DiveLook helps you look at a decision from several angles. DiveLook's own Topic AI Engines first remove personal
details from your question, then write the viewpoints with the help of a large language model (LLM). We want you to know
exactly what happens to your question. Here is the short version:
2. What we use and why
Your questions and answers
When you click "Dive", your question and settings (topic and intelligence
level) go to our server. There our Topic AI Engines first remove personal details from your question (see
below), then write the viewpoints using the language models of OpenAI,
our LLM provider, and send the answer back to your browser. We don't store your questions or
answers, in a database or in our logs.
Each question is handled on its own: while it is being answered it exists only in our server's memory,
and it is deleted as soon as the answer is sent. Nothing links one question to the next.
As our LLM provider, OpenAI works only on our behalf. Under its API terms it does not train its models on this data, and
it may keep it for up to 30 days to catch abuse.
If you ask with an account, each question also carries a one-way code made from your account. It is never your
email address, name or account number, it stays the same for your account, and OpenAI cannot turn
it back into who you are. It lets OpenAI's abuse checks stop one person's misuse without
blocking DiveLook for everyone. Legal basis: our legitimate interest in keeping the service safe and available
(GDPR Art. 6(1)(f)).
Legal basis: providing the service you ask for (GDPR Art. 6(1)(b)) and your consent (Art. 6(1)(a)).
Personal details are removed before any AI sees them
Before your question leaves our server, our own software replaces personal details with neutral
placeholders. This covers people's names, email addresses, phone numbers, national ID numbers, card and bank
account numbers, IP addresses, links, street addresses and social media usernames.
OpenAI, TypeSafe and the safety checks only ever see the version with placeholders. When the answer comes
back, our server puts your details back so the answer still reads naturally. Your details exist only in
memory while your request runs, and are never saved or logged. Places and company names are kept, because
the advice needs them and on their own they don't identify you.
No filter is perfect. It can miss nicknames, unusual spellings or details described in
words. You never need passwords, ID numbers or health records to get good advice, so describe your
situation in general terms.
Data minimisation and privacy by design (GDPR Art. 5(1)(c) and Art. 25).
DiveLook's knowledgebase
DiveLook has its own knowledgebase, with extra knowledge on how people make decisions, written by us. When
it fits your question, our engines use it in the background. This runs entirely on our own
servers, with personal details already removed: no outside service sees your question for it, and nothing
is stored.
Legal basis: providing the service you ask for (Art. 6(1)(b)).
Safety checks and topic suggestions
To block harmful or illegal requests, stop attempts to trick the AI, and offer helplines to anyone who may
be in crisis, your question and the answer (with personal details already removed) are checked
automatically. This is done by our own filters, by TypeSafe AI, Inc. and by OpenAI's
moderation service. While you type, TypeSafe may also read your question (again with personal details
removed) to suggest the best topic angle. TypeSafe does not train on your data. A blocked request is not
stored by us.
Legal basis: your consent (Art. 6(1)(a)) and our legitimate interest in a safe service (Art. 6(1)(f)).
Your account
You can read DiveLook without an account; asking a question needs one. For your account,
Supabase, our sign-in provider, stores:
- your email address, display name and an encrypted password;
- your plan (Standard or Pro) and, for Pro, the price, the next renewal or end date and the payment status that Stripe reports, so My Account can show them;
- the date of your last sign-in and last signed-in visit;
- if you changed your profile picture, the parts you chose (such as skin tone and hair);
- a record of your answers to these texts: the version you first accepted, your last 5 acceptances and when you last chose "Not now", each with its date and time, so your other devices don't ask again and we can show what you agreed to.
Your browser keeps you signed in until you sign out. Your profile picture is a drawn face, made in your browser from
your account ID and any parts you chose. No photo is uploaded, and no picture or avatar service receives anything. When you create an account or change its email
address, Supabase checks that it is an address you can use for an account (not a throwaway or alias address,
and not an inbox that already has one). Nothing extra is stored for this check.
Legal basis: our agreement with you (Art. 6(1)(b)).
When you delete your account, it is closed at once and deleted for good 30 days later. Until then you can sign
in and keep it, and nothing else can be done with it. We wait these 30 days so you can change your mind, and so
we can still answer payment and refund questions, deal with misuse (such as deleting an account to start again
after breaking the rules) and answer a lawful request from an authority. If you need it gone sooner, contact us.
Legal basis for the 30 days: our legitimate interest in handling payments fairly and keeping DiveLook safe (Art. 6(1)(f)).
When you sign in, create an account or reset your password,
Cloudflare Turnstile first runs a quick check that you are a person and not an automated program. For this, your browser loads Cloudflare's
check from Cloudflare's servers, and Cloudflare sees your IP address and basic browser and device details. It
never sees your questions or answers. Cloudflare uses this only to run the check, not for advertising.
Supabase receives only the result of the check.
Legal basis: our legitimate interest in protecting accounts from automated abuse (Art. 6(1)(f)).
Pro subscription (optional)
If you upgrade to Pro, you buy it on a secure page run by Stripe. Pro is sold through
Link, Stripe's payment service, which acts as the reseller (merchant of record): it takes
the payment, handles tax, sends receipts and answers payment questions. For that, Stripe receives your
email address, card details and billing address, and handles them as its own controller under its privacy
policy. On its own pages, including your Link account, Stripe may also ask for or show details such as your
name or phone number; these stay with Stripe and never reach us. We never see your full card number.
Stripe tells us whether your subscription is active. We save only that status and Stripe's customer and
subscription IDs with your account.
If you end Pro and ask for the unused days back after choosing not to accept an update, we email our team
your account email, account ID, Stripe's customer and subscription IDs and the versions involved, so we can
make the refund. We note the date of the request with your account.
Legal basis: our agreement with you (Art. 6(1)(b)), our legal duty to keep accounting records (Art. 6(1)(c)) and, for the record of your answers to these texts, our legitimate interest in being able to show what was agreed (Art. 6(1)(f)).
Help assistant (Pearl)
If you chat with Pearl, our AI help assistant in the corner of the page, each message and the last few
messages of that chat go to our server. Our server removes personal details in the same way as for your
questions, then sends the text to OpenAI, which writes Pearl's reply.
TypeSafe AI checks each message and each reply for safety, also with personal details removed.
Pearl only knows public facts about DiveLook. It cannot see your account, your questions on the workspace
or your answers. We don't store the chat: it lives only in your open browser tab and is gone when you
reload or close it. Harmful requests are refused automatically.
Legal basis: your consent (Art. 6(1)(a)) and our legitimate interest in helping you use DiveLook (Art. 6(1)(f)).
Live web check (optional)
Pro members can turn on the Live web check. It is off unless you turn it on, and your choice is kept only in
your browser. When it is on and your question needs current facts (for example whether a website can be
trusted), personal details are removed first, as described above. Only short search words, or the name of a
website you mention (never the rest of a link), go to Tavily, which
runs the web search. Your full question is never sent to the search service. The results are
used by the AI services listed in section 4 and shown with the answer as a list of sources.
When your question names a website, our server also looks up only that site name in free public records
about websites, such as domain registries, a public list of the most visited websites and a public malware
list. They receive the site name and nothing else, and the results stay in our server's memory for up to 24
hours.
To keep the service fair, our server counts how many web checks each Pro member uses per day,
under a scrambled code instead of your account. These counts are deleted when the day ends.
Legal basis: our agreement with you (Art. 6(1)(b)).
Ask this viewpoint
Members can ask one viewpoint of their answer short follow-up questions about the same situation, from that viewpoint's side; questions about anything else are declined. When you ask, that viewpoint's text, your
original question, your question about it and your earlier questions to that viewpoint go to our server.
Personal details are removed in the same way as for your questions, then the same AI services as for your
answer write and check the short reply (see section 4). Nothing is stored: the questions and replies live only
in your open browser tab and are gone after a different question, Reset or a reload. To keep it fair, our server
counts how many viewpoint chats each member uses per day, under a scrambled code instead of your account, saves
these counts on our server and deletes them when the day ends.
Legal basis: our agreement with you (Art. 6(1)(b)).
Assignment Planner
The free Assignment Planner is for members with a confirmed email address. Without an account you can
see example plans prepared in advance: they run no AI, and we receive nothing you type. When you make an assignment plan, your brief,
the level, type and word count you chose and any marking criteria go to our server. Personal details are
removed in the same way as for your questions, then OpenAI models write
the outline and TypeSafe AI checks the brief and the outline for safety, also with personal
details removed. When a brief is about a well-known person, such as an author or a historical figure, that name
can stay so the outline makes sense; the names of people you know are hidden. Nothing is stored: the plan lives only in your open browser tab and is gone when you close
the tool or reload. A PDF you download is made in your browser and is not sent to us.
To keep it free and fair, our server counts each member's assignment plans a day, under a scrambled code instead of your
account, saves these counts on our server and deletes them when the day ends (midnight UTC). Like a question, an
assignment plan carries the one-way code made from your account. Assignment plans go through the same misuse checks as questions,
so those the safety checks refuse count towards the pause described under "Technical and security data" below.
Legal basis: providing the service you ask for (Art. 6(1)(b)) and, for the counts, our legitimate interest in keeping a free tool fair and safe from misuse (Art. 6(1)(f)).
Task Planner
The free Task Planner is for members with a confirmed email address; without an account you can see
example plans prepared in advance, as for the Assignment Planner. When you make a task plan, the
task you describe and your answers to its questions (where you are, what is holding you back, the area and
who is doing it) go to our server. Personal details are removed in the same way as for your questions, then
OpenAI models write the plan and TypeSafe AI checks
the task and the plan for safety, also with personal details removed. The name of a well-known person the task
is about can stay; the names of people you know are hidden. Nothing is stored: the plan lives only
in your open browser tab and is gone when you close the tool or reload.
Its task plans are counted each day, and checked for misuse, in the same way as for the Assignment Planner, but
apart from assignment plans. The counts are saved on our server under a scrambled code and deleted when the day ends
(midnight UTC).
Legal basis: providing the service you ask for (Art. 6(1)(b)) and, for the counts, our legitimate interest in keeping a free tool fair and safe from misuse (Art. 6(1)(f)).
Conversation Planner
The free Conversation Planner is for members with a confirmed email address; without an account you can
see example plans prepared in advance, as for the Assignment Planner. When you make a conversation plan, the talk
you describe and your answers to its questions (who they are to you, what you want from the talk and what is
holding you back) go to our server. Personal details are removed in the same way as for your questions, the
other person's name too, then OpenAI models write the plan and
TypeSafe AI checks the talk and the plan for safety, also with personal details removed.
Nothing is stored: the plan lives only in your open browser tab and is gone when you close the tool or reload.
A talk often holds details about another person, and sometimes about sensitive matters such as health, family
or a relationship. The names of people you know and other personal details are hidden before the text reaches
an AI (only the name of a well-known person the talk is about can stay), and we never store what you write. Please share only what you need for the plan.
Its conversation plans are counted each day, and checked for misuse, in the same way as for the Assignment
Planner, but apart from the other tools. The counts are saved on our server under a scrambled code and deleted when the
day ends (midnight UTC).
Legal basis: providing the service you ask for (Art. 6(1)(b)) and, for the counts, our legitimate interest in keeping a free tool fair and safe from misuse (Art. 6(1)(f)).
Read aloud
When you press 🔊 on a viewpoint or Read in its Detailed View, the text on your screen goes to our own voice
service, which turns it into speech on our servers. No outside service receives it, and the text is never
logged or stored; the sound is kept only for a short time in our server's memory, so the same text is not
made twice. If our voice is not available, your browser's own voice reads instead; depending on your browser
and device, that voice may come from your browser or system maker.
Legal basis: providing the service you ask for (Art. 6(1)(b)).
Emails about your account and DiveLook
Besides the sign-in emails above, we may email you about important changes: to your account, your plan, the Pro
price, this policy or the Terms, or DiveLook itself. We never send ads or newsletters. Such an email goes to every
member with a confirmed address (or only to Standard or Pro members). We read the
addresses from your account when we send and don't keep a list of who got the email.
Legal basis: our agreement with you (Art. 6(1)(b)).
Contact form
If you message us, your name, email address, topic and message are delivered to our mailbox through our
email provider (Oracle Cloud Email Delivery or Resend). We use them only to reply to you.
Legal basis: our legitimate interest in answering you (Art. 6(1)(f)).
Sharing DiveLook by email
If you use "Share DiveLook" to email a friend, we send one invitation to the address you type, through
Brevo. The email contains only a link to DiveLook and, if you add them, your first
name and a short note (links are removed). We do not store the address or the note. To stop the same address
getting more than one invitation a day, we keep a scrambled, one-way code of it in memory until the end of the
day. Your friend is told how they got the email and will not hear from us again.
Legal basis: your request to send the invitation, and our legitimate interest in letting people tell others about DiveLook (Art. 6(1)(f)).
Optional feedback scores
Now and then, after an answer, we ask one short question, such as "Did these viewpoints help you think it
through?". You can skip it. If you tap a reply, we keep only: which question it was, your score (or which
viewpoint you picked), the day, the topic, the intelligence level and a few labels about
how the answer was made (such as which features were on). We use these numbers to
see which answers help people and to improve them. We never keep your question, the answer,
your account, your IP address or anything else that shows who you are, so a score can't be linked back to you.
The scores are stored on our own server, are not shared with anyone, and are deleted after 12 months. Your
browser remembers only the day you last saw a question, so you are asked at most once a day.
Legal basis: our legitimate interest in improving DiveLook's answers (Art. 6(1)(f)).
Technical and security data
Cloudflare (USA) is our content delivery and security provider: every visit to DiveLook
passes through its network before it reaches our server. Cloudflare delivers the pages quickly, keeps the
connection encrypted and protects DiveLook from attacks, bots and abusive traffic, for example by blocking
suspicious requests. For this it sees your IP address, the pages and features your browser asks for and basic
browser and device details. Your questions and answers travel encrypted between your browser, Cloudflare and
our server; Cloudflare only forwards them and does not keep or use them. It also works out which country a
request comes from, and if a page fails to load, your browser may send Cloudflare a short error report so the
problem can be found. Cloudflare works for us under a data processing agreement and never uses this data for
advertising.
Legal basis: our legitimate interest in a fast, secure and working service (Art. 6(1)(f)).
Like every website, our server receives your IP address and basic request details. We use your IP address
only in memory to stop abuse (for example, too many requests per minute) and never build profiles from it.
To show the Pro price in euros or US dollars, and to check that questions and chats are available where you
are, we read the country Cloudflare works out from your IP address for that one request; we don't
keep it. Our hosting provider may keep short technical logs for security. Our fonts and the sign-in library come
from our own server, not from Google or another outside service. The one exception is the
security check for sign-in and invitations, which loads from Cloudflare (see "Your account" above).
When you open an Insights article, we add one to that article's read count, which orders the "Popular"
list. The count is only a number. To count each reader once a day, we keep a scrambled code made from your
IP address and the article in memory only, and throw it away every day.
Each plan has a number of answered questions per day, counted by your account, so people who share a network (a school,
an office, shared Wi-Fi) each keep their own count. To stop misuse, such as one person using many new
accounts or one account shared by many people, we also keep simple daily counts per account and per network
(IP address).
For these numbers our server keeps only scrambled codes of the account and the network, never the account
or the IP address itself. The codes change every day, so days cannot be linked, and the numbers are deleted
when the day ends (midnight UTC). If many questions in a row are refused by the safety checks, new questions
are paused for a while; that pause is kept under a scrambled code of the account until a day after it ends.
Legal basis: our legitimate interest in a secure, working service (Art. 6(1)(f)).