy in our database (metadata, text for searching and a thumbnail of images); unless the customer turns on link reading (see below), WorkOS does not read the content of linked pages. When an item is deleted it is removed from our copy and its Notion page is archived; when a person is removed, what they saved privately is archived in Notion (recoverable for 30 days from its trash) and deleted from our copy, and what they had shared with the team passes to the person who takes over their work. Notion may keep copies in its version history and trash for the period set by its own policy. If a YouTube link is saved, the browser of the person using the Library loads the video thumbnail directly from YouTube (Google), which receives their IP address and the technical data of the request (WorkOS does not send the referring page), whether or not link reading is turned on. Saving from your phone: the Android “Share” button only opens WorkOS with the data pre-filled (the person decides and saves from their own session); for iOS Shortcuts and similar tools, each person can create personal tokens: we store only an irreversible fingerprint of the token (SHA-256), the name the person gives it (avoid personal data) and its creation, expiry, last-use and revocation dates; they expire after 365 days, can be revoked (they are revoked automatically when the person is deactivated and deleted when they are removed) and expired or revoked ones are deleted after 30 days; they only allow saving to that person’s private Library (they cannot read anything or share) and, when one is created, its owner gets a bell notification; attempts with an invalid token are limited by IP address, like other access. With the Android “Share” button, the shared content travels once in the request address, so it may remain in the web server’s access logs for their rotation period. AI reading of images (off by default; the customer turns it on): if it is on, when an image is saved that image —re-encoded and without camera metadata— and the names of the account’s topics are sent to OpenAI to transcribe the text it shows, describe it and suggest a title and tags; the result is stored with the item for as long as it exists and the person decides whether to apply the suggestions; nothing is sent to other AI companies, there is a daily spending cap per customer and the AI usage log records only the model, the tokens and the user, never the content. In the Library’s AI features (images, text summaries and convert to note), OpenAI receives the data with its no-storage option: it keeps them only as long as needed to handle the request and, for up to 30 days, to detect abuse, and does not use them to train its models. If Library items are cited when an AI-written post is created, their title, link, the person’s notes and texts, and the summaries, the description and the text read from the person’s own screenshots are sent, together with the rest of the post, to the AI provider that writes it (never the text extracted from third-party pages and videos; from an item shared by another person only its title, link, notes and text are used). A screenshot may show third-party data: the customer decides what is saved. That material is also written into the post’s research toggle in Notion, which anyone with access to that post can see, and a record is kept of which items were used in each post (only the item, post and person identifiers, and the date); that record is deleted when the item is deleted or the person is removed. Reading links (off by default; the customer turns it on): if it is on, when a link is saved WorkOS downloads that page from its servers (public https pages only, without cookies or credentials; the source site receives our server’s IP address and a “WorkOS-Biblioteca” identifier) and stores with the item the article text and its metadata (title, author, site and description); for YouTube links, our servers also ask YouTube for the video’s title and author (oEmbed service, which receives the server’s IP address and the video identifier); the text is also copied to the customer’s Notion page, as plain text only (no links or images). If AI is also turned on, that text —which is third-party content— is sent to OpenAI to obtain a summary, a title and suggested tags; and, for links to public YouTube videos, the video link is sent to Google (Gemini) in segments; Google retrieves and analyses the video (including the voices of the people who appear in it) to obtain a verbatim transcript. Suggestions are never applied automatically, the same daily spending cap applies, the text is sent only to OpenAI and a video link only to Google (and only if the customer also turns on video transcription, with a Gemini model assigned by the administrator), and the usage log records only the model, the tokens and the user. If a page states that it does not want its content copied or used with AI (a “noai” tag or rights reservation), it is not copied; nor is a page that the site’s “robots.txt” file does not allow to be downloaded. The text extracted from someone else’s page or video is private to the person who saved the link (while it exists it cannot be shared with the rest of the account), is deleted automatically after 90 days (or the period agreed with the customer), and the person can delete it at any time with the “Delete the extracted text” button (its copy in the Notion page is also deleted, if it is the one made by WorkOS); if the customer turns off link reading, all extracted text is deleted. The link, notes and summary are kept. Copy to Google Drive (optional, per person): if the person picks a folder in their own Google Drive with one of their connected accounts, the images they save from then on are copied there and, when the item is deleted, the copy is deleted permanently (without going through the Drive trash); if the item disappears from Notion, the copy is deleted 7 days after it is archived there (or 30 days if Notion can no longer find it), to leave time to recover it; if at that moment the Google account is disconnected or Drive does not respond, it is retried and, meanwhile, the copy stays in the person’s Drive, who can delete it by hand; if the person is removed, the file already copied stays in their Drive. Convert to note (off by default; the person asks for it item by item and it requires the Library’s AI): the item’s content —the article text or transcript, the text the person wrote or what was read from an image, together with its title and notes, up to 150,000 characters— is sent to OpenAI to generate a detailed summary, which is saved as a new note in the customer’s Digital Garden (in Notion), with the item’s tags and a link to its source. The note is independent of the item: deleting the item or removing the person does not delete the note. The generated summary is kept in our database only until the note is created (and, if the process fails, for up to 3 days); nothing is sent to other AI companies, the daily spending limit per customer applies, and the usage log records only the model, the tokens and the user. RSS feeds (off by default; the customer turns it on and each person adds their own): when a person adds the address of an RSS or Atom feed, our servers download it once a day from the originating site (with a conditional request, identifying themselves as WorkOS; the originating site receives our server’s IP address) and store, for that person only, the address and title of each feed (they may reflect the person’s interests: only they see them in the application, not even the administration of their account; only direct access to our database could show them) and, for recent entries, the title, the link and a short plain-text excerpt, as items pending review. Nothing is sent to AI companies. Accepting an entry creates a private link in their Library (and, if the customer has link reading turned on, that feature’s rules apply); when an entry is accepted or dismissed its text is deleted immediately and only a fingerprint remains so it is not offered again, which is removed after 90 days (365 if the feed does not date its entries); anything not reviewed is deleted after 60 days. A person’s feeds and pending items are deleted when the feed is removed, when the person is offboarded, 60 days after they are deactivated and 7 days after the customer turns the feature off; a feed that cannot be read is removed automatically.
- Publications and AI assistants: Kairos and the MCP connectors the user connects can search and read, with the user’s own permissions, the text of their publications in any status —including drafts and posts in review— but not their research. That text reaches Kairos’s AI provider or the connected assistant, like any other content the user asks them to read.
- Day close: if the customer uses time tracking, the day on which a person validates their hours, the total minutes and number of records validated, a technical fingerprint of those records (to detect later changes) and the time of the validation.
- Document signing: when a person approves a document in the client portal, their name, email, identity document number (stored encrypted), IP address, browser and the time of signing, together with a sealed copy of the document and a chained record that only keeps a fingerprint of the signer.
- Team profile: name, email, area and, optionally, the mobile phone number of the people on the client team.
- Email drafts: the drafts Kairos prepares at the user request, stored encrypted until they are sent or discarded.
- Time tracking and production log (optional module): only if the client company enables this module for its staff. It includes each person's name, the time and type (clock-in or clock-out) of every entry, manual corrections with their reason, the device they were recorded from, the quantities of pieces produced and a log of the administration actions on the clock-in terminal. The personal PIN or code is stored as an irreversible hash and as an encrypted copy that only the client company's administrators can view, and every view is logged. If the company also enables location when clocking in, the device asks for its position at the moment of clocking in and only with the device's own permission; approximate coordinates (rounded to about 100 metres), their accuracy and, if the company has defined its workplace, the distance to it are stored. There is no continuous tracking: the location is read only when the person presses clock in. From the clock-ins the system calculates, for each person and month, the hours worked against their schedule (deducting public holidays and approved absences; sick leave appears only as 'absence', with no type or reason), the hours balance that carries over from one month to the next, and the days with a clock-in but no clock-out, which do not count until a manager corrects them. This balance is not stored separately: it is calculated on every request from the clock-in ledger and only over the current calendar year: the balance resets to zero on 1 January and counts from the start date set by the company. The factory team each person works in each week is also stored (by rotation or by a manual change by a manager, with the reason and who made it); each worker sees the names of their team mates at the clock-in terminal. Each person sees only their own hours and balance at the clock-in terminal, after identifying themselves with their PIN; the client company (owner and technical administrators) sees those of its whole staff and receives a notice of clock-ins without a clock-out, and every report download and every view of the whole staff's hours is logged. In this module the data controller is the client company, as the employer, which decides the legal basis, informs its staff and handles their rights; we act as processor.
- Fathom meetings: if the person connects Fathom, their API key (stored encrypted) and, for each meeting, its title, date, attendees (name and email), link, summary and action items, which are copied as the meeting page in the customer Notion. The record of already imported meetings keeps only internal identifiers.
- Client profile (optional module): if the customer enables the feature, the identification and regulatory compliance data of its own clients (type of person, document, date of birth, address, occupation, source of funds, politically exposed person status, partners, representatives and ultimate beneficial owners), entered by the team or confirmed by the person when signing an offer in the portal. The content of the profile is stored encrypted on the platform; the company's own data (legal name, tax ID, address, phone, email, website, activities and income) is also stored on the company record in the customer's Notion. We record who last saved the profile and, for the declaration the person confirms when signing, their email, the date and their IP address. They are used only for the due diligence required by regulation.
- Economy invoices: if the customer uses the Economy module and connects Qonto, the issued invoices it imports from Qonto or records in its Notion (number, dates, status, amounts and lines), the client data of each invoice (name or legal name, tax ID, address and email, which may belong to individuals) and the links between CRM companies and Qonto clients. They are stored in the customer Notion database and in a copy in the platform database so they can be shown quickly. WorkOS reads those invoices from Qonto with the credential the customer connects and only sends data to Qonto when the user expressly asks for it (for example, to create a client or issue an invoice). This data is not sent to AI assistants or MCP connectors.
- SEPA direct debits and invoice sending (Economy): if the customer also connects its Qonto account with OAuth authorisation, WorkOS reads live from Qonto the direct-debit mandates of its clients (reference, status, dates, the last 4 digits of the IBAN and the signature link; they are not stored on the platform) and, only when an administrator expressly asks for it, creates or cancels mandates and collections. For each collection created, the platform stores a record with the invoice, amount, date, status, reference and the user who created it, so that nothing is collected twice. When a user sends an invoice by email, a record is stored with the invoice, the date, the user and the number of recipients (not their addresses); the email is sent from the user's own account.
- Cerebro AI (company profile generated with AI): if the customer turns it on (it is off by default; the customer chooses the company statuses that turn it on, and every company in those statuses is included, including those that later change to one of them, except the account’s own company and those the customer excludes one by one; the Owner is notified when the number of active companies grows), WorkOS reads the work content it already holds about one of the customer companies —document bodies, shared notes, meetings and project descriptions, plus the names of the company, its projects, tasks and contacts—, sends it to OpenAI after masking amounts, emails, phone numbers, IBANs, tax IDs, links and credentials, and stores the resulting facts: short sentences with their origin and a confidentiality level. Projects, documents and notes marked private, documents of restricted types, financial and tax data, email and messaging are not processed, and automatic filters discard texts about payroll, personal finances or health (they reduce the risk, they do not remove it). As an additional option, off by default, the customer can allow meetings and documents to be sent to OpenAI with their amounts and percentages unmasked (for documents, in a separate copy used only to extract their figures); those figures are stored only in strategic-level facts, which only the Owner can see and which are not published to the Wiki. Confidential and strategic facts are stored encrypted. In addition, people with access to the company record can add information to Cerebro AI by text or voice: what they type or dictate (voice is transcribed with OpenAI) is sent to OpenAI to turn it into facts of the chosen level and is stored as a team fact, together with the identifier of the user who added it. Meetings and notes may contain information about third parties (for example, attendees); the customer is responsible for having a legal basis for that content.
- Emotional state of the sender (inbox with AI): if the customer has the inbox turned on and the sender of an email is already a contact in their CRM, the AI infers from the tone of that email an approximate emotional state of the sender (one of 14 options, such as grateful, worried or frustrated; none if the tone is neutral or there is no human tone) and WorkOS shows it as an emoji next to the email, in the inbox and in the email view. Only that label is stored, with no text, in the platform database and as the icon of the email page in the customer Notion. It is not inferred for senders who are not in the CRM, nothing about health is inferred, and it is used only to help the customer serve their contacts better: it is not used to make decisions about people, for advertising, or shared with anyone other than the AI provider that already processes the email. The customer, as controller, must inform their contacts; the data subject can request its deletion (see section 7).
- Temporary performance measurement: for the customer companies designated for it, and only while the diagnosis lasts, we record the application's response times. On the server side, for each request: its duration, the number and type of calls to connected services (with identifiers replaced by generic placeholders) and the waiting time. On the browser side, for each user action: the section, the names of the functions called, their duration, the response status, the device type (touch or mouse), the network type declared by the browser and the time until the screen is painted, with the time rounded to the hour. They do not include content, texts, record identifiers, IP addresses or the identity of the person, and are used solely to diagnose and improve the speed of the service.
- Daily factory team report (optional module): only if the client company turns it on (it is off by default). Each factory team sends, from the time-clock station, a report of the day with the people who worked, the pieces that moved on to the next stage in the orders, warehouse production and, where applicable, extra tasks with a short note that can be typed or dictated by voice. It is a team measure, not an individual performance measure: WorkOS calculates no score per person and makes no automated decisions about anyone. We store who was present, who sent the report and at what time, the content of the report and the text of the note; the audio of a dictated note is sent to OpenAI only to transcribe it (see section 6) and is not kept. Team members see, at the time-clock station, the names and the content of their own team's report. We ask that notes do not mention health data or other people. For this module the data controller is the client company, as the employer, which decides the legal basis, informs its staff and handles their rights; we act as processor.
- Requests from interested people (custom plugin): if the customer enables this plugin, the requests it receives from people interested in its products: name, email, phone, project or campaign, stage, the person in charge, whether they gave consent to receive communications, and the dates of the last contact and of the next follow-up. They are stored in the customer's Notion and in a copy in our database so they can be listed and filtered; the copy does not include the message written by the interested person or the comments. Each team member sees the requests assigned to them and the account administration sees all of them; WorkOS enforces that visibility.
4. What we use them for
Only to provide the features the user has requested: displaying and organizing their work in the interface, synchronizing it with their Notion workspace, sending the messages they compose, managing their calendar events and keeping their files organized. Also to bill the service, meet our legal obligations and keep the platform secure. When the customer company turns on the inbox, WorkOS also automatically processes each incoming email with AI (see section 5). We do not build advertising profiles. If the customer turns on Cerebro AI, WorkOS also generates internal summaries of each customer company with AI from its work content (see section 3); no decisions with legal effects are taken about individuals and they are not used for advertising. If the customer has the inbox turned on, the AI also infers an approximate emotional state of senders who are already contacts in their CRM, only so that the customer can improve the relationship with them (see section 3); no decisions are made about people using that data.
4b. Support form
If you write to us through the support form on our public page, we process the name, email address and message you provide for the sole purpose of handling your enquiry and replying to you. The legal basis is your own consent when submitting it. We keep that correspondence for as long as necessary to resolve the matter and, afterwards, for the period during which liabilities may arise. We do not use it to send you marketing communications.
4c. Newsletter
If you subscribe to our newsletter on befocusy.com, we process your email address (and your name, if you give it) for the sole purpose of sending it to you. The legal basis is your consent, given when you sign up; if you are already a customer, we may inform you about services similar to those you have contracted, under article 21 of the Spanish LSSI. Delivery is handled by Substack, acting as a data processor, which does not use your address for anything else. We keep your data until you unsubscribe, which you can do with the link at the bottom of every newsletter or by writing to us. Newsletters may include open markers and tracking links that tell us, in aggregate, whether the message was read.
5. Google API data
WorkOS only accesses the Google accounts the user explicitly connects, and only the services authorized on the consent screen: Gmail, Google Calendar and Google Drive. Access can be revoked at any time from the Google account settings, and also by disconnecting the account inside WorkOS.
- Sign in with Google. If the user chooses “Continue with Google” to sign in to WorkOS, we only receive their email address (verified by Google) and their name, and use them solely to match an existing WorkOS account. We do not store any token from this connection or access any other data in their Google account.
- Gmail. We read the messages of the connected account to display them in the WorkOS inbox, link them to the matching contact and save them as records in the customer company's Notion workspace. If the company has the inbox turned on, each new email is automatically processed with AI to classify it, clean it up or translate it and, where relevant, suggest a task. If the sender is already a CRM contact, the AI also infers the tone of the email as an approximate emotional state (an emoji) so the customer can serve that contact better. Attachments are stored in Notion and, if the company sets it up, in its Google Drive. We mirror back to Gmail the actions the user performs here: marking as read, archiving, labelling or moving to trash. We only send messages the user composes and confirms; we never send email automatically or in bulk.
- Google Calendar. We display the calendars the user selects and create, reschedule or cancel events at their request, including attendees and video-call links. If the user marks a calendar with “creates task”, WorkOS periodically reads its events for the next 30 days and creates a task in Notion for each one, keeping its date up to date. To make the calendar load quickly, WorkOS temporarily stores, encrypted on its servers, the access token for that account (up to 50 minutes) and the list of events requested, with their title, time, location, description, links (to the event and to the video call) and attendee email addresses (up to 45 seconds). This copy becomes unreadable when it expires, is deleted from the server within 24 hours at most and is not used for any other purpose.
- Google Drive. We work inside the folder the user designates: we create its structure, upload attachments there, read the documents they choose to import, and move or rename folders when they reorganize their notes. We also detect files the user adds by hand to those folders and attach them to the matching record in Notion. Files that already existed in their Drive are neither moved nor deleted when imported. If the user deletes an attachment from WorkOS, that file is also permanently deleted from their Drive, without going through the trash. If the person chooses a folder in their Drive for the Library, we show the names of their folders and shared drives so they can select it, and we upload copies of the images they save there; when the item is deleted, the copy is deleted permanently (also if it disappears from Notion, 7 days after it is archived there or 30 days if Notion can no longer find it). We do not read any other files in their Drive with this function.
Limited Use
SAL DIGITAL SKILLS, S.L. (WorkOS / Befocusy Labs) use of information received from Google APIs, and its transfer to any other application, adheres to the Google API Services User Data Policy, including its Limited Use requirements. Specifically, data obtained from Gmail, Google Calendar and Google Drive:
- Are used only to provide and improve the features the user has explicitly requested, or that their company has turned on, and that are visible in the interface.
- Are not transferred to third parties, except to the providers strictly necessary to deliver those features, where required by law, for security reasons, or with the express consent of the user.
- Are not used or transferred for advertising purposes, including personalized advertising and remarketing.
- Are not read by any human, except with the express consent of the user, for security reasons, to comply with the law, or when aggregated and anonymized for internal operations.
- Are not used to develop, improve or train generalized artificial intelligence models. When a feature requires AI processing — for example, classifying an incoming email, summarizing it or structuring a document into a note — the content is sent to the AI provider solely to generate that specific result, which is delivered to the user or their company; those providers do not use the data to train their models and only keep it temporarily to detect abuse, under their own terms.
6. Providers involved
To deliver the service we rely on the following providers, each under its own data processing agreement:
- Hostinger — platform hosting, within the European Union.
- Google LLC — the user Gmail, Calendar and Drive, when they connect their account.
- Microsoft — email and calendar, as an alternative to Google.
- Notion Labs — destination of the customer working content.
- OpenAI and Google (Gemini) — text generation, summarization, audio transcription, transcription of public YouTube videos (Library only) and real-time web search for the AI features (including the Kairos assistant and, if the customer enables it, reading the web pages linked in a publication’s research). They do not train their models on data sent through their APIs.
- Zernio — WhatsApp Business and Instagram messaging, social media publishing, and storage of the attachments WorkOS hosts to display them in the platform.
- Slack — channels and direct messages, when the user connects their account.
- Stripe — payment gateway that some of our customers activate, through a custom integration, to bill their own end customers; it does not process your WorkOS subscription payment.
- Qonto — banking and e-invoicing that some of our customers connect to manage their own invoices. WorkOS reads from Qonto the invoices and their client data in order to import them and, only when the user asks, creates clients or issues invoices in that account. If the customer also connects its account with OAuth authorisation, it also manages SEPA direct-debit mandates and creates direct-debit collections, always at the express request of an administrator.
- Kit.com — newsletter delivery, when the customer uses that feature.
- Substack — delivery of our newsletter.
- Time-stamping authority (FreeTSA by default) — certifies the date of document-signature records. It receives only a fingerprint (SHA-256 hash), never the content of the documents or personal data.
- Brevo — marketing email platform, only if the customer turns on contact synchronization: it receives the email, name and phone number of the contacts the customer itself chooses to synchronize.
- Anthropic (Claude) and OpenAI (ChatGPT) — only if the user connects WorkOS to their AI assistant through a connector. This is a transfer that the user initiates and expressly authorizes: the tools they allow return to that assistant the data they ask it to look up (which may include emails, documents or tasks), under the terms of the assistant they choose. They can revoke the connection at any time.
- GNS — Latin American ERP, only if the client connects its account: WorkOS reads its customer list (name, tax ID, email, phone and country) to synchronise it with the CRM; no data is sent to GNS.
- Fathom — meeting recorder, only if the person connects their own account with their API key: WorkOS reads from Fathom the meetings recorded by that account (title, date, attendees, link, summary and action items) to create the meeting page in the customer Notion. Fathom is only sent the API key, the read requests and the webhook address. The summary and action items are sent to OpenAI or Google (Gemini) to translate them into the customer language.
- OpenAI (Cerebro AI) — for this feature the already cleaned content is sent with OpenAI no-storage option: it is retained only as long as needed to serve the request and, for up to 30 days (unless the law requires keeping it longer), for abuse monitoring. It is not used to train its models.
- OpenAI (daily factory report) — if the client company turns on the daily report and a person dictates a note, the audio is sent to OpenAI, using the platform account, solely to turn it into text; WorkOS does not store it. OpenAI may temporarily keep the request data for security and abuse-detection purposes, under its terms, and does not use it to train its models.
We never sell data to third parties under any circumstances. Some of these providers are located outside the European Economic Area; in those cases the transfer relies on the European Commission standard contractual clauses or on an adequacy decision.
7. How long we keep them
- Connection credentials are deleted as soon as the user disconnects the tool from WorkOS. If they revoke access from their Google account, the credentials stop working immediately and are deleted when the account is disconnected in WorkOS. Signing in with Google does not store any credentials.
- Working content is kept while the account is active. When the relationship ends it is deleted or returned to the customer, as they instruct.
- Technical logs are deleted automatically after 90 days at most.
- The temporary calendar copy (Google or Microsoft) expires on its own: the access token after 50 minutes at most, and the list of events after 45 seconds; what has expired becomes unreadable and is deleted from the server within 24 hours at most. If the user disconnects the account, that copy is no longer used and disappears when it expires, never later than those 50 minutes.
- The temporary copies of hours and timers (team workload, "My hours" and the "Close the day" status) are stored encrypted on the server for at most 30 minutes (90 seconds for the current day and for open timers; 60 seconds for the "Close the day" status) and are invalidated when time is recorded from WorkOS; what has expired becomes unreadable and is deleted from the server within 24 hours at most.
- Conversations with the Kairos assistant (including the content of emails, documents and attached files or other data the user asks it to look up) are stored by OpenAI for up to 30 days to keep the conversation thread, and are deleted afterwards. OpenAI does not use them to train its models or for any other purpose. When the user attaches a file, it is sent to OpenAI together with their request; Word, PowerPoint, legacy Excel and CSV documents are sent as they are, so they include their metadata (for example the author, comments or tracked changes). WorkOS does not keep the attached file: it deletes it from its servers as soon as it has been processed and, if it is never sent, within 24 hours at most.
- If the customer company turns on reading of record attachments, Kairos and MCP connectors (Claude, ChatGPT or others the user connects) can read, at the request of the user and only in records the user can already access, the files attached to notes, wiki, documents, projects, tasks, companies and contacts. PDFs, images and Office documents are sent to OpenAI to be transcribed. In Kairos, the text read becomes part of the conversation (kept at OpenAI for up to 30 days, as stated above); through MCP, it is handed to the assistant the user chose to connect (Claude, ChatGPT or another), which handles it under its own policy. WorkOS does not keep the file or the extracted text. Attachments that arrived from an email are never read with AI.
- WorkOS also keeps the history of conversations with Kairos so the user can pick them up again. Nobody at the customer company can read it, not even whoever administers the account. The WorkOS technical team may only consult it when the user themselves asks for it to resolve an issue, to investigate a security problem or abuse, or when required by law, and every access is logged.
- If the customer company turns it on, Kairos can generate spreadsheets and presentations (Excel, CSV or PowerPoint). To do so, the data the user asks it to include is processed in an isolated OpenAI environment with no internet access, which is deleted after about 20 minutes without use. WorkOS keeps the generated file so that only the person who asked for it can download it, and deletes it when the conversation is deleted or after 30 days at most.
- Billing data is kept for the periods required by commercial and tax regulations.
- Support chat conversations are kept while the account is active; they are deleted automatically after 12 months without activity. Automatic replies are generated by an AI provider (OpenAI or Google) solely from the conversation itself and the public Help Center articles: no data from the user’s workspace is sent to it.
- Notifications that have been read are deleted automatically after 30 days, and all of them, read or not, after 90 days.
- The last search is replaced by the next one, is cleared if the user empties the search box before closing it, and is deleted together with the user account.
- The absences, day balances and holidays are kept while the person is part of the customer's team: a person's absences and balances are deleted when they are removed from Team, and all of it when the customer is deleted. The supporting document of an absence is also deleted when the absence is cancelled or rejected and, in any case, 12 months after the absence's end date (the absence is kept; only the file is deleted).
- The Kairos daily summaries are deleted automatically after 90 days and, earlier, when the user or the customer is deleted. The platform administration team cannot read them: it only sees whether they were generated correctly and their AI cost.
- Day validations are kept while the person's account exists and are deleted together with it (or with the customer).
- Brand-framed images (the "Adapt to my brand" option in Publications) are stored on the server so they do not have to be generated again; they are deleted when the original image is replaced or removed and, in any case, after 60 days without use.
- The data of a document signer (name, email, encrypted document number, IP and browser) and the sealed copy are kept for as long as the signed document must remain verifiable; the chained record only keeps a fingerprint of the signer, so the signer data can be deleted or anonymised on request without breaking the chain. Encrypted backups of this data are kept for 60 days.
- Email drafts stored by Kairos are encrypted and deleted automatically after 30 days without being touched.
- The history of conversations with Kairos stored in WorkOS is deleted automatically when a conversation has had no activity for 12 months, and earlier if the user deletes it.
- The access-attempt records we use to limit abuse are stored only as a technical fingerprint (without the IP or email in clear) and are deleted after 24 hours.
- In the optional time tracking and production module, location coordinates are deleted automatically after 60 days. The rest of the records (clock-ins, corrections, production and clock-in terminal audit) are kept while the client company's account is active, because the clock-in ledger is immutable by design (corrections are added, not edited), and when the relationship ends they are deleted or returned to the company. The hours balance is calculated only over the current calendar year (it can be viewed up to 24 months back); this is not a retention period for the ledger. The same applies to the log of changes to the factory teams, which is immutable by design and is kept while the account is active. The legal retention period for these records is determined by the client company as the employer.
- Economy invoices and their lines are kept on the platform while the account is active (and in the customer Notion until they delete them). The temporary import queue from Qonto, which holds the client data of each invoice, is emptied as soon as each invoice is imported and, in any case, deleted entirely after 30 days.
- The direct-debit collection record is kept while the account is active, because it is what guarantees that an invoice is not collected twice. The invoice email sending record (invoice, date, user and number of recipients) is deleted after 365 days.
- The Qonto client data from the initial scan of "Initialise the economy" (name, tax ID, email and address) is deleted after 30 days at the latest. The customer profile is kept while the customer's account is active and deleted when the account is deleted; earlier, at the customer's request.
- Cerebro AI facts are kept while their source exists: they are deleted when the source changes, is archived or becomes private, when the company is archived, when a contact exercises the right to erasure or a team member is removed (the facts that cite them or mention them by name are deleted) and when the account is deleted. The run log is deleted after 180 days and the internal state of inactive companies after 400 days. When a company leaves the activated scope (it changes to a status that is not activated, its status is removed, it is archived or it is excluded), its Cerebro AI facts are deleted and its Wiki profiles are withdrawn automatically after 30 days (if it comes back sooner, the period is cancelled); the Owners are notified when the period starts and when the removal happens.
- The emotional state of the sender is kept for as long as the email it belongs to exists: it is deleted when the email is deleted or archived from WorkOS and, at the request of the data subject, when the right to erasure of a contact is exercised (it is cleared from all their emails and the icon of each Notion page goes back to the default), as well as when the customer account is deleted.
- The Library items (links, texts, images and their thumbnails) are kept for as long as the item exists and are deleted when it is deleted. The extracted text of other people’s pages and videos is deleted automatically 90 days after it was last read (between 7 and 365 as agreed with the customer), when the person presses “Delete the extracted text”, or when the customer turns link reading off. Capture tokens expire after 365 days, and expired or revoked ones are deleted 30 days later. For RSS feeds, unreviewed entries are deleted after 60 days; an entry’s text is deleted when it is accepted or dismissed, leaving only a fingerprint for 90 days (365 if the feed does not date its entries); the feeds of a deactivated person are deleted after 60 days, and all of a customer’s feeds 7 days after the customer turns the function off. The draft summary of “Convert to note” is kept only until the note is created and, if the process fails, for 3 days. Server backups (weekly snapshots, of which the last two are kept) may still hold deleted data until they are replaced. In addition, an encrypted copy of the Library tables (without the extracted text of third parties or the “Convert to note” drafts) is made every day and kept for 7 days on the same server; it may hold deleted data until it expires, and restoring it does not recover what belonged to people who have been removed.
- The temporary performance measurement data is kept for 30 days at most, and the measurement switches itself off after 14 days unless it is expressly renewed.
- The notes of extra tasks in the daily factory report are kept for 90 days and then deleted automatically (the extra task and the rest of the report remain). The audio of a dictated note is not kept: it is deleted from the server as soon as the transcription finishes, also if it fails. The data of the people present and of whoever sent or cancelled each report is kept for 24 months; after that, the list of people is replaced by their count and who sent or cancelled the report is erased. The production movements the report generates are part of the production ledger, immutable by design. People's rights are handled through the client company, as employer (see section 9).
- Requests from interested people (custom plugin) are kept in our copy while they exist in Notion. When a contact is erased from Integrations, or when the retention the customer configures archives them, they are deleted from the copy immediately; if someone archives them by hand in Notion, the copy stops keeping their name, email and phone as soon as it is detected (within 24 hours at most) and the row is deleted within about 40 days at most. If the customer stops using the plugin, we delete the copy on request.
8. Security
Each customer company is isolated from the rest: its credentials and content are never shared or mixed with those of any other customer. Connection credentials are stored encrypted, access to the platform always uses a secure connection, and internal access is limited to the staff strictly necessary to provide support.
9. Your rights
You can exercise your rights of access, rectification, erasure, objection, restriction and portability by writing to eduardo@befocusy.com. If the data was provided by the company you work for, address them as the controller; we will assist them. You may also lodge a complaint with the Spanish Data Protection Agency (aepd.es). If you work for a client company based in Poland, you may also lodge a complaint with the Polish data protection authority, the Urząd Ochrony Danych Osobowych (UODO, uodo.gov.pl).
10. Changes to this policy
If we change this policy, we will publish the updated version here and change the date above. If the change significantly affects how data is processed, we will also notify customers by email.