Genki Privacy Policy

Last updated: 17 October 2026 Version: 1.6

This Policy explains what personal data Genki processes, for what purpose, on what legal basis, who it is shared with, how long it is kept and what rights you have. It has been written to be read in full: if anything is unclear, section 16 gives you an address to write to.

Genki is a conversational assistant that connects to your own work accounts — Google, Microsoft and a business's WhatsApp Business account — and operates on them. That means most of the data Genki processes is not something you type into a form: it is your email, your calendar, your files and your contacts, which Genki reaches because you authorised it to and only with the permissions you granted. This Policy is mostly about that.

---

1. Who the data controller is

In this Policy, "Genki", "we" or "the controller" refer to that entity. "You" or "the user" refers to the person using the service.

---

2. The essentials

This summary does not replace the rest of the document, but it covers what most people want to know:

• Creating an account gives us access to nothing of yours. Connecting Google, Microsoft or WhatsApp is a separate, voluntary step, which you authorise on the provider's own screen and can revoke whenever you want.
• WhatsApp differs from the other two and is explained separately (section 3.2 bis): what gets connected is a business's WhatsApp Business account, and the assistant can send messages from its number. If you have an agent listening on that number, it also receives the messages people write to it, with the content inside the notification itself, and keeps the conversation for 90 days, encrypted, so that the agent knows what that person and the business said before. What Meta does not deliver — the history from before an agent listened on that number — it cannot read.
• The permissions requested by default are still broad, because the assistant needs to be able to act and not merely read: email, files, calendar, contacts and chat, with write access. They are, however, split one by one: the read/write pair for each service is requested separately, rather than the general permission that covers both, except for files and for contacts, where the provider publishes nothing in between. The permission for marking Gmail messages, which grants considerably more than is used, is not requested by default: only if you choose it. You can reduce them before authorising and afterwards too, from the providers screen, as many times as you like. Section 3.2 lists them one by one.
• We do not train artificial intelligence models with your data, neither our own nor anyone else's, and our model provider undertakes the same commitment contractually.
• Processing takes place in the European Union, both hosting and processing by the language models.
• We do not sell your data, we do not advertise and we do not build profiles.
• We do not read your email or your documents. Access by our staff to a user's content is prohibited except in four exhaustively listed cases detailed in section 11.
• The files the assistant works with are kept only as long as needed. When it reads, copies, fills in or sends a file — one you attach, or one it brings in from Gmail, Outlook, WhatsApp, Drive or OneDrive because you asked it to — we keep the original, encrypted, for at most 7 days, and its text for as long as it stays attached. It only leaves Genki on your instruction, in the chat or in an agent's configuration (section 3.5).
• If you disconnect a provider from Genki, or reduce its permissions here, we delete what was being synchronised from there, and only that: your collections still exist with what you uploaded by hand, with what you asked to have copied into them and with whatever comes from the other provider. Withdrawing access only from Google or Microsoft does not delete it, because they do not notify us (section 12). If you close your account, we delete all your content.
• If your account belongs to an organisation, the organisation answers for what is done inside it, and what you share in it — including what you synchronise into a shared collection — stays with the organisation even if you disconnect or leave (section 3.6).
• By default, the assistant asks you to confirm before every action with consequences, and if nobody answers, the action is cancelled. You can choose a more automatic mode per conversation. In bots and scheduled tasks no confirmation is possible, because you are not there: read section 8 before giving them instructions that delete things.
• Genki is an artificial intelligence system and it can get things wrong. Section 8 explains what actions it can perform and what control you have over them.

---

3. What data we process and why

3.1 Data you provide to us

3.2 Data we access in your Google and Microsoft accounts

Signing up for Genki does not mean giving access to your accounts. You can create an account and use the assistant without connecting any provider. Connecting Google or Microsoft is a separate, voluntary act, authorised on the provider's own consent screen, and revocable at any time.

We explain clearly what will be asked of you, because it is broad and we would rather you knew before authorising it: so that the assistant can act on your behalf — and not merely read — the configuration proposed by default requests read and write permissions over the services the product uses:

In addition, only if you choose them, Genki may request these other Google permissions. None of them is part of the default configuration:

Genki cannot request any permission that is not in these two tables: the server refuses the authorisation if asked for another.

The narrowest permission that does the job is the one requested, and in five places that narrowest one grants more than we use. Where the provider publishes read and write separately, Genki asks for both separately rather than the general permission that covers them: it does not request https://mail.google.com/, nor calendar, nor chat.messages, nor Chat.ReadWrite, because those grant capabilities no feature of the product exercises.

There are five exceptions, and we would rather explain them than list them:

• Drive files (drive). Between "only what this application created" (drive.file) and "everything" (drive), Google publishes nothing. And drive.file does not work here: the assistant acts on files that were already in your Drive, not on files it created, so with the narrow permission renaming, moving, deleting or sharing any of them fails. It is the broadest permission we ask you for and the one worth weighing before you grant it. You can leave it out: the assistant will keep working without touching your files.
• Microsoft files (Files.ReadWrite.All). The same, for the same reason: Files.ReadWrite reaches only your own OneDrive, so a folder shared with you or a SharePoint library would be invisible to the sync.
• Marking email (gmail.modify). Marking a message as read or starring it takes a single Gmail call, and Google publishes no permission below the one that authorises it: gmail.labels administers the definition of labels, not applying them to a message. So the permission that allows marking also allows reading, composing and sending any email, archiving it and moving it to trash: that is why Google's screen presents it as "read, compose and send". Genki uses it for none of that — it reads with gmail.readonly and sends with gmail.send —, only for marking. It is not Gmail's full permission — https://mail.google.com/, which adds IMAP and SMTP access and permanent deletion, is not requested — but it does grant considerably more than the product uses, which is why it belongs here and is not requested by default: only if you choose it. Without it, the assistant keeps reading and sending, and the only thing it cannot do is mark your email.
• Outlook mail (Mail.ReadWrite). Microsoft Graph does not publish mail writing separately: Mail.Read allows reading and nothing else, and everything the assistant writes in the mailbox — marking a message read, changing its importance and preparing the draft of an outgoing message or of a reply — lives in Mail.ReadWrite. It is the narrowest one that does the job, and it grants at the same time modifying and deleting any message in the mailbox, which no feature of the product does. You can leave Outlook mail out and the assistant will keep working with the rest of the account; what does not exist is a read-only version.
• Contacts, where neither Google nor Microsoft publishes a narrower write permission.

You can reduce those permissions, before and after. At the moment of authorising, Genki lets you choose permission by permission: you can grant, for example, only reading of email, or leave out Drive, or exclude contacts. The assistant will work with fewer capabilities, and will tell you so when you ask it for something it cannot do. We recommend granting only the permissions you need for your use case.

And you can change them later without disconnecting anything. From the providers screen you can see which permissions each account has granted and change the selection as many times as you like. It is worth knowing how this works underneath, because it has a consequence for your data: reducing a permission forces the whole authorisation to be redone, and Google and Microsoft do not do it the same way.

• With Google, Genki cancels the current grant and takes you back to Google's screen with the new selection. Until you complete that second authorisation, the link is left without a credential and the assistant cannot access anything in that account.
• With Microsoft, Genki cannot cancel the current grant (section 12): it takes you to Microsoft's screen with the new selection, and whatever you authorise there replaces in Genki what it had. If you leave that screen, the connection stays as it was. The earlier grant keeps appearing in your Microsoft account until you remove it yourself (section 7).

With Google, if you leave that screen without finishing, it does not stay half-done indefinitely: an automated process periodically sweeps links left without a credential and deletes the content that had been synchronised under the permission that is no longer in force, with the same scope described in section 12. Before applying the change, the application warns you what will be deleted if the new set of permissions leaves out reading of Drive or OneDrive.

Genki can only access what you have expressly authorised, and never services other than those listed.

What each access is used for: carrying out the tasks you ask the assistant to do — searching and drafting email, checking and creating events, working on your files, managing contacts —, indexing the documents you choose, opening the attachments and files you point it to so as to save them somewhere else, fill them in or send them (section 3.5), and running the automations and bots you configure. Those file features use the same permissions as the table and add none. No permission is used for any purpose other than providing you with the service (section 7).

Genki also receives change notifications (webhooks) from Google and Microsoft — new mail, modified event — when you have active automations, for the sole purpose of running them.

We also store the access credentials (tokens) resulting from that authorisation, encrypted. They are the key that lets Genki act on your behalf and we treat them as the most sensitive data we hold.

3.2 bis Data from the WhatsApp Business account, if you connect one

WhatsApp sits apart from the other two for two reasons: what gets connected is not a personal account of yours but a business's WhatsApp Business account, and the permission is granted by that business through Meta. Like the other two, it is a separate and voluntary step, and the assistant works without it.

What the assistant does with that access. Send text messages and files from the business's number, react to a message, mark it as read and read the approved templates. The files it can send include those it holds in Genki, including ones it brings in from your Drive, your OneDrive or your email (section 3.5): the file is uploaded to Meta together with the message, and is never sent as a public link. In doing so it processes the recipient's phone number and the content of what you ask it to send.

The messages people write to the business's number

This is where WhatsApp behaves differently from Google and Microsoft, and it is worth reading in full.

We do receive inbound messages, and only while you have an active agent with the WhatsApp trigger. Without such an agent nothing is processed: the notification reaches our server, we check that the number has no agent listening, and it is discarded without its content being looked at.

When there is one, the notification Meta delivers carries the message inside it, and that is the difference from the other two providers: with Google and Microsoft the notification says "something changed" and the assistant goes and fetches it; WhatsApp publishes nothing to fetch, so whatever does not travel in the notification does not exist. From each inbound message we receive:

• the text of the message — or the file type, if it is not text;
• the phone number of the person writing and the name on their WhatsApp profile;
• the message identifier and, if they were replying to an earlier message from the business, that one's identifier;
• the date and time.

With that, the agent carries out the instructions you wrote for it, which normally means replying. It does so on its own: it cannot stop to ask you. If it is missing something to carry them out, it can ask the person who wrote in, or end without doing it and note so in the agent's run history (section 3.3).

Documents and photos that arrive with a message are downloaded only if the agent needs them. The notification does not carry the file, only an identifier, and Genki only asks Meta for it when the agent is about to open it to carry out its instructions. And only documents — PDF, Word, Excel and text — and PNG, JPEG or WebP images: audio, video and stickers are never downloaded. From a photo, the text it contains is read by optical character recognition, on our own servers: the image is not sent to the language model (section 3.5). Whatever is downloaded is kept like any other file the assistant works with (section 3.5), tied to that run of the agent.

What is left of the message: the conversation, for 90 days. Every message that arrives at a number with an agent listening, and every message the assistant sends from that number — from an agent, from a scheduled task or from a chat — is kept in our database as part of the conversation with that person: the text, or the kind of file and its name when it is not text, the person's phone number and profile name, the date and the message identifiers. The text, the number, the name and the file name are stored encrypted, with the same encryption as the credentials, and of the number a cryptographic digest is kept as well, which is the only thing used to look the conversation up. It is what lets the agent, when it answers, know what that person and the business said before: the notification that wakes it carries the latest messages of that conversation and, if it needs more, it can ask for the earlier ones. Each message is deleted on its own 90 days later, by an automatic process that runs every hour; sooner if you unlink WhatsApp or close your account. Two further things survive the run, and the assistant writes both of them — not the person who wrote in: what the agent answered and the trace of what it did — which tools it used, whether they worked, how long they took, a short narration of each step and the names of what it consulted (section 3.3). The trace does not keep the arguments the tools were called with, but what the agent answered and what it narrated can mention what the person wrote and what it read of the conversation. Both disappear when you delete the agent, and with the account (section 12). It is the same rule as section 3.4, point 6, which governs email and documents. The documents and photos the agent has opened, as described in the previous paragraph, are kept for the periods in section 12 and are deleted sooner if you delete the agent.

On what legal basis. Towards you, performance of the contract (art. 6.1.b GDPR). Towards the person writing, who is not a Genki user, legitimate interest (art. 6.1.f): they started the conversation themselves by writing to a business's public number, and that the business should read and answer it is precisely what they expect in doing so. Their message is processed for that and nothing else. Section 4 explains what they can do if they would rather it were not.

What Genki still cannot read — not a promise, a technical fact. Meta does not deliver the history of a WhatsApp conversation: its interface only pushes messages as they arrive and does not allow them to be read back. What Genki knows of a conversation is exactly what it received and sent while an agent was listening on that number, and for the 90 days of the previous paragraph. Anything from before WhatsApp was connected, anything that arrived with no agent listening and anything the business wrote from another tool is not reachable from Genki.

What we store. The WhatsApp Business account identifier, the identifiers and phone numbers the business authorises, and the resulting access credential, encrypted, handled exactly like those of Google and Microsoft. Of the messages, the conversation of the paragraphs above, encrypted and for 90 days; of the documents and photos the agent opens, what section 3.5 describes.

What passes through Meta. The messages and documents you send are delivered by Meta, so they necessarily pass through its infrastructure. That processing is governed by the terms the business has with Meta, which we do not control. We do not use WhatsApp data for advertising, we do not sell it and we do not use it to train models, exactly as with everything else (sections 3.4 and 7).

What the business answers for, and we do not. WhatsApp only allows messaging people who have consented to hear from that business, and more than 24 hours after that person's last message it only allows previously approved templates. These are Meta's rules and the business connecting the account is the one that answers for them: do not ask the assistant to write to someone who has not authorised it. Sending a document is sending a message, under the same rules. And since inbound messages are data about that business's own customers, it is the business that has to tell them: that the number is handled by an automated assistant is information for the business to give, not us.

How to withdraw it. Unlinking WhatsApp from the application deletes the credential from our systems and the kept messages of that account's conversations, the assistant can no longer send anything and inbound messages stop being received. Deleting the agent that listens on that number, or removing its WhatsApp trigger, has the same effect on reception without unlinking the account; what was already kept stays until its 90 days are up. Meta, like Microsoft, publishes no mechanism that would let Genki cancel the granted permission itself, so the access will keep appearing in the business's settings until you remove it, at business.facebook.com/settings/whatsapp-business-accounts. We remind you of this in the application itself when you unlink.

---

3.3 Data generated by using the service

3.4 What we do not do

This section matters as much as the previous ones, and it is written so that you can hold us to it. Each commitment comes with what it means exactly and, where an exception exists, with the exception stated.

1. We do not train artificial intelligence models with your data. Neither our own models — we have none — nor anyone else's. Your conversations, your emails, your files and the vectors derived from them are not used to train, fine-tune or evaluate any model, at any time, not even in aggregated or pseudonymised form. Our model provider undertakes the same commitment contractually (section 6), and Google's Limited Use policy imposes it on us as an express prohibition regarding Gmail and Drive data (section 7). It is a structural commitment, not a revisable internal policy: if we ever wanted to do it, we would need your explicit prior consent, and we would ask for it as such.

2. Your data does not leave the European Union. Hosting is in European Union regions and processing by the language models takes place in European Union data zones, including the associated storage. We do not transfer your data to third countries. Four qualifications, because we would rather you knew them: content that our provider's anti-abuse system flags may be reviewed by authorised staff located in the European Economic Area, with storage in the European Union (section 6); the data you already held in Google or Microsoft is governed by the terms you contracted with them, which we do not control; whatever you ask to send to another person, by email or WhatsApp, goes wherever that person is (section 3.5); and whatever the assistant passes to an external tool (MCP) you have connected reaches the servers of whoever publishes it, wherever they are (section 9).

3. When you delete something, it is really deleted. Deleting a conversation, a file or a collection removes the records; in the case of collections, also the extracted text and its vector representations from the search database, and in the case of files, also their original from the store (section 3.5). They are not marked as deleted: they are deleted. On account closure all your content is removed — in an organisation, except what you shared in it (section 3.6) —, your own user record is deleted — no tombstone or empty row with your identifier remains — and the Google and Microsoft credentials are withdrawn (section 12).

What survives an account closure is the following, and there is nothing else:

• Your invoices, if there ever were any. It is the only thing that survives carrying personal data of yours, and it is worth knowing exactly which: the law requires an invoice to identify its recipient, so inside it remain your first name and surname, your email address and the internal identifier your account had, together with the period billed and the amounts. This is why an account that has been invoiced is never entirely anonymous, and we prefer to tell you rather than let you assume otherwise. They are kept blocked: available to the authorities and the courts if required, and not used by us in the ordinary course, for the period given in section 12.
• Your token consumption, but no longer as yours. These are counters — how many were spent, with which model and on what date — which contain no content from your conversations or your files. They are kept because they are the arithmetic backing of what was invoiced, but when the closure is executed they are reassigned in bulk to an internal account common to all closures: they stop being associated with you and are mixed with those of everyone else who has closed their account, so that it is no longer possible to isolate which were yours. With respect to those figures, and once that association is broken, we do not hold personal data.
• The service's technical and security logs, which contain your email address and your IP address associated with access attempts. We do not delete them on account closure because their purpose — detecting fraud, abuse and unauthorised access — outlives it: an attack against your account does not stop mattering because the account is closed. They are deleted automatically once the period in section 12 expires, without you having to ask.
• Backups: each backup of the database and of the cache is kept for 7 days, and each backup of the collections' vector store for 35 days. During that period the data is blocked and not used, and it disappears when the backup expires.

We point out a practical consequence of the above, because it works in your favour and is not obvious: if you have never been issued an invoice, all that remains of you are those security logs, and they expire on their own. Once their period has passed we hold no personal data of yours. Where an invoice does exist, retention is six years and by legal obligation.

4. We do not build profiles about you. We do not analyse your behaviour to classify you, score you, predict anything about you or segment you. Nor do we take automated decisions producing legal effects or similarly significantly affecting you, within the meaning of Article 22 GDPR.

5. We do not advertise and we do not sell data. We do not use your data for advertising or marketing purposes, we do not transfer or sell it to third parties for that purpose, we do not enrich it with external sources and we do not include analytics, tracking or advertising pixels in the application (section 14).

6. We do not read your email or your documents, and we do not keep them except for what you ask. These are two separate commitments and it is worth telling them apart, because they protect against different things.

Nobody reads them. No one on our team accesses a user's content, except in the four exhaustively listed cases in section 11, and every access is logged.

And they are not stored. When the assistant needs an email, a file or a message in order to answer you, it asks your provider for it at that moment, uses it within that turn and lets it go when the turn ends. The body of that email or that document is not copied into our database. If it is needed again in a later turn, it is fetched again; you can check this yourself in the trace attached to each answer, where the same call shows up once more. The one exception is WhatsApp conversations, because Meta does not deliver them to be fetched again: they are kept encrypted for 90 days, as section 3.2 bis explains.

What a conversation keeps is what you write, what the assistant answers and a bounded index of what has already been consulted, which the assistant itself writes as the turn closes so it does not have to ask for everything again: which resources have been touched, where they live, and the specific details that make each one distinguishable — an email's subject and sender, a document's figures or dates. It is an index and never the body: it is capped by design at around eighty words per resource and is forbidden from dumping full text. It is deleted with the conversation. In automations not even that exists: each agent run stands alone and carries nothing over from the last. Conversations and runs also keep the trace of what the assistant did, which stores the names of what was consulted — an email's subject, a file's name — but not its content (section 3.3).

There are two exceptions, and you decide both:

• The documents you choose to index into a collection: those are stored, with their extracted text and vectors, because storing them is exactly what they are for. Section 3.3 explains it, and they are deleted when you delete the collection.
• The files the assistant works with because you asked it to. When consulting is not enough and it has to open an attachment, save a file somewhere else, fill it in or send it, it brings it into Genki as a file: the original is kept encrypted for at most 7 days, and the text for as long as it stays attached. Section 3.5 explains it.

7. We do not use your data for anything other than providing the service. Not for development, not for testing, not for sales demonstrations, not for internal analysis of content. Development and test environments do not contain real user data.

3.5 Files the assistant works with

Besides answering about your documents, the assistant can work with files: open them, save them somewhere else, prepare a filled-in copy and send them to other people. It does so only on your instruction: in the chat, when you ask it to; in an agent or a scheduled task, when you have written it into its instructions and given it access to that connection.

Where they come from. From what you attach to a chat and from your connected accounts: attachments to Gmail and Outlook emails, documents and photos sent to you over WhatsApp (section 3.2 bis) and files in Google Drive and OneDrive. Documents — PDF, Word, Excel and text — and PNG, JPEG and WebP images are accepted, and what each one is is checked from its content, not its name: programs, compressed archives and documents with macros or protected by a password are not accepted.

Images are read as text. The text an image contains is extracted by optical character recognition (OCR), and that is done on our own servers, without sending it to any third party. What reaches the language model is that text, never the image. Otherwise, an image is kept and deleted like any other file in this section. Images inside the body of an email can be read without bringing them into Genki: they are read at that moment and nothing of them is kept.

What we keep of each one.

• The original, encrypted, in a private Amazon Web Services store in the European Union (Ireland), for at most 7 days from when it comes in. It is what allows it to be copied or sent as it is, without being rebuilt. After that period it is deleted automatically; sooner, if you remove the attachment, if you delete the chat, the agent or the task that brought it in, or if you close your account.
• Its extracted text, encrypted in our database, for as long as the file stays attached to a chat or to a run of an agent or a task. It is what the assistant reads. After the 7 days it can still read it, but no longer copy or send it: for that it would have to bring it in again.
• What is needed to manage it: its name, type and size, where it came from, a fingerprint of its content that avoids storing it twice, and the result of the malware scan.

What it can do with them.

• Save them to your Library, in one of your collections. From then on it is one more document in the collection (section 3.3) and stays until you delete it.
• Upload them to your Google Drive or your OneDrive, always as a new file and without sharing it with anyone. Also from one cloud to the other: a file from your Drive can end up in your OneDrive, and vice versa. Whatever is uploaded to your account is governed by that provider's terms.
• Prepare a filled-in copy: the fields of a PDF form, those of a Word document or the cells of an Excel workbook. The original is not touched: the result is a new file, with the same periods as any other. Only from the chat, and with your confirmation, can that copy be saved as a new version of the Drive or OneDrive file it came from, and the provider keeps the previous version. An agent cannot do this.
• Send them as an attachment, through Gmail, Outlook or WhatsApp.

Sending files to other people. It is the only one of these operations that takes the file out of your accounts, which is why it has its own controls. In the chat, sending goes through the control in section 8: in the default mode, the card shows you the file and the recipients, and nothing is sent until you confirm. In agents and scheduled tasks, where there is nobody to ask, files are only sent to the recipients you have authorised in advance in the agent's configuration, plus you and whoever wrote the message the agent is handling. Over WhatsApp, the file is uploaded to Meta together with the message; a link is never sent. Once sent, the file is in the hands of whoever receives it, and the provider that delivers it — Google, Microsoft or Meta — handles it under its own terms; what the recipient does with it is beyond our reach.

How they are protected. Every file goes through a malware scan before it can be read, copied, edited or sent, and one that fails it is not used and is deleted. The scan is done by an AWS service in the same region, so the file does not go out to any other provider to be scanned. There are no links to these files, neither public nor temporary: the assistant never handles the file or an address to download it from, only an identifier, and on every operation the server checks that you have access to that file. The full measures are in section 11.

What does not change. These features use the permissions you have already granted (sections 3.2 and 3.2 bis) and request no new ones: if you leave a permission out, the assistant cannot bring files in from that service or take them there. And the same commitments in section 3.4 apply: these files are not used to train models or for advertising, and nobody on our team reads them outside the four cases in section 11.

3.6 Organisation accounts

Genki can also be contracted as an organisation: a team, a school or a company that pays a seat for each person and shares agents, tasks, collections and other tools with them. If your account belongs to an organisation, this section concerns you, and where it says something different it prevails over the rest of this Policy.

An account is either personal or an organisation's, never both. An organisation's account is born when its invitation is accepted, and the invitation has to reach the same email address you sign in with. If you already had a Genki account, it is emptied on joining: its content, its subscription and its billing details are deleted, and your user, your profile, your password and your earlier invoices are kept. Whoever creates an organisation takes what they had into it, as their own private items.

Who answers for what.

• For what is processed inside the organisation — what its members create and share in it, what goes through its central account and its members' data as such — the controller is the organisation, which decides what it is used for and who sees it. Genki acts as processor, on the organisation's behalf and following its instructions, and will provide the data processing agreement under Article 28 GDPR at its request (section 19 of the Terms).
• Genki remains the controller of what it processes for its own purposes: identifying each person on sign-in, charging and invoicing the organisation, the security of the service and its logs, and meeting its legal obligations. The rest of this Policy applies to that processing as it stands.
• If you are a member, whatever you want to know or ask about what the organisation does with your data is a question for the organisation; about what Genki does, you can write to us at the address in section 16.

Who sees what. An organisation has three roles: the owner, who pays for it and can transfer or dissolve it; the admins, who manage the members, the central account and what is shared; and the members.

• Each person's private items are seen by nobody else, not even the owner or the admins: their conversations, which are never shared, and the agents, tasks, collections and anything else they created without sharing.
• Shared items are seen, used or modified by whoever their visibility says, which ranges from the owner alone to every member and distinguishes who may use them from who may modify them. Only the owner and the admins can share something and choose with whom.
• About the members, the owner and the admins see their first name and surname, their email, their role, their seat, when they joined and who invited them, and how much the organisation's shared items have spent. What each person spends on their own items they do not see.
• Billing is seen only by the owner. The organisation's invoices carry its name, its tax details and the owner's email.

The central account. The owner and the admins can connect Google, Microsoft and WhatsApp Business accounts on behalf of the organisation. Shared agents and tasks work with them — not with the accounts of whoever created them — and so do the owner and the admins in their own conversations. They keep working after whoever connected them leaves. Sections 3.2 and 3.2 bis apply in the same way, with one difference: the permissions are granted by the organisation, and the organisation answers for what is done with them. Each member's personal connections are seen by nobody else and are not used by shared agents.

What stays when someone disconnects or leaves. This is where an organisation departs from section 12, and it is worth reading carefully:

• What a member synchronises into a shared collection with their own Google or Microsoft connection stays in the collection even if they disconnect that connection, leave the organisation or have their account closed: it stops being updated, but its documents, its text and its vectors remain, within reach of whoever can see the collection, until the owner or an admin removes them or deletes the collection. What they synchronised into their private collections is deleted on disconnecting, as section 12 says.
• What was synchronised with the central account is not deleted when it is disconnected: it stays in the organisation's collections until someone deletes it or the organisation is dissolved.
• When someone leaves the organisation, or is removed from it, that person's account is closed at once — they can no longer sign in — and what was theirs is deleted at the end of the period the organisation has paid for; until then, the owner or an admin can undo the departure. Everything private of that person is deleted; what they shared stays with the organisation, and the shared agents and tasks that ran as them start running as the owner.
• When the organisation is dissolved, which only the owner can do, it carries on until the end of the paid period. On that day everything that belongs to it is deleted — what is shared, its collections and the central account, whose credentials are withdrawn — and its members' accounts are closed with all their content. The owner chooses whether to close theirs too or keep it as a personal account, with what they had created for themselves.

---

4. Data about people who are not Genki users

It is the inevitable consequence of an assistant that works on your email and your calendar: in doing so, Genki also processes data about senders and recipients of your emails, guests at your meetings, authors of documents and your contacts — people who have no contract with us. If you connect WhatsApp, this adds the phone numbers of the people the business writes to and the content of those messages and, if you have an agent listening on that number, also the people who write to the business, the name on their WhatsApp profile and what they write (section 3.2 bis). And if you ask the assistant to send files, the address or number of whoever receives them, as well as any third-party data those files contain (section 3.5).

• That data is processed solely to provide you with the service you asked for, and with the same protection as your own.
• It is not used for any purpose of Genki's own, nor to train models, nor for advertising, nor is it enriched with external sources.
• It is kept and deleted on the same basis as the rest of your account's content (section 12).
• Any of those people can exercise their rights by writing to the address in section 16. We tell you transparently about the practical difficulty: if their data is inside a Genki user's mailbox or files, the person who controls that content is that user, and our ability to locate or delete it separately is limited. We will handle the request as far as it is materially possible for us and will guide them on how to approach the controller of that data.

If you are a user: by connecting your accounts you are making third-party data available to Genki. You must be entitled to do so and to ask the assistant whatever you ask it about them.

---

5. Special categories of data

Genki never asks you for sensitive data. But your email, your calendar and your documents routinely contain information about health (a medical appointment), trade union membership, religious beliefs or political opinions, and it is foreseeable that Genki will process them when carrying out what you ask.

Genki does not tell that information apart from the rest, and that is deliberate. It does not identify which emails discuss health, nor classify anyone by their beliefs: it treats all content alike, carries out what you ask and draws no conclusion about you from it. There is no analysis separating the sensitive from everything else, because performing one would be a different and more invasive processing than the one you asked for.

And you decide whether this happens at all. Creating an account and using the assistant requires connecting no provider: the connection is a separate, voluntary step, you can grant the permissions one by one, reduce them later and revoke them whenever you want. Disconnecting the provider from Genki, or reducing its permissions here, deletes whatever had been synchronised from there, with the scope described in section 12. Authorising access knowing what your account contains is precisely what makes this processing lawful.

Practical recommendation: do not enter sensitive data into the chat that is not necessary for what you are asking, and grant Genki the most limited permission that works for you. If your work involves specially protected information, consider restricting read permissions to what is strictly necessary.

---

6. Artificial intelligence: how your data is used

This section tends to raise the most questions, so it goes into detail.

We do not train models with your data. Genki has no models of its own and does no fine-tuning with user information. We use third-party models through calls to their API: the text is sent, the response is received, and the processing ends there.

Where processing happens. All processing by language models takes place in Microsoft Foundry (Azure), in European Union data zone deployments. Both processing and the associated storage remain within the European Union.

Which models are used. Genki uses general-purpose language models for conversations and an embeddings model for document indexing, all deployed under the conditions described in this section. You can see in the application which model each conversation is using and choose among those available. Adding or withdrawing particular models does not alter any of the guarantees in this Policy: they all operate under the same provider, the same European Union data zone and the same contractual commitments, and none will be enabled that does not meet those conditions.

What is sent to the model. Your message, the conversation's context and, where relevant, fragments of your indexed documents, the text of the files it is working with (section 3.5) and the results from connected tools. Of an image, the text read from it is sent, never the image. The text of the documents you index is also sent to the embeddings model so they can be searched by meaning.

What our provider commits to. Under Microsoft's data policy applicable to models sold by Azure: prompts, responses and embeddings are not used to train any generative artificial intelligence model without the customer's permission or instruction, neither by Microsoft, nor by OpenAI, nor by third parties; they are not available to OpenAI; the service runs entirely within the Azure environment; and the models do not store prompts or responses. Microsoft acts as a processor under its Data Protection Addendum.

An exception we tell you about because it is true. Microsoft's service includes an abuse detection system. When that system raises a flag, the content may be subject to review, which is automated by default and involves no storage. In cases where automated review is not enough, human review may take place, and then the flagged content is stored in a monitoring repository segregated per customer, located in the European Union, accessible only to authorised Microsoft employees located in the European Economic Area through secure workstations. It is an exceptional case and outside our control, but we would rather you knew than claim without qualification that your content is never retained outside Genki.

Commitment on providers. We will not enable any artificial intelligence model provider other than those described without first reviewing its processing agreement, the location of processing and its data policy, and without updating this Policy.

---

7. Use of Google API data

Genki's use and transfer to any other app of information received from Google APIs will adhere to the Google API Services User Data Policy, including the Limited Use requirements. In particular:

• Data obtained from Gmail, Drive and other Google services is used exclusively to provide you with the assistant's user-facing features: answering you, doing what you ask in the chat and what you have configured in your agents.

• It only leaves Genki when you ask for it or set it up, and only to do what you asked, in these ways:

  • Copying it to another service of yours: your Drive, your OneDrive or your Genki Library.
  • Sending it by email or WhatsApp, as an attachment or inside the message. In the chat, to whoever you name. An agent or a scheduled task sends emails and messages to whoever the instructions you wrote for it say, without asking for confirmation; the list of authorised recipients in its configuration limits only the sending of files (sections 3.5 and 8).
  • Passing it to an external tool (MCP) you have connected, when what you ask — or what an agent you gave access to that tool does — requires it: for example, an email's subject and text to create a task in your project manager. Only to the tools you connected and authorised, and only what that action requires; what that third party then does is governed by its own terms (sections 8 and 9).

  Beyond that, it is not transferred to third parties, except to the processors we need to provide you with the service (section 9), for security purposes — for example, investigating abuse —, where required by law, or as part of a merger, acquisition or sale of assets and only with your prior, express consent.

• It is not sold, nor transferred to advertising platforms, data brokers or information resellers.

• It is not used for advertising purposes of any kind, including personalised, interest-based or retargeting advertising.

• It is not used to determine creditworthiness or for lending purposes.

• No human reads that data, unless you expressly consent for specific messages or files, it is necessary for security purposes — for example, investigating abuse —, a legal obligation requires it, or the data is aggregated and anonymised for internal operations.

• It is not used to develop, improve or train generalised artificial intelligence or machine learning models, nor any other model: we do not train models on your data (section 3.4).

• It is kept only as long as needed. An attachment or a file the assistant brings in to work with is kept for the periods in section 12 — the original for at most 7 days, the text for as long as it stays attached — and is deleted sooner if you ask.

Processing of data from Microsoft is governed by the same principles.

Meta Platform data. Where a WhatsApp Business account is connected, Genki processes Meta Platform data — the account identifier, the authorised numbers, the messages you ask it to send, the conversations it keeps for 90 days (section 3.2 bis) and the documents and photos an agent downloads — solely to provide you with the service, in accordance with the Meta Platform Terms and the WhatsApp Business Messaging Policy. It is not used for advertising, not sold or transferred, not used to train models, not used for profiling, and not combined with data obtained from other sources. A document or a photo received over WhatsApp only leaves Genki if you ask to save it to your Drive or your OneDrive, or to send it to someone.

You can revoke Genki's access to your account at any time from the application itself and, in addition, from your account's security settings: for Google, myaccount.google.com/permissions; for Microsoft, account.live.com/consent/Manage for a personal account or myapps.microsoft.com for a work or school account; and for WhatsApp, business.facebook.com/settings/whatsapp-business-accounts. For Microsoft and for WhatsApp that step is not redundant, for the reason explained in section 12. Withdrawing it only there, on the other hand, does not delete what was synchronised: for that you also have to disconnect the provider in Genki (section 12).

---

8. What the assistant can do and what control you have

Genki does not merely answer: it performs actions in your accounts when you ask it to. It can send email on your behalf, send files as attachments by email or WhatsApp, upload files to your Drive or your OneDrive, create and modify events, and delete files, calendars, events and contacts. Some of those actions are irreversible: a file that has been sent cannot be recalled.

Genki is an artificial intelligence system and it can make mistakes: misreading an instruction, choosing the wrong recipient or proposing an action you did not intend. We warn you of this expressly, as required by Article 50 of Regulation (EU) 2024/1689 on Artificial Intelligence.

That is why the system includes a control that holds the action back and asks you before executing it. In the chat you have three modes, chosen per conversation: they are remembered for the next time you open it and you can change them mid-conversation, so that the next message already behaves the new way.

In bots and scheduled tasks there is no confirmation card at all, and we say so bluntly because it is what you need to know before creating one: they run without you being there, so there is nobody to ask, and the control described above does not apply to them — nor does the one that protects deletions. What you can limit, and we recommend that you do, is which external tools each bot or task can reach: when creating them you choose which tool servers and which specific functions they may use, and whatever you do not tick is out of their reach. There are two further limits that do not depend on that selection: everything a bot sends — emails, WhatsApp, Teams or Google Chat messages, event invitations and shared files, with or without attachments — can only go to the recipients you have authorised in its configuration, plus you and whoever wrote the message the bot is handling, unless you tick that it may write to anyone; and it never replaces a file in your Drive or your OneDrive (section 3.5). Beyond that, the tool selection is today the only effective limit on what a bot can do unsupervised, so do not give a bot instructions that involve deleting anything you are not prepared to lose.

What a bot or a task can do is tell you things and ask you. It leaves you a notice, which you see in the application — at the top of the screen if you have it open, and in the Notices block of the home screen and on its own page — and which stays there even if you later delete the bot or the task. If its instructions tell it to confirm something before doing it, the notice is a question: it waits for your answer for a limited time — ten minutes unless the bot says otherwise, and never more than thirty — and if you do not answer, or answer no, it does not do it. This only happens when its instructions ask for it: it is not the chat's confirmation card and does not turn an unsupervised run into a supervised one. Genki also leaves you a notice when one of your Google or Microsoft connections stops working and has to be synced again.

And it is worth knowing exactly where our responsibility for those tools ends, because it is not a minor point. External tool servers — the so-called MCP servers — are chosen, connected and authorised by you. They are not ours, we do not operate them, we do not audit them and we have no way of knowing what that third party does with whatever reaches it.

What is our responsibility, and what Genki answers for:

• Holding encrypted the credential you entrust to us for that tool.
• Checking the address you type: that it points outside our infrastructure and that the connection is encrypted, before accepting it and again on connecting.
• Sending that tool nothing other than what the action you asked for requires.
• Warning you before connecting it, on the screen itself.

What happens afterwards, already on that tool's server — what it stores, for how long, who it shares it with, under what security measures and under what terms of use — is outside our control and outside our responsibility, and at that point your relationship is directly with whoever publishes that tool. Read their terms before connecting it, with the same care you have read these: connecting an MCP server is opening a door to what the assistant can reach on your behalf.

Every run of a bot or a scheduled task is logged — time, status and an extract of the result — and you can consult it from the application, filtering by date, which is how you know what the assistant did while you were away. Actions executed from the chat are also reflected in the conversation itself.

We ask you to bear in mind the allocation of responsibility: we are responsible for the system being properly designed, disclosed and controlled; you are responsible for the instructions you give the assistant and the permissions you grant it. The details of that allocation are in the service's Terms and Conditions.

---

9. Who we share your data with

We do not transfer your data to third parties. We do use providers who process it on our behalf and following our instructions, under a processing agreement pursuant to Article 28 GDPR:

Stripe processes plan payments. You type your card details on Stripe's payment page, which also runs the strong customer authentication that payment regulations require; Genki neither sees nor stores the card number. Stripe processes the payment data on our behalf to take the payment and, to meet its own legal obligations and prevent fraud, as an independent controller under its own privacy policy. In the European Union the service is provided by Stripe Payments Europe, Limited, based in Ireland.

VIES, the European Commission's system for checking VAT numbers, receives the VAT number of businesses in other European Union countries that ask for a full invoice. VAT rules require that check before a business in another Member State is invoiced without VAT, and only the number is sent.

In addition, your data is necessarily communicated to Google and Microsoft when you connect your own accounts, because they are your providers and that is where your emails and files live, and when you ask the assistant to save a file there or send it by email. That processing is governed by the terms you have contracted with them.

The same is true of Meta if you connect WhatsApp: it is the party that delivers the messages, so the recipient's number, the content of the message and the documents sent pass through its infrastructure. That relationship is the one the business has contracted with Meta (section 3.2 bis).

External tool servers (MCP). If you decide to connect third-party tools to the assistant, you establish that connection, under your own responsibility and with the permission you grant each one. Genki acts as a mere conduit: it transmits what you ask for and does not control, and cannot control, what that third party then does with the information. The exact split between what we answer for and what the tool's publisher answers for is set out in section 8. You are warned of this before connecting a tool, on the screen itself.

Three things are worth stating precisely about those connections:

• We hold your credentials. The API key or token the tool requires is stored encrypted in our systems, with the same treatment we give Google and Microsoft credentials (section 11).
• Removing a connection deletes your credentials and, where the provider allows it, also revokes the authorisation with them. On account closure the same is done with all of them.
• You can decide which tools each bot or task can reach, server by server and function by function (section 8).

The address of an MCP server is written by you, so it is a destination we do not choose. Before accepting it, and again at the moment of connecting, we check that it points outside our infrastructure and that it uses an encrypted connection, so that a mistyped address — or one written with bad intent — cannot become a route into our internal systems.

We will also disclose data where there is a legal obligation, in particular to judges, courts, law enforcement or competent authorities.

---

10. International transfers

The processing carried out by Genki takes place in the European Union. Hosting is in European Union regions and processing by the artificial intelligence models takes place in European Union data zones.

On that basis, we do not carry out international transfers of data as a general matter. To the extent that any of our providers, which belong to corporate groups with a parent outside the European Economic Area, may access data from outside for support or administration tasks, such access relies on the safeguards of Chapter V GDPR: adherence to the EU–US Data Privacy Framework where applicable and standard contractual clauses approved by the European Commission, with any supplementary measures required.

The exception is the email you write to us. The mailboxes we answer you from are hosted by Proton AG in Switzerland, a country the European Commission recognises as providing an adequate level of protection (Article 45 GDPR), so that transfer requires no additional safeguards.

Data you already held in Google or Microsoft is governed by the location terms you yourself have contracted with those providers, and falls outside our control.

---

11. Security and internal access

We apply the technical and organisational measures required by Article 32 GDPR, including:

• Passwords stored by irreversible cryptographic derivation with salt.
• Google and Microsoft access credentials encrypted before being stored.
• Files encrypted at rest: their extracted text in our database, and their original in a private AWS store in Ireland encrypted with keys managed in AWS KMS, with no public access and reachable only by the service that manages it. Sensitive configuration is also encrypted in the database.
• Malware scan of every file before it can be read, copied, edited or sent: one that fails it is not used and is deleted. It is done by an AWS service in the same region.
• Logging of access to the file store.
• The assistant handles no files and no links, only identifiers, and on every operation the server checks that you have access to the file requested (section 3.5).
• Vector database on a separate encrypted volume with keys managed in AWS KMS.
• Platform secrets held in an encrypted secrets manager, with no credentials in the code.
• Encrypted communications over TLS and an isolated internal network, with no administration ports exposed to the Internet.
• Your session credential is not accessible from the page. Your session identifier travels in a cookie that the browser does not allow any JavaScript code to read — not ours, not a compromised dependency's, not one an attack might try to inject into the page — accompanied by a token that prevents another website from making requests on your behalf.
• Verification of destinations we do not choose. When the service has to call an address decided by a third party — that of an external tool you configure, or one an authorisation server gives it — we verify that it points outside our infrastructure before opening the connection and again on opening it.
• Sessions of limited duration and progressive lockout after failed access attempts, with rate limiting.
• Signing out really signs you out. When you leave, your session is revoked on the server, not merely erased from your browser: it stops being accepted across all our services immediately, even if someone kept a copy of the credential. Moreover, a session cannot be extended indefinitely however much it is used: it has an absolute cap after which you must identify yourself again.
• Operations with no way back ask for the password again. Disconnecting a provider, removing an MCP server with its credentials or deleting a bot require that you have identified yourself within the last few minutes, even if your session is still open. An unattended computer with the session open is not enough to destroy any of that.
• Permission control: Genki can only access what you authorise, you can reduce the permissions before granting them (section 3.2) and you can revoke them at any time, both from the application and from your Google or Microsoft account.

Access by our staff to your content. A general prohibition applies to accessing the content of a user's conversations, emails or documents. Access is allowed only in four cases: (i) where you expressly request it as part of a support request and in order to resolve it; (ii) where it is indispensable to investigate a security incident; (iii) where a legal obligation requires it; and (iv) over aggregated or anonymised data. Every access is logged with identity, reason and date, and everyone with access is bound by a duty of confidentiality.

Security breaches. We have an internal incident management procedure. If a security breach occurs that poses a risk to your rights, we will notify the Spanish Data Protection Agency within 72 hours and will inform you without undue delay where the risk is high.

If you find a vulnerability, you can report it through the channel we publish at genki.digital/.well-known/security.txt, following the RFC 9116 standard: it gives the address to write to.

---

12. How long we keep your data

The general principle is simple: we keep your data for as long as you have an account and need it for the service.

Deletion on withdrawing a permission. If you disconnect a provider from Genki, we delete everything that had been synchronised from that source — the folders and files it brought, with the extracted text and its vector representations —, we delete the credentials from our systems — and in Google's case we also revoke them with the provider itself, with the qualification about Microsoft detailed below — and we cancel the notification subscriptions. If instead of disconnecting it you reduce its permissions here and leave out reading of Drive or OneDrive, we delete what was synchronised from there just the same. Synchronised data does not outlive the connection that brought it.

Withdrawing access from the provider itself is not enough to delete it. Google and Microsoft do not notify Genki when you withdraw access from your account (section 7): the connection stops working, but what was synchronised stays in your collections, and the credential — which no longer works — in our systems, until you also disconnect the provider from Genki. Do so if you want it deleted; the step-by-step is in How to disconnect your accounts and delete your data.

The scope is worth stating precisely, because it is more surgical than is usually claimed and it works in your favour: your collections are not deleted. What disappears from them is what was being synchronised from the affected provider; what you uploaded by hand, what you asked the assistant to copy into them and what comes from the other provider stays intact. The application warns you of all this before you confirm the change. If you reconnect the provider, you will need to synchronise again whatever you want available once more.

The files the assistant has already worked with do not depend on the permission either, because they were brought in on a specific instruction of yours (section 3.5): they stay attached where they were, with their periods — the original disappears after 7 days in any case —, and are deleted when you remove the attachment or delete the chat, the agent or the task. Whatever you asked to upload to your Drive or your OneDrive is in your account with that provider, and whatever you sent is with whoever received it.

With Google, the same happens if you start changing permissions and never complete the new authorisation: the link is left without a credential and an automated process sweeps it periodically, deleting that same content. You need do nothing.

Closing your account. You can close it yourself, from Account → Cancel account, without writing to us or waiting for anyone to process it. An organisation account is not closed that way: a member closes it by leaving the organisation, from the same place, and the owner by dissolving it (section 3.6). Because the service is paid monthly, closure is not immediate and that works in your favour: your account keeps working normally until the month you have already paid for ends, and until then you can change your mind and get it back intact — from the same screen, by identifying yourself again with your password or with the provider you sign in with. The application shows you the exact date on which it will be executed before you confirm, and asks you to type your email to confirm it.

When that month expires, and in a single operation: we check that the last period has been invoiced, the notifications Google and Microsoft send us are cancelled, your credentials are withdrawn — those for Google, for Microsoft and for any external tools you had connected; those for Google we also revoke with the provider itself, and Microsoft offers no equivalent way of doing so, with the exact consequences detailed below — and all your content is deleted: profile, conversations, messages, files — their originals in the store too —, document collections and vectors, bots, tasks, their execution history, your notices, your subscription and your settings. Your own user record is also deleted, without leaving an empty row in its place. From then on the closure cannot be undone.

The order of that operation has a consequence we would rather explain: before anything is deleted we check that the last period has been invoiced. Each month's invoice is issued when it is charged, at the start of the period (section 13 of the Terms), so it will normally exist already. If it did not, the closure halts and nothing is deleted — described a few paragraphs below — rather than leaving you without the document evidencing what you have been charged.

Only two things survive, on the terms in section 3. One by legal obligation: your invoices, which carry inside them your name, your email and the identifier your account had. And another by legitimate interest in security: the technical logs holding your email and IP address, which are not deleted with the account and are removed once their own period — the one in the table above — expires. Your token consumption is kept, but detached from you and mixed with that of all other closures, so that it can no longer be isolated. Backups expire on their own — after 7 days for the database and the cache, after 35 for the vector store —, and meanwhile the data is blocked and not used.

A clarification about revoking permissions, because the closure screen summarises and here the detail is warranted. Your Google and Microsoft credentials are deleted from our systems in both cases: from that moment Genki cannot access anything of yours. What differs between one provider and the other is what happens on the provider's side:

• Google publishes a standard revocation mechanism and we use it: the authorisation is cancelled with Google immediately.
• Microsoft publishes no equivalent mechanism we can invoke. We would rather tell you exactly what that means instead of brushing it off with "it expires on its own", because that would not be true: any live access tokens stop working in about an hour, and the renewal credential becomes unusable through disuse within up to 90 days — we do not keep it, so nobody can use it —; but the authorisation you granted back then does not expire by itself and will keep appearing in the list of applications with access in your Microsoft account until you remove it.
• Meta behaves like Microsoft, with one further wrinkle: it publishes no mechanism that would let us cancel the grant either, and the permission belongs to the business rather than to a person. Unlinking deletes the credential from our systems — from that moment Genki can send nothing — and the access keeps appearing in the business's settings until it is removed there, at business.facebook.com/settings/whatsapp-business-accounts.
• That is why we ask you to remove it yourself, and we tell you so in the application when disconnecting the provider and when confirming the closure, with the direct link: personal account at account.live.com/consent/Manage and work or school account at myapps.microsoft.com. It takes a couple of clicks and it is the only way for no record of the permission to remain (section 11).

If that last period did not appear as invoiced, the closure halts and nothing is deleted, so as not to leave you without the document evidencing what you have been charged. It is an anomalous situation that we review manually, and about which we would inform you.

---

13. Your rights

You can exercise the following rights at any time:

• Access: know what data of yours we process.
• Rectification: correct data that is inaccurate.
• Erasure: ask us to delete it. You do not need to ask us: you can exercise it yourself and immediately from Account → Cancel account, with no intermediary and no response time (section 12).
• Restriction of processing in the cases provided by law.
• Objection, in particular to processing based on our legitimate interest.
• Portability: receive your data in a structured, commonly used format.
• Withdraw consent at any time, without affecting the lawfulness of prior processing.

What you can do without writing to us. Besides closing your account, from the application itself you can at any time:

• Sign out, which revokes the session on our servers and not only in your browser.
• Disconnect Google or Microsoft, or change the permissions granted without disconnecting them, with the deletion of associated content described in section 12.
• Remove an external tool (MCP) together with its credentials.
• Delete a conversation, a file, a collection, a bot or a task, or remove an attachment from a chat, which are really deleted (sections 3 and 11).
• Choose how much confirmation you want before the assistant acts, conversation by conversation (section 8).

We would rather those controls were in your hands than dependent on an email from us.

13.1 How to delete your data

This section is here so it can be linked on its own, because it is what reviewers of the application ask for and what you need if you ever want to leave. There are three ways to delete, depending on what you want deleted:

1. A specific item — a conversation, a file, a collection, a bot or a task: delete it from its own screen. It is really deleted, including the extracted text and the vectors of collections (section 3.4). An attachment is removed from a chat with the × next to it in the message bar: its original and its text are deleted, unless it is still attached somewhere else.
2. What is synchronised from a provider: disconnect Google, Microsoft or WhatsApp from Connections → Native, or reduce their permissions. The content synchronised from that source is deleted and your credentials are removed (section 12). The files the assistant has already worked with are not deleted this way, but as point 1 says. The step-by-step for each provider, and how to also withdraw access from the provider itself, is in How to disconnect your accounts and delete your data.
3. Everything, account included: go to Account, press Cancel account and confirm the closure by typing your email address. The application shows you the exact date it will run, which is the end of the month already paid for, and until then you can change your mind. On the day, all your content is deleted, and so is your user record, with the only two legal exceptions in section 12: your invoices and the security logs. If your account belongs to an organisation, Account shows Leave the organization instead of the closure: leaving closes it, as section 3.6 describes.

None of the above requires writing to us, and there is no processing delay. If you would still rather ask in writing, or if you are not a Genki user and want us to delete data about you held in someone else's account, write to privacy@genki.digital and we will reply within the period given in the next paragraph. In that second case, read section 4 first: it explains frankly how far what we can do reaches.

For everything else, write to privacy@genki.digital stating the right you are exercising. We will reply within a maximum of one month, extendable by two further months for complex requests, in which case we will inform you. We may ask you to prove your identity, without requesting more data than necessary to do so.

If you consider that we have not properly handled your request, you can lodge a complaint with the Spanish Data Protection Agency (C/ Jorge Juan 6, 28001 Madrid — www.aepd.es), without prejudice to going to court.

---

14. Storage in your browser and cookies

Genki installs no third-party cookies, includes no analytics or advertising tools and loads no external resources in the application.

It does use local browser storage in order to work:

Not everything it stores is a cookie, and the difference matters because they are not cleared the same way and do not last the same. All three kinds, each with its own:

Local storage is information that stays in your browser until you delete it; session storage, information that disappears on its own when the tab is closed. Neither travels to our servers on every request, unlike cookies.

Desktop notices. If you turn them on — the application offers it the first time you get a notice, never when you sign in —, your browser shows on your desktop the notices that arrive while Genki is open in a tab you are not looking at. Only the notice's title and who sent it are shown: the text does not leave the application, because the system may show the notification on a locked screen. The permission is kept by your browser, not by us, and is withdrawn from its settings. With Genki closed nothing arrives.

The consent panel covers four categories: necessary, preferences, analytics and marketing. The analytics and marketing categories are provided for the future and are inactive today: no technology of those categories is installed in the application. If any is ever added, this section will be updated and you will be asked for your decision again before it is activated.

You can accept everything, reject everything or choose by category, and change or withdraw your decision at any time just as easily, from Settings, under Privacy, in the application.

Signing out leaves no trace. On leaving, all session cookies are removed from the browser, and the session is additionally revoked on our servers (section 11). What remains are your preferences and your decision about categories, which is precisely what it would make no sense to ask you again every time.

---

15. Minimum age

Genki is intended for professional use and may not be used by anyone under 18 years of age. If we become aware that an account has been created in breach of this requirement, we will delete it.

---

16. Contact

For any question about this Policy or the processing of your data:

privacy@genki.digital Genki Solutions, S.L. — Plaza de España, 15, 2.º B, 37439 San Cristóbal de la Cuesta (Salamanca), Spain

---

17. Changes to this Policy

We review this Policy at least once a year and whenever a relevant change occurs: a new provider, a new purpose, a change of artificial intelligence model or its provider, or a regulatory change.

Where the change is substantial we will inform you by a visible means in the application or by email before it takes effect. The document header always shows the version and the date of the last update.