Data Processing Agreement
Last updated: 20 November 2026
This agreement is part of the Blixild AI Terms of Service. It covers personal data in the content you send to Blixild AI and in the answers the models return. Blixild (Jacek Garbuliński Blixild, ul. Komunalna 3, 37-403 Pysznica, Poland. VAT PL8652593912, REGON 545317054.) processes that data on your behalf, as your processor. You are its controller or, if you process it for someone else, their processor.
1. What it covers
1.1. “Your content” is what you send to the API at api.blixild.com or with Run and Chat on the dashboard (text, and images where a model takes them), and the answers the models produce from it.
1.2. This agreement applies to personal data in your content where the GDPR (Regulation (EU) 2016/679) applies to your processing of it.
1.3. It doesn't cover the data we process as controller for our own purposes: your account, sessions, API keys, usage records, payments and emails. The Privacy Policy covers those.
1.4. If you process the data as another controller's processor, we are your sub-processor (another processor under Article 28(4) GDPR). You confirm that your controller has authorised this, and you pass its instructions to us through this agreement.
1.5. Content you send us outside the service, for example by email to support, isn't covered. Leave personal data out of it.
2. What we do with your content
2.1. We process your content only to produce the answer to each request and send it to you: we receive the request, pass it to the model on one of our GPU servers and return the answer. Annex 1 gives the details.
2.2. We don't keep your content after the answer is sent: we don't save it to any file, log or database (Annex 1 says what stays in the GPUs' memory meanwhile). We don't look at it, and we don't use it to train or improve models or for any other purpose of our own.
2.3. We process it only in the EU: on our server in Finland, which Hetzner runs for us, and on our GPU servers in Poland. Between them it travels encrypted.
2.4. The processing of each request lasts from its arrival until its answer is sent. This agreement lasts as long as you use the service, and ends when your account is closed.
3. Your instructions
3.1. We process your content only on your documented instructions: the Terms with this agreement, and each request you send (the model you choose and what you ask of it). This includes transfers to countries outside the European Economic Area: we make none.
3.2. If EU or Member State law requires us to process your content otherwise (for example, to hand it over under a court order), we'll tell you before we do, unless that law forbids telling you on important grounds of public interest.
3.3. If we think an instruction of yours infringes the GDPR or other EU or Member State data protection law, we'll tell you immediately.
3.4. You are responsible for having a legal basis for the content you send, for informing the people it concerns, and for your instructions complying with the law.
3.5. Don't send personal data to Bielik: the terms of use set by its makers forbid processing personal data with it. Don't send special categories of personal data (Article 9 GDPR) or data on criminal convictions and offences (Article 10 GDPR) to any model unless we have agreed it with you in writing.
4. Confidentiality
4.1. Only people we authorise can reach the systems that process your content; today that is Blixild's owner alone. Each person we authorise is bound to confidentiality, by contract or by law, and processes your content only on your instructions.
5. Security
5.1. We take the measures Article 32 GDPR requires, appropriate to the risk. Annex 2 lists them. We may change them, but not in a way that lowers the protection of your content.
6. Sub-processors
6.1. You give us general written authorisation to engage other processors (sub-processors). Annex 3 lists those we use now.
6.2. Before we add or replace one, we'll email you at least 14 days ahead. Until the change, you may object, with your reasons, by writing to contact@blixild.com. If we can't meet your objection, you may close your account, and we'll refund the paid credit you haven't used.
6.3. We bind each sub-processor by contract to data protection obligations no less protective than this agreement's, in particular to sufficient guarantees of appropriate technical and organisational measures. If a sub-processor fails to meet them, we remain fully liable to you for its performance.
7. Requests from the people concerned
7.1. We keep no content after the answer is sent, so we hold none to give access to, correct or delete. If someone asks us to exercise their rights over data in your content, we'll pass the request to you without undue delay, if we can tell it concerns you, and we won't answer it on the merits without your instruction.
7.2. Beyond this, taking into account the nature of the processing, we help you as far as possible, by appropriate technical and organisational measures, to answer requests from the people concerned.
8. Security, breaches and impact assessments
8.1. Taking into account the nature of the processing and the information we have, we help you meet your obligations under Articles 32 to 36 GDPR:
- security of processing: the measures in Annex 2;
- personal data breaches: if we become aware of a breach affecting your content, we notify you without undue delay, by email to your account's address. As far as we know them, we give you what Article 33(3) asks for: the nature of the breach, with the categories and approximate number of people and records concerned; its likely consequences; and the measures taken or proposed. What we don't know yet, we send as we learn it;
- impact assessments and prior consultation (Articles 35 and 36): we give you the information in this agreement and its annexes, and answer your reasonable questions about the processing.
9. At the end
9.1. When the service ends, we delete or return your content, as you choose, and delete any copies, unless EU or Member State law requires us to keep them. As we keep no content after each answer, nothing is left to return. The content of a request still running when your account closes isn't kept after it ends. We'll confirm this in writing if you ask.
10. Information and audits
10.1. We make available to you all the information necessary to demonstrate that we meet Article 28 GDPR and this agreement: this agreement, its annexes, and written answers to your questions.
10.2. We allow for and contribute to audits, including inspections, by you or by an auditor you mandate. An audit starts with the information in 10.1 and our written answers; an inspection takes place only if they aren't enough. For an inspection, you ask in writing at least 30 days ahead; inspections take place at most once a year, unless a personal data breach or a supervisory authority's request gives reason for another; the auditor is bound to confidentiality; an inspection gives no access to other customers' data, or to security details whose disclosure would put the service at risk; each party bears its own costs.
10.3. Business customers who need a signed copy of this agreement or answers to a security questionnaire can write to contact@blixild.com.
11. Liability, precedence and changes
11.1. Liability between us under this agreement follows section 8 of the Terms. This doesn't limit liability towards the people concerned under Article 82 GDPR.
11.2. Where this agreement and the rest of the Terms differ about personal data in your content, this agreement prevails.
11.3. We change this agreement as we change the Terms (section 10): by email at least 14 days ahead.
11.4. Polish law applies (section 11 of the Terms).
Annex 1. The processing
- Subject matter: producing the models' answers to your requests.
- Duration: each request, from its arrival until its answer is sent; this agreement, while your account exists.
- Nature: we receive the request over HTTPS on our server in Finland; check the key and your credit; pass the request, encrypted, to the model on one of our GPU servers in Poland; the model computes the answer in the GPU's memory; the answer returns the same way. The language models keep computations from recent requests in the GPU's memory (a cache that lets a later request with the same beginning be answered faster) until later requests take the space or the model restarts. Each customer's cache is kept apart: a request can reuse only the computations of the same customer's earlier requests. They aren't written to disk, and no one looks at them. A model that takes images finds and outlines objects in them and returns their boxes, scores and outlines; it doesn't identify people, and its answer contains no image.
- Purpose: giving you the answers you ask for; nothing else.
- Types of personal data: whatever personal data you put in your content: in text, for example names, contact details and the contents of documents, messages and records; in images, where a model takes them, people who may appear in photographs, workers among them: their likeness and anything else in the picture that identifies them, such as a face, a name badge or a number plate; and what the models write about these people, and the outlines drawn around them. Special categories of data and data on criminal convictions and offences only where we have agreed it with you in writing (clause 3.5).
- Categories of people: the people your content concerns, for example your customers, employees and correspondents and the people named in your documents or shown in your photographs, and you and your own users.
- Places: our server in Helsinki, Finland (Hetzner); our GPU servers at our premises in Poland; between them, the traffic travels encrypted, through Tailscale.
Annex 2. Security measures (Article 32 GDPR)
- Encryption in transit: HTTPS (TLS) between you and our server in Finland; WireGuard (Tailscale) between it and the GPU servers.
- No content stored: request and answer contents are held only in memory while the answer is computed; they don't go into usage records, logs, databases or files, and the servers don't swap memory to a disk, except where the law requires us to keep them.
- Access: each customer's requests carry their own API key, and we store keys encrypted (AES-256-GCM); each GPU server accepts requests only with our Finnish server's key; that server can reach only the models' ports on the GPU servers (Tailscale's access rules); the administration interfaces can't be reached from the internet; servers accept logins by SSH key only; every server has a firewall.
- Separation: the services on our server in Finland run in containers without administrator rights, under an unprivileged account; each customer has a limit on requests at once, and each customer's prompt cache on the GPUs is kept apart.
- Availability: every model is checked every minute, with alerts on failures; the databases are backed up every night, encrypted (they hold no content).
- Testing: the service's automated check runs after each change to the models' configuration and sends marked requests, answered and refused, to every model, then searches every server for the mark; we review the network and the servers' exposure (last on 6 and 8 November 2026).
- People: only Blixild's owner has administrative access (section 4).
- Places and physical access: our server is in Hetzner's data centre in Helsinki; the GPU servers are at our premises in Poland, which aren't open to the public, and only people we authorise work on them. The servers' disks hold no customer content.
Annex 3. Sub-processors
- Hetzner Online GmbH (Germany) runs our cloud server, through which every request and answer passes. The server is in Helsinki, Finland.
- Tailscale US Inc. (USA) carries the encrypted traffic between our server in Finland and our GPU servers when they can't connect directly; it can't read it. We allow only its relay servers in the EU.
Resend (email), Stripe (payments), Google and GitHub (sign-in) never receive your content.