Effective date: 15 November 2026 Controller: Opafex Ltd ("Opafex", "we", "us"), 1 Coldbath Square, Farringdon, London, EC1R 5HL, company no. 17330428. Privacy contact: privacy@opafex.com. ICO registration: ZC196031.
This policy explains how we handle personal data for the Opafex platform ("Service"). We act in two roles: as a controller of data about our own account holders, and as a processor of the data our customers route through the AI agents they build. Where we act as a processor, our customer is the controller and our Data Processing Agreement governs that processing.
1. Who this applies to
- Account holders (our customers — businesses and individuals who sign up).
- Visitors to our website, opafex.com, including anyone who sends us an enquiry, completes our fit questionnaire or audit intake, or uses the chatbot on our site (see §2).
- End individuals whose personal data a customer's agent processes (e.g. the customer's own contacts, email/SMS recipients, leads). For these people, the customer is the controller; contact them first, but you may also contact us at the address above and we will route your request.
2. What we collect and why (controller role)
| Category | Examples | Purpose | Lawful basis (UK/EU GDPR) |
|---|---|---|---|
| Account & identity | email, name, password (hashed), license key | Create and secure your account | Contract (Art 6(1)(b)) |
| Billing | Stripe customer ID, invoices, usage/token counts | Take payment, meter usage, tax | Contract; Legal obligation (tax records) |
| Authentication & security | session tokens, IP address, user-agent, device/location, audit logs | Keep your account secure, detect abuse, meet our security duties | Legitimate interests (Art 6(1)(f)); Legal obligation |
| Support & comms | messages you send us, product emails | Respond to you, operate the Service | Contract; Legitimate interests |
| Product usage | feature/telemetry events, error reports | Operate, debug and improve the Service | Legitimate interests |
We do not sell personal data. We do not use account holders' content to train AI models (see §6).
Visitors to opafex.com. If you send us an enquiry or an audit intake, we collect your name, email, company, your message or answers, and any documents you upload. We store them on our platform and email them to our own inbox so we can reply. The fit questionnaire stores your multiple-choice answers without your identity; if you then leave your details, we add your name, email and whether you want to hear from us. If you use the chatbot on our site, our AI model provider (Google) processes your messages to answer them, and any email or contact details you give it are kept like an enquiry. Card payments are handled by Stripe; we never see card details. Our website is hosted by Vercel, which also provides our cookie-free visitor analytics. Lawful basis: legitimate interests, or steps before a contract when you ask us for a service.
The document check on our site. The document check at opafex.com/document-check reads your file in your own browser; the file itself is never uploaded. Before any text is sent, your browser replaces names, addresses, contact details and account numbers with placeholders, and shows you the exact text that will leave. Our model provider (Anthropic) processes that text to produce the findings, and their terms exclude training on it. We store none of it: not the text, not the findings, not the name of your file. The report is built on your own computer. Pseudonymised text can still be personal data, which is why we send as little as possible and keep none of it. If you ask to hear when the rules change, we keep your name, work email, firm, the regulators you picked, the rule references your documents were checked against and the kinds of document, so we can email you when one of those rules changes. Nothing from any document is kept with it, and we do not know which of your documents relied on which rule. Every one of those emails has a link that stops them, and you can ask us to delete your details at any time. Lawful basis: legitimate interests, and your consent for the emails.
Our plugin in ChatGPT. If you use the Opafex plugin in ChatGPT, ChatGPT decides what to send us from your conversation. The plugin accepts only fixed choices, such as your sector, team size or the kinds of data involved, and numbers you give for a payback estimate. It does not accept names, contact details or free text, and we do not receive your conversation. Requests come from OpenAI's servers, not your device, so we do not see your IP address. We use each request only to return the answer. We log which tool was used, the sector if given, and the time, to count how often the plugin is used; we cannot tell who you are from that log. Lawful basis: legitimate interests. Links in the plugin's answers are tagged so we can count visits that come from it. OpenAI's handling of your conversation is governed by OpenAI's privacy policy.
If we contact your business. We contact businesses about our audits and services. We find them through the
Companies House register, job adverts and their own websites, and sales partners working for us may add what they
find or are told. We hold the business's name and the name, role, work email, work phone number and notes of the
person we are in touch with. Before we call a number we check it against the Telephone Preference Service and the
Corporate Telephone Preference Service, through a screening provider. Contact details in our sales partners'
workspace are deleted after 12 months with no activity. Lawful basis: legitimate interests, in offering our
services to businesses. You can object at any time by telling the person who contacted you or by emailing
privacy@opafex.com. We then stop, and keep only what we need to make sure we do not contact you again.
3. Customer content we process (processor role)
When you use an agent, we process whatever you and your agent's data sources contain — prompts, conversation messages, uploaded documents, knowledge-base content, and data pulled from integrations you connect (e.g. CRM, email, GitHub). This may contain personal data about third parties. You are the controller of that content; you must have a lawful basis and provide any required notices to those individuals. Our handling is governed by the DPA.
4. AI processing and third-party model providers
The Service is powered by AI. When you interact with an agent, you are interacting with an AI system, and your content is sent to third-party large-language-model (LLM) providers to generate responses. By default your agent's AI processing stays in the EU. PII-bearing calls are pinned to a paid, no-training tier either way. The full list is in our Sub-Processor List. Who processes what:
- Mistral (France) serves every agent in the EU region, which is the default.
- AWS Bedrock, through an EU inference profile, is the in-region alternative to Mistral. Your uploaded files are stored with AWS in London.
- Anthropic (US) is the primary provider for an agent you set to "no preference" in its Settings, with Google and OpenAI (US).
- Google indexes your documents and search queries in London (Vertex AI), unless you are on Private Mode or your account handles sensitive data, whose documents are not sent for embedding and are found by their words instead. It answers the website chatbot, in the US or in London with the US as its fallback, and it generates images, in the US, even for an agent in the EU region.
- OpenAI (US) indexes an agent’s own learned memory and translates interface text, whichever region you choose. It describes uploaded images for an agent set to "no preference".
- Mojeek (UK) answers an agent’s web searches.
Groq is available only as a Bring-Your-Own-Key option.
EU processing. An agent in the EU region — the default — is served by models running in the EU: Mistral (France) first, with AWS Bedrock as its alternative. Those model calls never fall back to a provider outside the region; if no in-region model is available, the call fails rather than leaving it. What the list above says happens in the US, the website chatbot, translation, an agent's memory index and image generation, still happens there.
You can reduce or eliminate this exposure:
- Bring Your Own Key (BYOK): route AI calls through your own provider account.
- Private Mode: keeps content on your own database, uses your keys and/or a local model, and restricts our telemetry to non-content metadata only. Ask us for current availability and pricing.
How the agent checks its own work. Some of what an agent does is checked by a separate, narrow AI decision — for example, whether a finished task or workflow did what was asked. These checks run on an EU-hosted model (Mistral), see only the run they are checking, and today only record their answer: they do not approve, block or change anything. They never make decisions about people.
5. International transfers
We are based in the UK and serve UK, EU and US users. Some sub-processors — notably the US LLM providers and parts of our infrastructure — process data in the United States.
Our hosting provider, Railway Corporation, stores all platform data at rest. We have an executed Data Processing Addendum with Railway dated 14 August 2026, which incorporates the EU Standard Contractual Clauses and the UK International Data Transfer Addendum, so that transfer is covered by appropriate safeguards.
For the remaining US sub-processors, principally the AI model providers, we rely on the safeguards in each provider's own data processing terms, which incorporate the SCCs and UK Addendum. We are still completing that documentation — the executed agreements and a transfer risk assessment — and not every agreement is yet on file. If the transfer position for a specific provider matters to your decision to use the Service, write to privacy@opafex.com and we will tell you exactly where it stands rather than give you a general assurance. Private Mode, BYOK and local models are available to keep content within your chosen region.
6. Model training
We do not use account holders' content to train our own models, and we will not enable any "improve the model with your data" processing without separate, opt-in consent.
We select AI providers' paid API tiers, whose published terms state that data submitted through the API is not used to train their models, and the platform enforces this in code: calls that may carry personal data are pinned to those paid tiers and are blocked from free/consumer tiers whose terms permit training. We are in the process of obtaining written confirmation of the no-training position from each provider; until that is complete, what we can state is the tier we buy and the control we enforce, not a warranty on the provider's behalf. Those providers keep what they receive for their own standard period under their terms, and we have no zero-data-retention agreement with any of them.
"Did this do what you wanted?" After a task or workflow finishes, you may be asked whether it did what you wanted. Your answer is recorded against that run and is used to measure how far the agent's own checks can be trusted. If you add a comment, it is saved to that agent's knowledge so the agent can take it into account next time; you can see, hide or delete it on the agent's Intelligence page, and it is deleted with the agent. This is not model training: the comment is stored as text the agent can read, like an instruction you write. Answering is optional, and dismissing the question records no answer.
7. Retention
We keep personal data only as long as needed for the purposes above. Key points: account data is kept for the life of your account and deleted within 30 days after you close it, apart from the records we must keep, such as billing, which have your details removed. If you ask us to erase your data from an account that continues, records that belong to the account (approvals you decided, changes you made to the agent's knowledge, widgets you saved) are kept with your identity removed; a note you typed stays as written unless you ask us to remove it. If your subscription ends and you have not closed the account, it is paused rather than closed: nothing on it runs, you can still sign in, and everything on it is kept for 12 months so you can restart where you left off. If it has not been restarted after 12 months, the account and its data are deleted. You can ask us to delete it sooner at any time; security and audit logs are kept up to 24 months; billing records are kept as long as tax law requires (~6–7 years). Customer content is retained per your instructions and the DPA, and deleted on your request or account closure. Deleting a conversation also removes what the agent learned from it (the summary, facts and decisions it drew from that conversation); clearing old conversations in bulk removes the transcripts but keeps what was learned, and says so before it runs. Notes the assistant saves for itself while you chat are not linked to a single conversation, so deleting one does not remove them; they are deleted with the agent, or on request. Records of the agent's checks — which check ran, its answer and a pointer to the run, never the content it looked at — are kept for 12 months and then deleted automatically, and are deleted with the agent on request; aggregate accuracy figures contain no personal data and are kept. The ChatGPT plugin's usage log (§2) is kept by our hosting provider for no more than 30 days. Enquiries and audit intakes, with any documents uploaded, are deleted 24 months after we were last in contact; contact details left on the fit questionnaire are removed after the same period, leaving only the anonymous answers. Our records of the businesses and people we have contacted about our services are deleted 24 months after we were last in contact, except what we need to make sure we do not contact you again, which we keep for as long as we need it. Conversations with the chatbot on our site are deleted after 30 days. If you asked to hear when the rules change, those details are deleted 24 months after we were last in contact, which means 24 months after the last email we sent you, or after you stopped them; asking us to delete them removes them straight away.
Trading desk. If your account is a trading desk, its trading record — the proposals it raised, your decision on each, the orders it placed, the trades it journalled and the desk's own audit log — is append-only: an entry cannot be edited or individually deleted, including on request, because a record that can be rewritten is not a record. It is kept while your account is open, for up to six years from each entry, in line with UK practice for financial records. Closing the account deletes the desk at once, with its trading record and the broker key you entered, rather than within the 30 days that apply to the rest of your account. Orders already at your broker stay there and stay yours: we do not cancel or close them, and an exit the desk itself was holding, rather than your broker, will no longer happen.
8. Your rights
Subject to law, you can request: access to your data, a copy in portable form, correction, deletion,
restriction, objection, and withdrawal of consent. Contact privacy@opafex.com. We respond within
one month.
Complaints. You can complain to us about how we use your personal data: email privacy@opafex.com with
"complaint" in the subject. We acknowledge a complaint within 30 days and respond without undue delay. You can
also complain to the UK ICO (ico.org.uk) or your local EU supervisory authority at any time.
US residents (California and other states): you have rights to know, access, delete, correct, and opt out of "sale"/"sharing" — we do not sell or share personal data as those terms are defined, and we honour opt-out signals.
9. Security
We use encryption in transit (TLS) and encryption at rest for credentials and API keys (AES-256-GCM), access controls, audit logging, and tenant isolation. No system is perfectly secure. If a breach puts personal data at risk we tell the ICO within 72 hours where the law requires it, and we tell you without undue delay if it is likely to put your rights at high risk.
10. Cookies
See the Cookie Policy. We use only strictly-necessary cookies for login/session, and our website sets none of its own: its visitor analytics (Vercel Web Analytics) are cookie-free. We set no non-essential or advertising cookies, so no cookie consent banner is required. Error/performance monitoring (Sentry) sets no cookies and does not record your session.
11. Children
The Service is not directed to children under 16 and we do not knowingly collect their data.
12. Changes
We will post changes here and, for material changes, notify account holders. Continued use after the effective date constitutes acceptance.