AI transparency

This notice explains how Wispu uses artificial intelligence, what we do with the text you give it, and who is accountable for what the system produces. It covers our duties under Regulation (EU) 2024/1689 (the EU AI Act) and the transparency rules of the other regimes we work under. It sits beside the Privacy Policy, which governs personal data, and the Terms of Service, which govern the contract.

1. What Wispu does with AI

Wispu keeps a structured record of your brand — your voice, your point of view, your customers, your offers and your published work — and uses it to draft content. In practice the system does four things:

• It reads what you give it. Notes, voice notes, articles, saved posts, your public website and the performance of content you already published.
• It structures that into a Brand Context. The Context is data about your business, written in your words, that you can read and correct.
• It drafts. Posts, articles, emails and adaptations for each channel, generated by a general-purpose AI model working from that Context.
• It waits for you. Nothing leaves Wispu until a person approves it. Publishing is a human action.

2. The models we use, and where your text goes

Wispu does not train or host its own foundation models. Its AI runs on commercial API accounts with Anthropic (Claude) and OpenAI, and on other models we combine per task through the Vercel AI Gateway. Wispu chooses which model does each task, and that choice can change.

The current list, with each provider’s role and location, is on the Sub-processors page and is the version that governs.

No one trains models on your content. No model or provider uses your content for training. Wispu sends every request with the Vercel AI Gateway setting that forbids training (disallowPromptTraining). On top of that, the commercial API terms of Anthropic and OpenAI exclude training on API data by default. Your Brand Context, your drafts and your history are never sold and never used to train public models.

How long providers keep your text. Anthropic and OpenAI may keep API inputs and outputs for up to 30 days, only to monitor abuse and misuse and for safety. Then they delete them. They keep them longer only when an abuse investigation or a legal obligation requires it. The reason is simple: these providers check that no one uses their models for prohibited purposes, and they need a short window to do it. Wispu does not use those copies. Requests to other models that we pay for with the Vercel AI Gateway’s own credits run with zero data retention.

AI calls. The voice agent is an optional feature. The WhatsApp calling provider does not record audio by default and keeps only call metadata: the numbers, the duration, the cost and the reason the call ended. The real-time voice model is OpenAI Realtime, under the same rules: it does not train on your calls and may keep them for up to 30 days for abuse monitoring. Wispu keeps call transcripts and summaries for the life of your account, because the full history is your Brand’s knowledge base, and Wispu’s value depends on it. They are deleted when you delete them or close your account.

3. Our role under the EU AI Act

Wispu is an AI system placed on the market by Kyba Ventures LLC under its own name, so we act as a provider of that system. We are also a deployer of the general-purpose AI models listed above. You, as a customer, are a deployer of Wispu when you use it in your own business.

Wispu is not a high-risk AI system. It is not used for any purpose in Annex III of the AI Act: it does not select candidates or manage workers, it does not decide access to education, credit, insurance, benefits or essential services, and it is not used in law enforcement, migration or the administration of justice. Wispu writes marketing content and waits for a person to approve it.

That places Wispu in the transparency tier of the Act. The obligations that follow are the ones in Article 50, and we meet them as set out below.

4. Transparency: what we tell you, and what you must tell your audience

You are interacting with an AI system

Where you talk to Wispu — the assistant, a drafting action, a suggestion — you are interacting with an AI system, not a person. The interface says so at the point of use.

Wispu’s output is artificially generated

Text, images and audio that Wispu produces are artificially generated or manipulated content. Inside the product we mark drafts as generated and keep the record of which model produced them. Where the format allows a machine-readable marker, we apply it.

Your duty when you publish

When you publish AI-generated text to inform the public on matters of public interest, the AI Act requires disclosure that the content is artificially generated. That duty is yours, because you decide what ships. It does not apply where the content went through human review and a natural or legal person holds editorial responsibility for it — which is the workflow Wispu is built around, since nothing publishes without your approval.

Wispu does not remove, weaken or hide any disclosure you add, and does not strip provenance metadata from media you upload.

5. What Wispu does not do, and what you must not do with it

Article 5 of the AI Act prohibits certain practices outright. Wispu does not perform any of them. Specifically, Wispu does not:

• use subliminal, manipulative or deceptive techniques to distort behaviour;
• exploit the vulnerabilities of a person because of their age, disability or social or economic situation;
• score people socially, or infer or categorise them by biometric data;
• infer emotions in a workplace or an educational setting;
• build facial-recognition databases, or perform remote biometric identification.

You must not use Wispu for those purposes either, and the Terms of Service make that a condition of use. You must also not use it to impersonate a real person or organisation, to generate content that presents fabricated events as real, or to produce anything unlawful in the market you publish into.

6. Human oversight and accountability

Every publishing path in Wispu ends at a person. Drafts are drafts until someone with access approves them, and the approval is recorded with the account that gave it. Connected AI assistants — through MCP or our plugins — can propose and prepare, and they cannot publish on their own.

This matters for responsibility as much as for compliance: what goes out under your name is your editorial decision, and the product is built so that decision cannot be skipped by accident.

7. Accuracy, limits and what the output is not

Generative models produce plausible text, not verified fact. Wispu output can be wrong, out of date, or confidently invented, including names, numbers, quotations, citations and claims about your own business. Read it before you publish it.

Wispu output is not legal, medical, financial, tax or other professional advice, and must not be presented as such. Where a claim you publish has to be substantiated — a result, a statistic, a comparison with a competitor — the evidence is your responsibility.

Model behaviour changes when providers update their models. We may change the model behind a feature to keep quality, cost or availability acceptable, and the current providers are always the ones listed on the Sub-processors page.

8. Personal data and automated decisions

How we handle personal data, on what legal basis, for how long, and how we move it between countries is set out in the Privacy Policy. Two points belong here as well:

• No solely automated decisions with legal effect. Wispu makes no decision about a person that produces legal effects or similarly significantly affects them within the meaning of Article 22 GDPR. It drafts content for a human to approve.
• Do not paste what you would not send. Prompts and uploaded material are sent to the model providers above. Do not put special-category data, credentials, or other people’s confidential information into a prompt.

9. AI literacy

Article 4 of the AI Act asks providers and deployers to make sure the people who operate their AI systems understand them well enough to use them responsibly. Our team is briefed on what the models can and cannot do, on the review step that stands between a draft and a publication, and on the practices this notice prohibits. The product itself carries the same message at the point of use rather than only in this document.

10. Who owns the output

As between you and Wispu, your inputs and the content Wispu generates for you are yours, on the terms set out in the Terms of Service. Note that the copyright status of AI-generated material differs by country, and that in several jurisdictions purely machine-generated output may not attract copyright at all. Your own edits and your own material are unaffected.

11. When these rules apply

The AI Act entered into force on 1 August 2024 and applies in stages: the prohibited practices and the AI-literacy duty since 2 February 2025, the general-purpose AI model rules since 2 August 2025, and the transparency obligations in Article 50 from 2 August 2026. We already operate to the Article 50 standard described here rather than waiting for the date.

12. Questions, complaints and changes

Ask us anything about how the system works, or tell us it got something wrong, at team@wispu.app. If you are in the EU you may also complain to your national market-surveillance authority, and about personal data to your data-protection authority.

We update this notice when the system changes or the rules do, and the date at the top is the version in force.