Privacy
How SlyOS handles your correspondence, in plain terms. Last updated 4 October 2026.
Your messages are used to draft replies for you, and to build the memory, the rankings and the relationship figures the product shows you. SlyOS does not train models on them, does not sell them and does not share them with other customers — but drafting is not something we do alone, so they are sent to the model, database, storage and push providers listed at slyos.world/legal/subprocessors. A reply reaches a recipient only when you press Send, or through an automatic reply you switched on yourself for a channel you named — nothing is switched on to begin with, and a card held for review is never sent automatically. You can delete your account yourself from Settings, and export most of what we hold; what an export leaves out is listed under “Getting your data out” below.
Who controls your data
The controller of your personal data is Belto Inc., a Delaware corporation (file number 10589231) with its registered office at c/o LegalCorp Solutions, LLC, 1221 College Park Drive, Suite 116, Dover, Delaware 19904, United States, operating from California. Reach us at support@belto.world.
This policy explains what SlyOS reads, why, how long it is kept and what you can make us do about it. It forms part of the Terms of Service.
What we collect
This is the full list, and it is longer than the sentence a privacy policy usually gives here, because the product reads correspondence and correspondence contains everything.
Messages. The body and the subject of every message on a channel you connect, inbound and outbound, the time it was sent, the title of the conversation, the links found inside the text, the drafts SlyOS writes for you, and the text of what you actually sent.
Files that arrive with a message, and what is read out of them. The file itself, its name and size, a summary, the text extracted from a document, a description and the text read out of an image, and a transcript of a voice note. Stickers, shared social posts, link previews and disappearing media are not kept.
Contacts. The display name of each person you correspond with, and their phone numbers and email addresses. SlyOS also derives and stores figures about each of them: how formally you write to them, a trust tier, and a relationship strength.
Memory. Facts a model extracts from your messages — about you, and about the people who write to you — each with a confidence score, each tied to the message it was taken from, and each stored alongside a mathematical representation of the text used for search.
Search indexes. Messages, extracted text, memory and uploaded documents are turned into embeddings — numeric representations that make the text findable by meaning. An embedding is derived from the words, so we treat it as the content it came from.
Calendar, if you connect one. Events are read from it: title, description, location and times. When you approve one, SlyOS also creates an event on it, including the email addresses of the attendees you invite.
Websites SlyOS opens for you. The site each live browser was opened for, and when it was opened, last used and closed; the sign-in a site gave you, if you signed in there; and a record of each step the browser took and of each card it asked you to press, a purchase included; and, when SlyOS used a browser you were not watching, pictures of the pages it was on, kept as files in the conversation that asked for them. “SlyOS browsing the web as you” below says more about each of those.
Your writing, as a corpus. To learn how you write, SlyOS imports messages you sent: their text, who they were addressed to, and the incoming message recovered from the quoted block — which includes the other person’s words and address. Where we measure how well a draft imitates you, the record keeps your real reply, the model’s attempt and the reasoning that compared them.
Documents you upload, split into passages and indexed, and the conversations you have with the SlyOS assistant.
Details you save for reuse. In “My details” you can store things you type in yourself — a label, the value, and a note about when SlyOS should use it. You can mark one as sensitive. They are offered to the model when it drafts, which is what they are for.
Your account. The address you signed up with, a recovery address, what you tell us about yourself — job title, a description, a default sign-off — your plan and billing state, and a record of which version of these documents you accepted and when. If you joined through somebody’s invite link, we record who invited you, and if you invite others we record whom; we use that, with your first payment and the address you signed up from, to decide whether an invite reward is due, and the person who invited you sees only a masked form of your address.
Devices and technical records. For each device you sign in on: a name, the platform, the app and operating-system version, and a push token. Usage events carry a device and session identifier. Sign-ins, login approvals, your acceptance of these documents and every entry in our security log carry the IP address and the browser or app string the request came from.
Product usage — drafts generated, edits made, features used — so the value dashboard can be accurate.
Visits to this website. The page path, the referring site’s host, campaign tags in the link you followed, a two-letter country derived at the edge, and one of three words for your screen size. No IP address and no browser string are stored for a website visit. An identifier that lets repeat visits read as one person is stored only if you accepted it. If you join a waiting list or write to us, we keep the address you used and what you wrote.
What you send us through the report control. Every SlyOS app has a control for reporting a bug or telling us something. What you type goes to us as you typed it, together with which kind of report you picked, the screen you were on, your platform and your app version. Somebody here reads it in a staff console. It is kept for as long as the account exists and goes when you delete the account. Whatever you paste into that box, you have handed to us — so treat it as you would an email to support.
Why we hold it
To show you the messages that need an answer, to draft replies in your voice, and to remember the people and commitments that make those drafts useful.
Your correspondence is never used for advertising and is never sold. It is analysed, though, and it would be dishonest to call that nothing: SlyOS ranks what needs an answer, extracts confidence-scored facts about the people you write to, and computes per-contact figures — how many exchanges you have had in the last thirty days, how quickly you typically answer, how long someone has been waiting, and whether a relationship is going quiet. Those figures are a product feature and they are sold as part of a plan. All of it is computed for your account alone, is never pooled with another customer’s and is never sold. It IS used to make decisions inside your own account, and it would be misleading to say otherwise: importance and relationship strength set how a conversation is ranked, which decides who you are shown first and when, and a per-contact switch decides whether SlyOS drafts for that person at all. Those are decisions about what YOU see. None of it is shared with anyone else, used outside your account, or used to make a decision with a legal effect on the person it is about.
Legal bases, if you are in the UK or the EEA
We rely on performance of a contract to read and draft on the accounts you connect, because that is the service you asked for; on legitimate interests to keep the service secure, prevent abuse and improve reliability; on consent for the analytics identifier described under “Cookies and analytics”, which you withdraw by clearing this site’s stored data in your browser, because there is no switch for it on the site today; and on legal obligation where we must retain records, for example for tax. Marketing email is not on that list because, as the marketing section says, it works on an opt-out rather than an opt-in, and we are not going to label it consent when it is not.
Where we rely on legitimate interests, you can object at any time and we will stop unless we have compelling grounds.
Your voice profile
The profile that learns how you write belongs to your account alone. It is not pooled with other customers, not used to improve anyone else's drafts, and never used by SlyOS as training data for any model.
Automated processing and model training
SlyOS uses automated systems, including large language models operated by our sub-processors, to read your material and produce drafts. SlyOS does not train any model on your content, does not sell it, and does not share it with other customers. Our model providers receive it in order to return a result to you; what they may do with it afterwards is governed by their own published API terms, and SlyOS has not signed a separate agreement that overrides them. The providers we use are listed at slyos.world/legal/subprocessors.
Exactly what goes to an AI provider, and which one. SlyOS asks you for this in the app, and this is the same list:
Anthropic (Anthropic PBC) receives the material a draft is written from: the recent messages in the conversation you are replying to — which includes what the other person wrote — the facts SlyOS has stored about that person and about you, and the description of your writing style. If the conversation carries a document or a photograph, the file itself is uploaded, because the draft is written by the same model that reads it rather than by a parser of ours.
OpenAI (OpenAI, L.L.C.) receives every message and draft that is turned into a search embedding, and the same drafting material as a fallback when Anthropic is unavailable. It also receives the audio of a voice note sent to you, because that is how it is transcribed — so the recorded voice of the person who sent it goes to OpenAI as sound and comes back as text.
What that means for the people who write to you. A conversation has two sides, so the material above necessarily includes messages, files and voice notes from people who are not SlyOS users and who have agreed to nothing. We would rather write that down here than leave it to be worked out.
And what SlyOS reads for you when you ask it for something. Your conversations with the SlyOS assistant, and the words on each web page its browser opens for you, go to Anthropic, and to OpenAI when Anthropic is unavailable, because the model decides what to do next from them. On a checkout those words include what is in the basket, the total, your delivery address and the last digits of a card.
The protection those providers give it, stated exactly. Both receive it for inference only — to return a result to you — and both are bound by the data-processing terms they publish, which SlyOS relies on and has not overridden by any separate agreement of its own. Neither is permitted under those terms to use what SlyOS sends to train a general model. What we will not tell you is that we have negotiated anything stronger than their published terms, because we have not. That includes retention: we have taken no zero-retention arrangement with either of them, so what we send sits with each provider for as long as that provider’s own published terms allow, and nothing we have signed shortens it. Our sub-processor register will not settle it for you either — it records who receives what, where they process it and whether message content is among it, and it carries no retention period for anyone on it. An earlier version of this sentence sent you there for an answer it does not contain.
If you do not want this, do not use SlyOS. Drafting in your voice is what the product is, and there is no version of it that does not send your material to a model. Declining the permission in the app leaves you able to sign out or delete your account. Be exact about what declining does and does not do: that screen stops the app in front of you, not the service behind it. On a channel you have already connected, reading and drafting carry on while you decide, which means material can still go to a model provider in that window. Deleting your account is what stops it.
We do build a model of how you write, and we do extract facts — about your own life, and about the people you correspond with — in order to draft in your voice. That work is per account and serves only that account. If we later train a model on your material for your own use, it stays yours and is deleted with your account.
None of this produces a decision with a legal or similarly significant effect on you. By default SlyOS proposes text and you decide whether to send it. If you switch automatic replies on for a channel yourself, a draft on that channel can be sent without a further step from you — except where the card was held for review, which is never sent automatically. You choose the channels, nothing is switched on to begin with, and you can switch it off again at any time.
Storage and encryption
Data is encrypted in transit and at rest. Device channels such as iMessage and SMS are read on your own device — by the Mac companion, or by the Android app if you have granted it notification access. Our staff can see your ACCOUNT — the address you signed up with, your plan, the channels you connected, and counts of what the product did for you — and every one of those reads is logged. They cannot see your messages, your drafts, your contacts or your subject lines: the database itself refuses those columns to the staff console rather than trusting anybody to look away.
Anything you wrote becomes readable by us only if you send it to us yourself, and there are two ways to do that: writing to support and quoting it, or typing it into the report control described under “What we collect”, whose text a person here reads in a staff console. There is no control in the product today that hands a draft to us for review, and no ticket system behind it. An earlier version of this policy described two such routes and a consent you could withdraw; the tables for them exist in our database and nothing writes to them, so the description has been removed rather than leave a mechanism named here that you cannot actually use.
Your account also has an encryption key of its own. Be clear about what it covers: the logins we hold for the channels you connected, and your two-factor secret. Those are the things that become permanently unreadable in the live service the moment the key is destroyed. Not everywhere and not immediately: for 7 days after that, our database provider can still wind the whole database back to a moment when the key was intact, which is set out under “Retention and deletion” below. Your messages are not encrypted under that key — they are protected by the encryption our database and storage providers apply to everything they hold, not by a key only you have.
Full Disk Access, and what it is for
On macOS the SlyOS companion may ask for Full Disk Access. macOS requires that permission before any application can read the local databases in which Messages and some other apps store your conversations. SlyOS uses it for exactly that and nothing else: it reads the message stores for the channels you have connected.
It does not browse your documents folder, your photo library, your keychain or the rest of your disk. It does read one kind of file off the disk and upload it: the files attached to the messages it is reading. A photograph, a voice note or a document somebody sent you in Messages is stored on your Mac rather than inside the message, so reading the message means reading that file — and SlyOS uploads it, up to 25 MB, so that it can be summarised, transcribed or read the way it would be on any other channel. Stickers, link previews and disappearing media are not uploaded.
You grant the permission yourself in System Settings and you can revoke it there at any moment; revoking it stops those channels reading and breaks nothing else.
Two more things the Mac companion does
It asks for your macOS Contacts. That is a second permission with its own prompt, separate from Full Disk Access. The message stores on your Mac hold handles, not names — a phone number or an Apple ID — so without this, every iMessage conversation is headed by a number. The companion looks a handle up in your address book and uses the name it finds. The name is sent to SlyOS with the message, as the sender’s name. Nothing else from a contact card is read, nothing is invented for a handle that does not match, and refusing the permission costs you the names and nothing else.
It drives WhatsApp Web in your own linked session. If you connect WhatsApp on the Mac, the companion runs WhatsApp’s own web client in a window it controls, in a session you linked by scanning a QR code exactly as you would to use WhatsApp on any computer. That is how it reads those conversations, and how a reply you send from SlyOS is typed and sent there under your own account. The session is yours and lives on your Mac; unlinking it in WhatsApp on your phone ends it.
The Android app and your notifications
On Android, channels such as WhatsApp, Telegram, Signal and SMS have no way to let another app read them. SlyOS reads them through Android's notification access, which you grant yourself and can revoke at any time in Android Settings. The app explains this and asks before it sends you to that screen.
While it is on, SlyOS reads notifications only from the messaging apps you have connected — notifications from everything else on your phone are ignored. For each message it reads, it sends us the text of the message, the time it arrived, the sender's name and the conversation's title as your messaging app shows them. It also sends the sender's phone number or email address when the notification carries one, so that the same person is recognised across your conversations. And it sends the files people send you — photos, voice notes, documents — the file itself, not just a note that one arrived.
Where a notification bundles several messages together, as messaging apps do when more than one arrives, all of the incoming messages in that bundle are sent, not only the newest. SlyOS does not go and fetch the rest of a conversation, it does not read your phone's address book — it holds no permission to — and it sends nothing from apps you have not connected.
This includes messages written by people who do not use SlyOS. We hold them for the same reason we hold anything else you receive — to show you what needs an answer and to draft a reply for you — and under the same terms: not pooled with other customers, never training data, and deleted with your account. Be exact about that last one, because these people agreed to none of it: deleting takes their messages out of the live service straight away, and our database provider keeps 7 days of history the database can be wound back into, so for 7 days more the whole database is still restorable to a point before the deletion, their messages with it. We do not lift one account out of that history. After that the provider has nothing left to restore them from. Revoking notification access stops the reading immediately.
Push notifications, and what Apple and Google see
To tell your phone that something has arrived, SlyOS has to hand a notification to the operating system’s own push service, and there is no way to deliver one without going through it. Today that service is Apple’s, and only an iPhone receives one: the Mac companion registers no push token at all, and the app is not built for iPad. SlyOS does not deliver push notifications to Android phones yet, and nothing has gone to Google for one. When it does, Google’s Firebase Cloud Messaging will carry them, on the terms set out below.
What that service receives is a token identifying your device, our own internal ids for the conversation and the card, a count of how many cards are waiting, and the line of text you see on the lock screen. That line names the person: the contact name we hold for them, or the conversation’s title, or — for email, where no contact matches — the subject line. So a correspondent’s name, or the subject of a message somebody sent you, passes through Apple — and will pass through Google in the same way once Android push is switched on. Neither ever receives a message body and neither ever receives draft text.
SlyOS browsing the web as you
SlyOS can open web pages and read them, and it can click and type on them for you in a browser you can watch and take over in SlyOS. When it comes to something that would send, post, book or pay, it stops and waits for you to press it, on a card below the page. A page can still send something by itself — as SlyOS types, or a while after it clicks — and SlyOS cannot always stop that. The browsing happens on a machine we run, not on your phone or your computer. It is a Chromium instance on its own server, it has no public address, and it is reached only by SlyOS itself.
We do not store your passwords. There is no password field and no password column anywhere in SlyOS. When a site asks you to sign in, SlyOS stops and hands the page to you: you see a picture of it, you tap and type on it, and what you type goes to that site and nowhere else. It is never written to our logs, never stored, and never returned in a response. A card number or a card’s security code is the same: SlyOS will not type one, and when you type it yourself on the live view, what you type goes to that site and is not written to our logs or to the record described below.
What we do keep afterwards is the session that sign-in produced — the cookies that site set in the browser, and what the site saved in the browser’s own storage for itself — so SlyOS does not have to ask you again. They are encrypted with a key belonging to your account before they are written down, they are held against the exact site they came from, and they are only ever sent back to that site. You can see every site SlyOS is signed in to, and remove any of them, at Settings → Sign in for SlyOS. Removing one deletes the stored session immediately; SlyOS is signed out there from that moment.
We keep a record of what the browser did on your behalf: the kind of step, the address, and — where something was filled in — the name of the field but never what was typed into it. That record is how you and we can both answer the question of what your agent did, and it is deliberately incapable of holding a secret. When SlyOS stops on a card for you to press — a purchase included — the record also keeps the site the page would have sent to, the name SlyOS gave the thing it was about to press, and how the card ended: pressed, declined, or left until it ran out.
When SlyOS takes you to a checkout. Anything SlyOS can tell would buy stops on a card that says it will, or may, spend money, with the amount as the page shows it, and goes ahead only if you press that card — each purchase with its own press, whatever you have switched on. SlyOS can miss that something buys, or misread the amount, so look at the page before you press; the Terms, under “Acting on your behalf”, say more. The purchase is made by you with that site, and SlyOS does not take the payment. SlyOS reads a checkout page the way it reads any page it opens, so what the page shows — what is in the basket, the total, your delivery address, the last digits of a card — is part of what it reads, and it goes to the AI providers named above. When SlyOS can tell that pressing something would add or save a card or a bank account, or move money that buys nothing, such as a transfer or money sent to a person, it refuses, even if you press.
The pages SlyOS opens are read as information, never as instructions. A page that tells the agent to send something, buy something or go somewhere is reported to you rather than obeyed, and SlyOS refuses to open addresses inside our own network.
When SlyOS acts without asking you first
By default SlyOS prepares and you press. A draft waits, a meeting is proposed, a form is filled in only when you say so. Nothing leaves your account and nothing happens on a website unless you have asked for it in that moment.
You can change that, and some devices need you to. A pair of smart glasses and the SlyOS keyring have no screen and no button to press — you speak, and it either happens or it does not. So Settings has “What SlyOS may do on its own”, where you switch on, separately and by name: putting meetings in your calendar and inviting the guests; sending a message you asked for in your own words without showing you the draft first; and building changes into your Lovable projects. There is a switch for websites there too, and today it changes nothing: whether it is on or off, SlyOS works on websites as the section above describes.
Every one of these is off until you switch it on, each is switched on on its own, and switching one on does not switch on the others. Turning them all back off is one change on that same screen, and it takes effect on the next thing SlyOS does. None of them covers a purchase: whatever is switched on, a purchase waits for your own press on its card.
When SlyOS does act on its own, it is still your action in law and in fact: it is done with your authority, from your account, and you remain responsible for it, exactly as you are for a message you send yourself. Everything it does is written into the conversation you can read in the SlyOS app, so an action taken while you were not looking is still one you can find.
The browser extension and LinkedIn
If you install the SlyOS browser extension, it runs on www.linkedin.com and nowhere else. It reads the messaging surface only — not your feed, not profiles, not search, not your connections — and only a conversation you have opened yourself. It does not open conversations on your behalf and reads nothing when the tab is closed.
From a conversation you open it sends us the text of each message in it, who sent each one as the site’s own identifier together with their display name, when each was sent, and the site’s own identifier for the conversation. It reads no cookie — not LinkedIn’s and not ours. What it keeps in your browser is the device session, which channel it posts into, up to 500 ids of messages already sent so the same one is not sent twice, and for a reply currently being placed, a digest of its text rather than the text. It writes a draft into the site’s own message box and stops there; it cannot press Send.
Other people’s messages
A conversation has two sides, so what SlyOS reads necessarily includes messages written by people who are not our users. We process them only to serve you — to draft your replies and to answer your questions about your own correspondence. What we hold about them is the same list as everything else above: their words, their name, their phone number or address, the files they sent, model-extracted facts about them, and the per-contact figures described under “Why we hold it”. All of that is built for your account and for nobody else’s. We do not build profiles of them for anyone else, do not sell them, and do not use them to advertise.
Where the law of your jurisdiction requires the other party’s consent before a communication is processed by a third party, obtaining that consent is your responsibility.
If you are not a SlyOS user and you want to know what we hold about you, or want it erased, write to support@belto.world. We would rather tell you exactly where that stands than make a promise the product cannot keep. The staff tools at Belto are deliberately built so that no one here can read contact names or addresses — that is what keeps a customer’s correspondence private — and the same design means we cannot search for you across accounts to find where you appear. Erasing a correspondent happens inside one customer’s account, and the control for it is now on the person’s own screen in the SlyOS web app. When the account holder uses it, we delete that person’s contact record, their messages in that account, their identifiers and the facts we had derived about them, and we keep a one-way keyed hash of their identifier so that the next message from them does not quietly rebuild the contact record, the identifiers or the derived facts that were just deleted — that hash is not the address and cannot be turned back into one. Five limits, stated rather than glossed. The customer’s own replies in those conversations are theirs and are not deleted. Anything already sent to a model provider cannot be recalled. The erasure reaches the live database straight away but not the backups — our database provider keeps 7 days of history the database can be wound back into, we do not edit that history and we do not restore selectively from it, so for 7 days what was just erased is still restorable from a point before the erasure, and after that the provider has nothing left to restore it from. It does not reach the corpus that teaches SlyOS how the customer writes: the customer’s own sent messages quote what the other person wrote, and that person’s words and their address were recovered from the quote and are kept there, as is their address as a recipient. And it stops the profile rather than the collection — if that person writes to the customer again, the new message is still stored in the customer’s account, with no contact record, no identifier row and no facts derived from it, kept under that account’s retention window of 365 days unless a different one has been set, and sent to the model providers if a draft is written on that thread. What we still cannot do ourselves is find you. Our staff tools are built so that nobody here can read contact names or addresses — that is what keeps a customer’s correspondence private — and the same design means we cannot search across accounts to locate where you appear. So a request from you reaches the account holder rather than a switch we can throw; write to support@belto.world and we will tell you honestly what we were able to do.
Sub-processors
SlyOS uses model and infrastructure providers to generate drafts and run the service. Your content is sent to them only to produce a result for you. SlyOS itself never trains any model on your messages. What a provider may do with content sent to its API is governed by that provider's own published terms, and we have not signed a separate agreement that overrides them. The current list is published at slyos.world/legal/subprocessors.
Two things about that list are worth saying here rather than leaving to be discovered. Files go to the model providers too, not only text: a document or a photograph is sent to Anthropic to be read, and a voice note is sent to OpenAI to be transcribed — including files sent to you by people who do not use SlyOS. And the push service described above — Apple’s today, Google’s when Android push is switched on — receives the notification line, which names a person.
International transfers
We are established in the United States. Every account is hosted in the United States today — a European region exists in the product but is not open — and our sub-processors operate there and elsewhere. So if you are in the UK or the EEA, your data is transferred out of it.
The register at slyos.world/legal/subprocessors records, provider by provider, what each one receives, where it processes it and whether message content is among it. No column of that table answers whether a data-processing agreement is in place, so the table itself will not tell you. For the two model providers that see message content it records that no separate agreement has been signed with them and that their own published API terms govern. We would rather say that here than name a transfer mechanism we cannot show you.
Cookies and analytics
Our applications set a small number of strictly necessary cookies to keep you signed in and to protect against cross-site request forgery. These are required for the service and are not used to track you.
On our public website we count visits ourselves. Be clear about the order things happen in: the visit is counted as soon as the page loads, before you are asked anything. What that count records is the page path, the referring site’s host, any campaign tags in the link you followed, a two-letter country worked out at the edge and one of three words for your screen size — no IP address, no browser string and nothing kept on your device. What we ask consent for is the one thing that would make you identifiable across visits: a small id stored in your browser so that repeat visits read as one person. That id is written only if you press Allow, and declining costs you nothing.
Your answer is remembered in your own browser. There is no control on this site today that lets you change it afterwards — the only way back to the question is to clear this site’s stored data in your browser, which also deletes the id. We would rather write that down than leave you looking for a switch that is not there.
Do Not Track, and what third parties see. We do not respond to Do Not Track or to Global Privacy Control signals. Nothing in SlyOS reads either one, so a browser set to send them changes nothing here, and you should know that rather than assume a header is doing work for you. We do not run advertising and we operate no tracker that follows you to anyone else’s site. What our public pages do is fetch things from Google as they load: the web font described next, and — on /vision and /pitch — the small app logos on those pages, which come from Google’s favicon service. Each of those requests hands Google your IP address and the page you were on, and what Google does with it is on Google’s terms rather than ours.
Our front page at slyos.world loads a web font from Google’s font service, and so do our other standalone pages — /vision, /pitch, /onepager and /phone. Opening any one of them makes your browser request that font from Google — which gives Google your IP address and the page you were on. That happens on first paint, before the cookie question is answered. The signed-in app and these legal pages do not do it; they serve the same typeface from our own servers.
Marketing email
Marketing email goes to two groups: people who joined a waiting list, and people who hold an account. If you joined a waiting list, the message you got back said your address is on the list and that we will write when there is something to send — so treat that as the terms, and not as a confirmation step you still have to complete. We are describing this as it works rather than as opt-in, because it is opt-out.
Every marketing message carries an unsubscribe link. Unsubscribing takes effect immediately, and we keep the record of it against the address rather than against an account, so it survives your account being deleted and cannot be undone by a later import. It is checked when the list of recipients is built, and that is the check doing the work. There is a second guard in the sending path and we would rather tell you it is not doing anything yet than let you count it: it reads a register that nothing writes to today, so it refuses nothing. It becomes real when we build campaigns.
Service email — receipts, security notices, and messages about your account — is not marketing and continues regardless, because we are required to send it.
Retention and deletion
Disconnecting a channel stops collection immediately.
While your account is open, message content expires. Every account has a retention window, and unless a different one has been set for yours it is 365 days; it can never be set below seven days. A job runs every hour and, once a message is older than that window, permanently removes from the live service its text, its search index, the links taken out of it, the facts derived from it, and the attachments that came with it — the file, its name, the summary, the extracted text and the transcript. The message row itself survives without its content, so counts and timings keep their history and the value dashboard does not lose last March. No SlyOS app currently offers a control to change the window; the iPhone and Android apps show you the number your account is using.
What that window does not reach, because saying so is more useful than implying it covers everything: the corpus of your own sent messages that teaches SlyOS how you write — including the incoming message recovered from each quoted block, and the other person’s address — the text of replies as sent, drafts, and your conversations with the assistant. None of those expires on a timer: they are kept for as long as the account exists, and they go when you delete it. Nor does it reach contacts: a person’s name, their phone number or address, and the figures derived about them are kept for as long as the account exists, even after every message they sent has had its text removed. Usage events are kept for 24 months and then the whole month is dropped.
Deleting your account is something you do yourself, from Settings. If you have been asked to accept a new version of these documents and have not, the acceptance screen blocks the rest of the app on every surface — phone, web and the Mac panel — but that screen itself carries a Sign out and a Delete my account control, so you can leave or erase the account without agreeing to anything. Deletion removes everything held under your account from the live service straight away — messages, contacts, memory, the voice corpus, drafts and the assistant’s history — and the files stored for you are queued for deletion by prefix and then checked to be gone. The encryption key for your account goes with it, which makes the logins we held for your channels and your two-factor secret permanently unreadable in the live service, and in any copy of the database from after that moment. Be clear about the limit: we keep no snapshots of the database and have no snapshot schedule set; what our provider keeps instead is 7 days of history that the whole database can be wound back into, so for those 7 days there is a point from BEFORE you deleted — and at that point the wrapped key and the encrypted value sit side by side, with the key that unwraps them ours rather than yours. Those credentials are not unreadable there. They stop being reachable 7 days after you delete, like everything else in that history, because after 7 days the provider has nothing left to restore them from.
Backups are the honest hard part. Your message text is not encrypted with a key of your own, so it cannot be made unreadable in that history the way a credential can. We do not edit that history and we do not restore selectively from it — we do not lift one account out of it. So here is the whole of it: your rows leave the live database the moment you delete, and for 7 days after that the content you deleted is still restorable from a point before you deleted it. After 7 days the history has rolled past that point and our database provider has nothing left to restore it from. The same goes for the text the hourly job removes while your account is open — it leaves the live database and stays restorable for the same 7 days. So does what that job strips from an attachment: the file itself sits in storage rather than in the database, but its name, its summary, the text read out of a document and the transcript of a voice note are ordinary database columns, and they are in that history like everything else. That 7 days is a setting on our provider’s project rather than anything our own code enforces, and it is what that setting says today.
Some records outlive deletion, on purpose, and there is more than one — the deletion page lists them, and the 7 days of restorable history above is one. The largest is our security log. We keep an append-only security log of administrative actions — a sign-in, a channel connected, a settings change, a send, a deletion. It cannot be edited or deleted by anyone, including us, and there is no job that expires it. When you delete your account the columns naming you are cleared, but the entries remain, and they carry the IP address and browser or app string the action came from. Some entries also carry a conversation’s subject line, the name of a file you uploaded, or the address of a mailbox you connected.
Getting your data out
You can start an export from Settings on the web app on any plan, without asking us. What it needs is a working sign-in: a closed or suspended account cannot reach Settings, and neither can an account that has been asked to accept a new version of these documents and has not — that screen stands in front of the whole app, export included. It is built and returned to you in one go rather than queued for later.
The archive contains your contacts and their identifiers, your threads and messages, your attachment records, cards, drafts, the outbox, the facts SlyOS extracted, your persona, your rules and your baselines, and your security log, which is capped at ten thousand entries and takes the oldest ones, so an account with a longer history does not get all of it.
It does not yet contain everything, and here is what is missing: the corpus of your sent messages that the voice profile is built from, your conversations with the assistant, documents you uploaded, your follow-ups, the details you save for reuse, the reports you have sent us from inside the app, the list of channels you connected, your agent settings, your subscription record, the device records we keep for you, and the profile fields on your account. If you need any of those, write to us. We would rather name the gap than let “export everything” stand.
Your rights
You can access, correct, export, restrict or delete your data, and object to processing. Export and account deletion you can do yourself; for the rest, write to us and we will answer within 30 days.
If you are in the UK or the EEA you may also lodge a complaint with your supervisory authority — in the UK, the Information Commissioner’s Office.
Children
SlyOS is not for anyone under 18. We do not ask your age and the product holds no age signal of any kind, so this rests on what you tell us when you sign up rather than on a check we perform.
There is a second and larger point, and it follows from what this product is. Because SlyOS reads both sides of a conversation, some of what it holds was written by people who never signed up, and some of those people will be under 18. Nothing here can tell. If you are a parent or guardian and believe we hold your child’s messages, write to support@belto.world — and read “Other people’s messages” above first, because the limits described there apply to this request too, and we are not going to pretend otherwise in the section where it matters most.
Changes to this policy
We may update this policy. Where a change is material we will tell you in the product or by email before it takes effect, and the date at the top always says when it last changed.
Contact
Privacy questions go to support@belto.world, or use the contact form.