Privacy Policy

DeepShrink is run by DeepShrink Inc. (in the process of being incorporated) ("we", "us"). We are the controller of your personal data. This policy covers the DeepShrink website and the iPhone and Android apps.

You can reach us about your data at privacy@deepshrink.com.

We collect what you give us and a small amount of technical data:

We use your data only for the purposes below. We do not sell it, we do not use it for advertising, and we do not use what you write to train AI models.

What you write about your feelings, mood, sleep, symptoms and life can reveal information about your health. Under the GDPR (Article 9) this is special category data, and we process it only with your explicit consent.

Before you start, we ask for that consent with the checkbox "Store and use what I write to support my reflection". Without it, DeepShrink stores none of your reflections. Two further consents are optional and off unless you turn them on: suggesting things to remember from your conversations, and finding patterns across your check-ins. You can change both at any time in Settings.

You can withdraw your consent at any time in Settings → Privacy. Withdrawing it stops us storing or processing anything new you write. It does not delete what you have already written: export it, reset your reflection data or delete your account if you want it gone. Urgent help, your settings, export and deletion keep working.

If you turn off memory suggestions, pending suggestions are deleted and kept memories stay until you delete them. If you turn off patterns, we stop working them out and hide past weekly look-backs; they stay stored until you reset your reflection data or delete your account.

When we change how we explain this consent, we ask you to confirm again before you write anything new.

One more consent is off unless you turn it on: your map (You, Your map). With it, DeepShrink keeps what seems to help you, worked out from how you use the app as counts and fixed labels, never your words or anything you keep to yourself, and uses it to fit the companion's tone, the order of exercises and the nightly question to you. It is never used for ads, prices or offers. Turning it off deletes the map. Statistics about which approach helps whom include only people who also allow usage statistics, in groups of at least 20.

The DeepShrink companion is an AI, not a person. AI models run by our AI providers generate its replies, summarise long conversations, suggest and keep memories (only if you turned that on, or when you ask it to remember something), check your messages for safety, translate a message and its reply when needed, and word your weekly look-back (only if you turned that on). Trends and patterns are calculated without AI.

To do this we send our AI providers the text needed for each task: your message and the conversation so far (or a summary of its older part), the name you asked the companion to use, your instructions to it, the memories you keep and, in your first conversation, some of your onboarding answers. Your check-ins and notes are never sent to the companion. The safety check receives what you write in conversations, onboarding, guided exercises, check-ins and diaries: first a fast screen of each message with up to three earlier messages, then, when that screen sees a possible risk or is unavailable, a fuller check with up to six earlier messages. The weekly look-back receives your scores, emotions, exercise titles and memory labels, never your notes. We never send your email address or account name.

Your journal entries reach an AI in three ways only: the safety check reads an entry when you finish writing it, "Talk about this" starts a conversation with it, and Ask your reflections can read it to answer a question (only if you turned that on). Switch on "Keep out of AI" for an entry and only the safety check still reads it: no conversation, answer or pattern uses it. Entries you import from Day One or Apple Journal are read on our servers without AI, and are kept out of AI unless you choose otherwise.

Today our AI providers are Together AI (first choice for almost every task), Replicate (first choice for translating your message and the reply when the companion cannot write your language, and a backup for some tasks if Together AI is unavailable) and TypeSafe (the fast safety screen, which receives every message described above). They receive the text only, never your email address, username or account id. Together AI does not store the text it receives, and Replicate deletes the inputs and outputs of each request after one hour by default. None of them uses what you write to train AI models.

We may change which AI provider handles a task. Before we add a new one, we update this policy and ask you to confirm your consent again.

No decision with legal or similarly significant effects on you is made by AI alone. The safety check only changes what DeepShrink shows you and pauses reminders for a while. AI replies can be wrong; you decide what to do with them.

Dictation and Read aloud work on your device. When you dictate or talk out loud, your browser, or our app, turns your voice into text on your device: your voice is not sent to us or to any provider, and we receive only the words you send or save. Read aloud uses a voice on your device and sends nothing. The one exception is a choice you make: where your browser cannot turn speech into text on your device, members on a paid plan can turn on Better voice in this browser. What you say is then streamed from your browser to Soniox, in the United States, only to turn it into text; Soniox keeps neither the audio nor the text and never uses them to train models. It counts toward your AI allowance, and you can turn it off at any time in Settings → Privacy.

Your conversations are private. Staff can only open one under strict, logged rules:

Your data export includes a staff access log that lists each time a staff member opened one of your conversations, with the date. If you send us something yourself, for example in a support request, staff read what you send.

We use these service providers. Each may use your data only to provide its service to us, except where we say it acts on its own account. We never send what you write to emergency services, your contacts, advertisers or data brokers.

We host DeepShrink in the EU. Some providers above process data outside the EU and the UK, including in the United States: our AI providers, Soniox if you turn on Better voice in this browser, payment providers, Apple and Google, and Sentry and PostHog if used.

Where that happens, we rely on an adequacy decision (such as the EU–US Data Privacy Framework, for certified companies) or on the European Commission’s standard contractual clauses with the UK addendum. You can ask us for a copy of the safeguards.

We keep your account and what you write for as long as you have an account, unless you delete it sooner or choose to delete chats automatically. Everything else we keep has a time limit, listed below.

You have these rights under the GDPR. If you live in the UK, the UK GDPR and the Data Protection Act 2018 give you the same rights.

To use a right, use the controls in Settings or write to us at privacy@deepshrink.com. We answer within one month. If you think we have not handled your data properly, you can complain to the data protection authority where you live or work. In the UK that is the Information Commissioner’s Office (ICO).

DeepShrink is for people aged 18 or older. Before you can use it, you confirm that you are 18 or older. If you say you are under 18, you are signed out and none of your reflections are stored.

If we learn that someone under 18 has an account, we will delete it.

DeepShrink is not an emergency or crisis service. Nobody watches your conversations in real time.

When the safety check finds signs that you may be in danger, DeepShrink shows crisis lines and emergency numbers on your screen. It never sends anything to anyone on your behalf: not to emergency services, not to the people in your safety plan, not to anyone else.

If you are in danger, call your local emergency number. The Urgent help page lists crisis lines and works without an account.

Every request is limited to the signed-in user’s own data, and automated tests check that one user can never reach another’s. Passwords are stored as hashes, and you can turn on two-factor sign-in.

Logs are set to hide what you write. Reminders and background jobs carry ids, never your words. Keys for our AI providers are stored encrypted. Staff access to conversations works as described above.

No system is perfectly secure. If a breach puts your rights at risk, we will tell you and the authorities as the law requires.

If you join the waitlist, on the home page or the waitlist page, we store your email address, your name if you give one, when you joined and confirmed, and the IP address and browser details of the request. We send one email to confirm the address, and it joins the waitlist only when you open the link in that email. After that we use it only to send your invite. The legal basis is your consent, which you can withdraw at any time by writing to privacy@deepshrink.com.

To stop the form being misused, we count sign-up attempts per IP address and per email address for one hour.

On the public pages (the home page, pricing, Urgent help and these legal pages) we count visits with our own copy of Plausible Analytics. For each visit it records the page without anything after it in the address, the site you came from, your country and your type of device and browser, and it counts waitlist sign-ups. It sets no cookies, stores no IP address and does not follow you across sites, so it needs no consent.

We will update this policy when DeepShrink changes. If a change matters to you, we will tell you in the app or by email before it takes effect. If we change how we explain a consent, we ask you to confirm it again.

DeepShrink Inc., privacy@deepshrink.com.