AI Transparency

Medipher sends your text to large language models through the Lovable AI Gateway. The gateway currently routes to models from Google (Gemini) and OpenAI (GPT). We may switch or update models over time when a better option improves quality, safety or reliability.

We never expose your text to more providers than needed to produce your explanation.

1. You paste text into the Explain tool.
2. Your browser sends that text to our server.
3. Our server combines it with a strict, source-bounded system prompt (rules about what the AI may and may not say) and forwards it to the model.
4. The model returns a structured, plain-English response.
5. Your browser renders the response. We do not save the pasted text or the response.

AI models are statistical — they predict what words are likely to come next. This means they can:

• misread part of a document, especially when text is fragmented or handwritten;
• omit information that is present in the original;
• occasionally invent details ("hallucinate");
• use language that sounds more confident than the underlying evidence supports.

For these reasons, Medipher is designed to be a companion to your original document and your clinician — never a replacement.

• Source-bounded prompting. The model is instructed to only explain what the document actually says, and to say "the document does not state this clearly" rather than infer missing information.
• No diagnosis or reassurance. The model is told not to diagnose, reassure, or say things like "everything is fine".
• Structured output. We ask the model for a fixed shape (summary, at-a-glance, questions to ask, etc.) so it is less likely to drift into freeform speculation.
• User feedback loop. Every explanation invites feedback, which we use to improve prompts and templates.

We ask AI providers not to train their models on your inputs. However, we cannot currently guarantee provider-side settings like Zero Data Retention from within our stack, because those typically require a direct enterprise contract with the model provider. We think transparency here matters more than marketing, so we are telling you plainly what is and is not currently true.

Please avoid pasting information you would not be comfortable sharing with a third-party service. If you need clinical-grade guarantees for handling patient data, this is not the right tool for that use case today.

If anything Medipher tells you feels off, or contradicts what your clinician has said, trust your clinician and the original document. Ask your GP, doctor, nurse or pharmacist. See our Medical Disclaimer.