Vent First Journal

Privacy Policy

Last updated: 2026-11-03

This policy describes what the app stores, what leaves your device, and how long anything is kept. Where we cannot yet state something truthfully, it is marked as pending rather than filled in with a plausible number.

Who we are

Vent First Journal is operated by Individual Entrepreneur Bondar Oleksandr Viktorovych, a sole trader registered in Ukraine.

If you have a question about this policy or about your data, write to bav@vcity.com.ua.

The short version

What we collect, and why

What you write

Your journal entries, your check-ins, the messages in a conversation and the answers you tap during onboarding.

This is the app. We store it so that your journal is there when you come back, and so that a reply can take into account what you already said.

An installation identifier

When the app first runs, it creates a random identifier for that installation and a credential that stays in the device's secure storage. That identifier is the whole of your "identity" here — it is not your Apple Account or Google Account, not your device's advertising identifier, and not linked to any of them.

Your interface language

So a reply comes back in your language.

Your subscription state

Whether a subscription is active, which one, and when it renews — so that paid access works and can be restored on a new device. We never see your card or your payment details. Payment happens inside Apple's system on iPhone and inside Google Play on Android, and neither gives us those.

Google Play purchases, on Android

On Android a subscription is bought through Google Play. To check it, the app sends our server the purchase token Google Play issued and the product, and the server asks Google Play for the purchase's current state. When the purchase is made, Google Play is also given a reference we derive from our internal user record with a keyed hash — not your installation identifier itself — so that the purchase can be matched back to this installation.

We erase the copy of the token kept with the purchase record when our server records that the purchase has been acknowledged to Google Play — whether our server or the app confirmed it. If confirmation keeps failing, the copy becomes due for removal after Google's three-day acknowledgement period and is erased by the next regular clean-up. What we keep after that is the subscription state above and a fingerprint of the token (a SHA-256 hash), linked to you until you delete your data.

The purchase check described above does not depend on what follows. While Google Play notifications are switched on for our server, Google Play also tells it when a subscription renews, is cancelled, expires or is refunded. These notifications arrive through Google Cloud Pub/Sub, and once a day our server also asks Google Play for the list of refunded purchases. A notification carries its own copy of the purchase token and has a 32-day processing period from receipt. Its token is removed when processing ends, or by the regular clean-up after that period expires — for example, while the purchase is not yet linked to an installation. The rest of the record becomes due for deletion 32 days after processing ends and is removed by the regular clean-up. Notifications carry no name, no email address and nothing you wrote.

A pseudonymous abuse-prevention token

To stop automated abuse of the app's public endpoints we need to count requests. Your raw IP address is not stored in the rate-limiter's table. What is stored is a keyed pseudonym — an HMAC derived from the address, with separate key spaces per policy. The row becomes due for deletion two minutes after its counting window ends (windows last not more than about an hour), and is removed by the regular clean-up.

This token is a pseudonym, not anonymous data: it is derived from an identifier, and we describe it as a pseudonym rather than as anonymised or irreversible. It is kept apart from your journal and is not used to look anything up about you.

A cache key for one stretch of conversation

When a request goes to the AI provider, it carries an opaque key so the provider can reuse its prompt cache instead of re-reading the beginning of the conversation every time.

It is not a direct identifier: it is a keyed hash on a separate server-side subkey, and neither the installation identifier nor the raw session number ever leaves us. But it is stable across one stretch of conversation and it goes to the provider, so the provider is able to group the requests of that stretch together. We say this plainly rather than calling the field an implementation detail.

If you report a reply

In app versions that offer "Report this reply", you can report a reply from Jolly that our server has stored: press and hold it, or use that action with a screen reader. Nothing is sent until you tap "Send report".

What a report contains. That one reply (the team reads the copy our server already stores), the reason you choose, and a comment if you write one. Your own messages are not attached. But if the reply repeats something you wrote, that part is in what the team reads.

Who reads it. A person on the Vent First team reads the reply you reported and your comment — that reply only, not the rest of the conversation. We use reports to review and fix how the app replies, not to train a model of our own. A report is not sent to the AI provider. There is no account, so we have no way to answer you about a report; you can still write to us.

How long. A report is kept until you delete your data. Deleting your data removes your reports too.

A report does not depend on the AI processing consent below: it goes to our team, not to an AI provider, and only when you tap "Send report".

Your consent, and how to take it back

What you write in the journal, together with the state you tap before sending it, can say something about your emotional or mental state. That is a special category of personal data, and "you agreed to the terms" is not a basis for processing it. So the app asks you separately.

Before the first send. The first time you would send a journal entry or a message to Jolly, a screen explains what is processed and asks you to agree. Nothing is pre-ticked, and the two buttons are "I agree" and "Not now". Until you agree, your text, your selected state and the check-in do not leave your phone, and our server does not read them — not even to classify them. If you choose "Not now", the screen closes and the app keeps working: settings, support and deleting your data are all still there. It is not raised again by itself — sending again shows a brief notice with a way back to it — and you can agree later.

What does happen before you agree. The app may create the pseudonymous installation identifier and credential described above, so that a request can be signed at all. That is not your journal text, not the state you select, and not an Apple device identifier. Nothing you write in the journal or the conversation, and no state you select, reaches our server or an AI provider until you agree.

Taking it back. Settings → AI processing consent withdraws it. New journal entries and messages to Jolly stop immediately on the phone, and the server refuses them from then on. Withdrawing does not delete your journal — that is a separate button with its own question — and you can agree again if you change your mind.

Reporting a reply is separate from this consent. A report goes to our team for review, not to an AI provider, so this consent is not what allows it: in app versions that offer reporting, you can report a reply that is already stored even after withdrawing. Nothing is sent until you tap "Send report", after the report screen has shown you the reply and said that a person will read it. The reply may repeat something you wrote, and your comment, if you add one, is read too — see If you report a reply.

Emergency information does not depend on this. The screen that points to local emergency services and crisis lines is part of the app, needs no consent and makes no network request.

We record only that consent was given, which version of this text it was given for, and when — against the same anonymous installation the app already uses. No IP address, no device fingerprint and no new identifier is created for it. Deleting your data removes the consent record with everything else.

Where your text goes

To compose a reply, the text needed to answer you is sent to external AI service providers used to generate replies. We do not use these providers for advertising, and we do not give them your data for their own marketing.

Who the recipients are. The category is third-party AI infrastructure and model providers. We name the category rather than a list of models on purpose: models are replaced as better ones appear, and a page that reads like a catalogue would be out of date the week after it was written — while the category, the purpose and the safeguards stay the same.

What that means concretely today. In production the text goes to a single AI provider: DeepInfra, reached directly, with no gateway or broker in between. It is the only AI provider we send your entries to and the one we hold the agreement with; the companies that run its infrastructure are named further down. There is no automatic second attempt at another provider: if that one call fails, you get a short local message instead of a reply, and no other provider is contacted. OpenRouter receives no production journal text. Its credentials are not present in the production environment at all, and the server refuses to start if they are found there. Amazon Bedrock and Google Vertex are not part of the route either: earlier versions of this page named them as possible serving endpoints, and that described a gateway route which is no longer used.

Which models, and which are excluded. The models in use are open-weight ones served by DeepInfra itself. Models that DeepInfra resells on behalf of Anthropic or Google are excluded, and the server refuses to start if one is configured — because DeepInfra's own no-training commitment names those two families as carve-outs, so the protection this page relies on would not cover them.

How the retention limit works — and how it does not. With DeepInfra it is contractual, not a per-request switch. Their terms commit them not to retain, store or log the data we send beyond the period strictly necessary to process and return that request, and that commitment stands at the account level rather than being something our code asks for on each call. Read it for exactly what it says: a limit on how long our data may be held, not a promise that no copy of it is ever made. Their own documentation goes further and describes inference as running in memory rather than being written to disk — that is their account of their systems, and we cannot verify it from outside. We say this plainly because the previous version of this page described a per-request field, and that was a different provider's mechanism. What we can verify on our side is which provider is configured and that no second one is contacted; the retention commitment itself is theirs.

This describes the route that is switched on today, not every route that could ever be switched on. Before a different route is enabled, this page and the actual transmission settings are checked and, where they differ, updated first.

What we do not promise here. We do not promise that no copy exists anywhere on a provider's side for any purpose, and we do not promise the absence of technical metadata. Operating an API produces records of its own — request timing, volume, billing and abuse-prevention data — and a provider may retain content briefly where its own terms require it. These are their terms, not a guarantee we are able to give on their behalf.

Service providers generally

We use service providers only to operate the app. What binds them is their published terms together with the settings on our account — described above — and applicable law. We are a small self-serve customer of the AI provider, which is the level at which that company contracts: we have not negotiated a separate agreement with it, and we do not claim to have.

Hosting and database. A dedicated server rented from OVH, located in Canada. The virtual machines on it are managed by us, not by the hosting provider.

Who else may be involved. DeepInfra is the only AI provider your text is sent to and the direct recipient of the request. It is not, however, a company that runs on hardware it owns alone, and it publishes the list of companies working on its behalf. That list — last updated by them on 6 September 2024 — names three: Amazon Web Services and Google Cloud Platform as infrastructure, and Stripe as its payment processor. It is published at docs.deepinfra.com/account/subprocessors, given here as an address rather than a link because this page keeps a closed list of destinations it will offer you.

What that means for your words. Running on someone else's infrastructure means that AWS and Google Cloud may process your text on DeepInfra's behalf while a reply is being generated. They are not recipients we chose, we hold no agreement with them, and we cannot describe for you the terms DeepInfra has with them — what we can tell you is that DeepInfra names them as its subprocessors, which is the honest shape of it: your words are handled inside a service that itself runs somewhere. DeepInfra lists Stripe only as its payment processor, not as an infrastructure or AI-processing provider. VentFirst purchases are processed by Apple on iPhone and by Google Play on Android. What that list settles is Stripe's role; we have not seen the contract behind it, so we do not tell you that no access is technically possible — only what the roles are.

Google Cloud here is not the Google models we exclude. Two different roles are easy to read as one. Above, Google Cloud is infrastructure that DeepInfra's own service runs on. Separately, models that DeepInfra resells on behalf of Anthropic or Google are excluded from our route and the server refuses to start if one is configured, because the no-training commitment this page relies on carves those two families out by name. Infrastructure is not resale, and excluding the second does not remove the first.

Transfers. Our server is in Canada, for which the European Commission has recognised an adequate level of protection for commercial organisations. DeepInfra operates from the United States. The public release is offered outside the EEA, the United Kingdom and Switzerland; we do not claim a transfer mechanism we have not put in place. If that public availability changes, the transfer basis, processor arrangement and this policy are settled before the territory is enabled. Separately, a small, invitation-only closed test of the Android app may include selected testers in any country enabled for that test, including the EEA, the United Kingdom and Switzerland. Only people we invite can join, and this does not make the app publicly available in those countries.

Apple processes your purchase on iPhone. We receive from Apple only what is needed to grant access — which product, and whether it is active. On Android, Google Play processes it; what we exchange with Google and how long we keep it is described under Google Play purchases.

Your voice

Speech is recognised on your device, by the system's own recogniser, with on-device recognition required. Audio is not uploaded, not stored by us, and not sent to any third party. Only the text that appears on screen — and only after you send it — reaches our server.

Microphone and speech-recognition permission are asked for by iOS the first time you use dictation; on Android, the system asks for microphone permission. On Android, dictation needs the phone's on-device recogniser, which is available from Android 13; where it is not, dictation is not offered rather than falling back to a recogniser that sends audio away. You can refuse; all journaling features remain available by typing.

What we do not do

How long we keep things

Journal, check-ins, conversation, onboarding answers, installation identifier, language, subscription state
Until you delete them.
Reports you make about a reply — the reply, the reason and your comment
Until you delete your data.
The copy of a Google Play purchase token kept with the purchase record
Erased when our server records the acknowledgement. If confirmation keeps failing, due for removal after three days and erased by the next regular clean-up.
Google Play notifications about a purchase — renewal, cancellation, expiry, refund
Their own copy of the purchase token: removed when processing ends, or by the regular clean-up after the 32-day processing period from receipt. The record without the token: due for deletion 32 days after processing ends, then removed by the regular clean-up — until then the same notification delivered again is recognised.
Fingerprints of Google Play purchase tokens
Kept — see What stays on purpose.
Records that let an interrupted send be recovered without duplicating it
24 hours.
Abuse-prevention counters (the keyed pseudonym above)
Due for deletion two minutes after their counting window ends (windows last not more than about an hour), then removed by the regular clean-up.
Service-wide daily cost totals (no identifier of any kind)
Kept; they contain nothing about a person.
Per-installation daily usage counters (a keyed pseudonym of the installation)
Expire at the end of the day (UTC) plus an hour and are removed by the regular clean-up; deleting your data does not remove them sooner.
Records of Apple's subscription notifications
Kept; there is no automatic clean-up of them today — see What stays on purpose.
Server and proxy technical logs — the request time, the route, the outcome, the duration, the IP address and technical request identifiers, never journal text
Not more than 14 days.
Database backups — a deletion is applied to the live database at once, but a backup taken before it may still hold the data until that backup expires
Taken once a day, kept 14 days, then deleted automatically.

Deleting your data

Open Settings → Delete my data inside the app.

This removes your journal entries, check-ins, conversations, onboarding answers, reports you made about a reply, installation credentials and the other data linked to you — your user record and everything attached to it, including the subscription right recorded on our side and, on Android, the record of which Google Play purchases are yours. It happens in a single transaction: either all of it goes or none of it does, so a half-deletion cannot be reported to you as success.

The same steps, with what to do if the app is no longer on your phone, are on Delete your data.

Backups, said plainly

Your data is removed from the live database at once. Backups created before that moment may still contain it until they expire, on the schedule in the table above. If an older backup is ever restored, the deletion has to be applied again to the restored data — that step belongs to the operating procedure, under which a restore is exercised into a separate temporary database after each migration and its result recorded.

There is no second layer of copies: the production host is a dedicated server, the virtual machines on it are managed by us, and the hosting provider takes no whole-machine snapshots.

So we do not tell you that pressing Delete leaves no copy anywhere at that instant. It leaves no copy in the database the app reads and writes, and the backups age out.

Your phone's own backup. On Android, the phone may back up app settings to your Google Account. For this app that can include the installation identifier and settings such as the theme — never what you write and never the credential, which is excluded from backup. Deleting your data clears them on the phone; a backup made earlier keeps them until Android replaces it, and on its own it cannot reconnect to the deleted data.

What stays on purpose

Some records deliberately survive a deletion. None of them is anything you wrote, and each is named here.

Cost counters, of two kinds. The service-wide daily totals of what replies cost us hold no identifier at all — not your installation, not your address, nothing that leads back to a person — and are kept. The per-installation daily counters that enforce the daily limits are keyed by a pseudonym of the installation — an HMAC, not the identifier itself. They survive a deletion; they expire at the end of the UTC day plus an hour and are removed by the regular clean-up. Deleting them at once would turn "delete my data" into a way to reset a daily limit.

One record about a refunded purchase. If Apple refunds a subscription, we keep one compact record per Apple subscription transaction chain for which a refund was recorded, and this is all of it: which Apple environment it came from (production or sandbox), Apple's original transaction (chain) identifier, the purchase date of the refunded version, the date it was refunded, and the times we wrote the record. Several refunds in the same chain merge into that one record. There is no receipt in it, no signature, no Apple Account, no product name, no Vent First user, profile, installation, or device identifier, and not a word of anything you wrote.

We do not call that transaction identifier anonymous. It is Apple's identifier for a purchase chain, and Apple can connect it back to the buyer. What we can say is narrower, and you can hold us to it: after your deletion our copy carries no Vent First user, profile, installation, or device identifier and is not linked to your journal, your conversations or your profile — everything else we held about you is gone, and nothing in our database points from it to a person.

It is kept for as long as a signed refunded transaction can be reused, which in practice means indefinitely — we would rather tell you that than name a short period we do not keep to.

Why it exists, and it is the only thing we use it for: a refunded purchase can be presented to us again, from the same phone or from a new one, and without that record the refund and the subscription would both stand at once.

Records of Apple's subscription notifications. Apple tells our server when a subscription renews, expires, is refunded, and so on. For each notification we keep a record: Apple's notification identifier, its type and subtype, the Apple environment, the date Apple signed it, a SHA-256 fingerprint of the notification as received, Apple's original transaction (chain) identifier, how and when we processed it, and — only for a refund or a revocation — the purchase and refund dates of that version. They exist so that each notification is processed once and in the right order. There is no receipt in it, no Apple Account, no product, no Vent First user, profile, installation, or device identifier, and nothing you wrote. A deletion does not remove these records, and there is no automatic clean-up of them today, so in practice they are kept indefinitely. As with the refund record, we do not call the chain identifier anonymous: Apple can connect it to the buyer.

A fingerprint of each Google Play purchase token. For every Google Play purchase our server has recorded, we keep the token's SHA-256 fingerprint, whether it came from Google's live or test environment, and — if Google replaced that purchase with a newer one — the fingerprint of the replacement and when that happened. There is no token in it, no product, no Vent First user, profile, installation, or device identifier, and not a word of anything you wrote.

We do not call this fingerprint anonymous either. Google Play holds the token, and anyone who holds the token can compute the same fingerprint. What we can say is the same narrower thing as above: after your deletion nothing in our database points from it to you. It is kept indefinitely, for one reason: so that a purchase Google has already replaced cannot be presented to us again, after a deletion, as if it were a current one.

Google Play notifications, said plainly. When you delete your data, every notification still waiting to be applied that we can connect to your installation — through the purchases recorded as yours and the earlier purchases they replaced — loses its purchase token at once; the record stays, without the token, until it becomes due for deletion 32 days later and is removed by the regular clean-up, so that the same notification delivered again cannot bring the token back.

Some notifications cannot be connected to you that way: one about a newer purchase Google issued as a replacement that had not yet been linked to your installation when you deleted your data, and any notification that arrives after the deletion. These keep the purchase token until processing ends or their 32-day processing period expires, and their record is removed by the regular clean-up once it is due, 32 days after processing ends. If such a purchase is later restored on a new installation, a waiting notification can be applied to that installation.

So pressing Delete does not leave us with no technical or financial records at all. It removes your journal and the records that point to you, on the terms above; the records named here stay, on purpose.

Deleting your data does not cancel your subscription. Apple or Google Play owns that — see below.

Your subscription

Manage or cancel at any time in Settings → Subscription in the app, or in iOS Settings → your name → Subscriptions. Cancelling stops the next renewal; it does not refund the period you already paid for. Refunds are handled by Apple.

On Android, manage or cancel in the Google Play app: your profile picture → Payments & subscriptions → Subscriptions. Cancelling stops the next renewal, and refunds are handled by Google Play. A subscription you keep stays with your Google Account after you delete your data, and a new installation of the app can be linked to it again by restoring the purchase.

Your rights

Depending on where you live you may have the right to access, correct, delete or export your data, to object to or restrict processing, and to complain to a supervisory authority.

Deletion is available inside the app and needs no request. For anything else, write to bav@vcity.com.ua. Because there is no account, we may not be able to connect a request to a specific installation without information from the app itself — we will tell you what is needed rather than guess.

The legal basis for processing, where that concept applies, is the performance of the contract between you and us (running the app you asked for) and our legitimate interest in keeping the service from being abused. For the part that can say something about your emotional or mental state — the state you tap and what the app derives from what you write — the basis is your explicit consent, asked for separately and withdrawable at any time, as described above.

Children

This app is for people aged 16 and over, everywhere it is offered, and the app asks before anything else. If you tell us you are under 16, the app does not continue: no account and no identity are created, and no request is sent to us. The app does keep a small technical note on the device — that the answer was given and which minimum it was given under — so it does not ask again on every launch and can ask again if the minimum changes. If you later tell the app you have turned 16, it asks the question once more.

What this app is not

Vent First Journal is a self-reflection journal. It is not therapy, not medical care, not a diagnosis, and not a crisis service. If you may be in danger, contact your local emergency number or a crisis line in your area.

Changes

If this policy changes materially we will update the date at the top and, where the change affects what leaves your device, say so in the app.

Contact

bav@vcity.com.ua · Individual Entrepreneur Bondar Oleksandr Viktorovych, sole trader, Ukraine