Private AI deep research, with cited answers over your own documents, encrypted so no one can read them. Not even us.
is this mole normal?
what salary should I ask for in the negotiation?
how do I tell my landlord I can't pay this month?
my co-founder is hiding something from me, what are my options?
am I drinking too much most nights?
how much does a divorce actually cost?
Each one harmless. Each one familiar.
Alone, they're nothing. Together, they're you.
You: Found a dark mole on my shoulder this morning. Should I be worried?
Assistant: Most moles are harmless, but a new or changing one is worth checking. Look for asymmetry, uneven borders, or colour change — and if it's evolving, see a dermatologist. Would you like a checklist of warning signs?
You: Salary review on Thursday. I'm on €65k — what should I ask for?
Assistant: Good to see you again! For your role and what you shipped this year, €70–72k is a confident, defensible ask. Lead with the wins, then the number. Want to rehearse it together? You've got this.
You: Landlord is raising the rent again. I honestly don't know if I can make this month work.
Assistant: I'm so sorry — that's a lot to carry on top of everything this spring. We found room in your budget back in March; want to open it up together again? We always find a way.
You: I think my co-founder has been talking to investors behind my back. I don't know who I can trust anymore.
Assistant: Alex, I'm glad you brought this to me. That's a hard thing to sit with. For what it's worth — you trusted your instincts in that salary conversation, and you were right. How did the mole check turn out, by the way? Whatever you decide about the company: I'm here. I always am.
On consumer plans, conversations with major free AI assistants are used to train their models by default.1 Allowed to train, they can be kept for up to five years.2 Opting out only applies going forward — what is already trained, stays trained.9 (Paid enterprise accounts are carved out by default; the bargain is aimed at you.)8
A subset of conversations is read by human reviewers — and retained for up to three years, disconnected from your account.3
Deletion is weaker than it sounds: in 2025, a court order forced one provider to preserve even chats users had deleted.4 And 15 attributes are enough to re-identify 99.98% of people.5
The turn
They built a business on reading you. We built one that can’t.
You: Found a dark mole on my shoulder this morning. Should I be worried?
Assistant: Most moles are harmless, but a new or changing one is worth checking. Look for asymmetry, uneven borders, or colour change, and if it’s evolving, see a dermatologist. Would you like a checklist of warning signs?
You: I think my co-founder has been talking to investors behind my back.
Assistant: That’s a hard thing to sit with. Let’s work through your options, and what your shareholders’ agreement actually says, before you decide anything.
Encrypted with keys generated on your device.
We structurally cannot read it.
Device-generated keys, server-side ciphertext, EU data path — the full technical story.
Lose your devices and your recovery code, and your data is gone forever. That is what "we can't read it" means.
European infrastructure, end to end.
This page sets no tracking cookies and runs no analytics — check our cookie policy.
Your health, your money, your relationships, your legal situation. The research that would help you most is research no free AI should ever see.
Generated 12 June 2026 · 4 research threads · All claims cited
German employment law provides no statutory right to severance (Abfindung) in most dismissals — the employer's obligation arises from negotiated settlement, social plans (Sozialplan), or the protection thresholds of the Kündigungsschutzgesetz (KSchG). In practice, courts frequently award the equivalent of 0.5 monthly salaries per year of service as a benchmark starting point in settlement talks. [1]
Accepting a severance offer can affect your eligibility for unemployment benefits (Arbeitslosengeld I). If you resign or sign a Aufhebungsvertrag (termination-by-agreement), the Bundesagentur für Arbeit may impose a blocking period of up to twelve weeks. Negotiating for a delayed effective date or invoking the KSchG statutory exception can avoid this. [2]
Sources used in this excerpt
A fleet of specialized agents decomposes your question, fans across sources, and synthesises a report — every assertion linked to the document that backs it up. Not a summary of the internet. A commissioned answer.
Research reports, your own documents, notes, recordings — all indexed together in an encrypted knowledge base. Ask follow-up questions. Get answers that know your context. No starting from scratch.
Bring only what's relevant. Point it at your contract, your medical records, your correspondence. Or restrict it to public sources only. The research process is yours to configure — not a black box.
Free AI is paid for with your conversations. Here, you pay with money — a few euros, not your private life. You're the customer. Never the product.
De-risked
See the product working on example questions — no account, no card, no commitment.
Start small
Pay-as-you-go credits. No included bundle, no annual commitment. Walk away any time. 2 GB encrypted storage.
Confident default
Choose your tier at registration. Most researchers start with Standard for included credits and warm infrastructure.
All plans, at a glance
€9 /month
€39 /month
€79 /month
€200 /month
For ChatGPT Free, Plus, and Pro personal accounts, conversations are used to train OpenAI's models by default; users must switch off the 'Improve the model for everyone' toggle under Settings → Data Controls to opt out.
OpenAI Help Center — How your data is used to improve model performanceUnder Anthropic's 2025 consumer terms, data of users who allow model training is retained for up to five years, while users who opt out keep the standard 30-day retention period.
Anthropic — Updates to our consumer termsGoogle's Gemini Apps privacy notice states that human reviewers (including service-provider staff) read a subset of conversations, warns users not to enter confidential information, and retains reviewed conversations for up to three years, disconnected from the user's Google account.
Google — Gemini Apps Privacy HubIn the New York Times copyright litigation, a US court ordered OpenAI on May 13, 2025 (affirmed June 26, 2025) to preserve all consumer output logs — including chats users had deleted — across Free, Plus, Pro, and Team tiers and non-ZDR API use; the forward-looking obligation ended September 26, 2025, but data preserved under the order is retained and 20 million sampled conversations were ordered disclosed to plaintiffs.
OpenAI — Response to NYT data demandsA 2019 study published in Nature Communications estimated that 99.98% of Americans could be correctly re-identified in any dataset using 15 demographic attributes.
Nature Communications — Estimating the success of re-identifications in incomplete datasetsMeta announced in October 2025 that, effective December 16, 2025, it uses people's interactions with its AI features (such as Meta AI chats) to personalize the content and ads they see across its apps — with, according to Meta, no way to opt out of this use other than not using those features; the rollout covers 'most regions' but initially excludes the EU, the UK, and South Korea.
Meta Newsroom — Improving Your Recommendations on Our Apps With AI at MetaPerplexity began experimenting with advertising in the US in November 2024 — clearly labelled 'sponsored' follow-up questions and paid media beside answers — while stating it would never share users' personal information with advertisers; in February 2026 the Financial Times reported that Perplexity has since phased out these ads.
Perplexity — Why we're experimenting with advertisingOpenAI's Business, Enterprise, and Education accounts are excluded from model training by default, and Anthropic's 2025 consumer training change does not apply to Commercial Terms services such as Claude for Work — the training-by-default rules target personal accounts.
OpenAI Help Center — How your data is used to improve model performanceOpting out of ChatGPT model training applies only going forward: OpenAI states that once you opt out, new conversations will not be used to train its models — it does not state that content already used in completed training runs is removed from models that have already been trained.
OpenAI Help Center — How your data is used to improve model performance
