How we use
AI models.

01 · Model providers

DraftGuard uses third-party language models through OpenRouter to analyze and rewrite text on a per-request basis. The model provider may change as we improve quality, latency, cost, or reliability. We do not operate our own foundation model.

02 · Draft handling

Draft text is sent only to produce the result you requested. We do not store draft text in our application database, and we do not use your draft to train a model.

Browser storage may keep your active draft locally for the current tab/session so the workspace can continue after navigation.

03 · AI limitations

AI output can be incomplete, inaccurate, overconfident, or mismatched to your context. You must review all rewrites before using them. DraftGuard does not guarantee a detector score, academic outcome, publication decision, admissions result, or factual correctness.

04 · Sources and personal detail

DraftGuard can tell you that a claim needs support and can name the kind of source that would help. It must not invent citations, quotes, statistics, case studies, interviews, or personal experiences. Add only facts you can verify.

05 · Safety behavior

We use product rules and prompts to discourage detector bypass, fabrication, plagiarism, and academic dishonesty. If you find output that violates this policy, send details to support@whispen.io.