What we keep, and who reads it.

• We keep what you give us to find you a match, and nothing to sell.
• Your problem is never shown to another company. The platform and its AI model read it; that is how matching works.
• Figures are compared inside the platform. The other side learns that they fit, never what they are.
• One kind of cookie: the one that keeps you signed in. No ads, no trackers.
• We do not use what you write to train AI models.

Your account. Your work email and a password, which Supabase stores hashed; we never see it. If you sign in with Google, we receive your email and name from Google instead.

Your company. Name, website, what it does, industry, size, city, languages and logo — mostly read off your website and confirmed by you. The name and role of the person who takes meetings, shared only once both sides agree to meet.

Your problems. The text, the interview you had with the assistant, the part of the business it sits in, and its terms: budget ceiling, contract formats, start date and hard requirements.

Your services. What you sell and who it is for, and its terms: price floor, contract formats and when you are free.

Matches. For each pair: a score, an explanation written for the other side, the transcript of the two agents, what they settled, the status of the double opt-in, and when each side last opened it.

The directory. Public facts about companies read from their own websites — name, domain, description, logo. It holds no information about people.

“Private” is enforced in the code and the database, not only promised in a prompt:

• The database lets only your company read or change its problems. Another company’s account cannot fetch them at all.
• The other side’s agent is a separate AI call that is never handed your problem, not even a summary. It knows only what your agent says out loud.
• Every transcript is checked for your wording before it is stored. One that fails is thrown away, never shown.
• Budgets and price floors are compared in one place in the code. Neither figure enters the other side’s prompt or screen.
• A counterparty stays anonymous — industry, size, city — until you both agree to meet.

The platform itself can read your problem, and so can the AI model it sends it to. We say “withheld” rather than “encrypted” for exactly that reason.

• To run the service you signed up for — interviews, matching, negotiation, briefs. This is the contract between us (GDPR Art. 6(1)(b)).
• To keep the directory of public company information, and to keep the service secure and working. This is our legitimate interest (Art. 6(1)(f)); you can object to it.
• We do not use your data for advertising, we do not sell it, and we do not profile people.

These providers process data on our behalf, only to run Crossdesk:

• Supabase — the database and sign-in.
• Vercel — hosting, server logs and page-view counts.
• The AI model provider — currently DeepSeek, reached through an Anthropic-compatible interface. It receives problems, services, interviews and transcripts to run the assistant and the agents, and processes them outside the European Economic Area.
• Tavily — reads a public website when our own reader cannot open it.
• Google — only if you sign in with Google.
• Stripe — only if your company pays the fee for an accepted meeting. Card details go to Stripe and never reach us.

Crossdesk sets one kind of cookie: the Supabase session that keeps you signed in. It is needed for the service to work, so there is nothing to consent to and no banner.

Page views are counted with Vercel Analytics, which sets no cookies and stores no personal data. There are no advertising or third-party tracking scripts.

To see what is slow or broken, the server writes short lines to its log: which AI step ran, how long it took and how many tokens it used; which pages of a website were read and the profile drafted from them; how many turns an interview had. They hold no problem text. When the privacy check stops a transcript, the log records which line failed and why, never the words themselves. Vercel keeps logs for a limited time and then deletes them.

When you sign up, Crossdesk reads your company’s public website to pre-fill your profile; nothing is saved until you confirm it. The same reader builds the directory. It obeys each site’s robots.txt and reads a handful of pages, not the whole site. If your company is in the directory and you want it corrected or removed, ask us.

• Your account, company, problems, services and matches: for as long as the account exists.
• A transcript that fails the privacy check: not at all — it is discarded before it is stored.
• Server logs: for the hosting provider’s short retention period.
• When you ask us to delete your account, your company and everything under it goes too.

Under the GDPR you can ask us to:

• show you what we hold about you, and give you a copy you can take elsewhere;
• correct it, or delete it;
• stop using it for our legitimate interest, or limit how we use it.

If you think we got it wrong, you can complain to the Estonian Data Protection Inspectorate (Andmekaitse Inspektsioon, aki.ee).

When this policy changes in a way that matters, we tell you before the change applies. The date at the top says when it last changed.