How research is handled here.

The PRQT corpus includes protected, licensed research — published works and, as it is added, current manuscript material. Each Paraqyt account receives only the research access it is authorized to use.

When Paraqyt’s research process decides a work is relevant, it asks to open it. That request is checked on our servers against the account’s current, active access before anything is read. The assistant cannot grant itself access, and a request for a work outside your access is refused without revealing more than the public site already describes.

Material you aren’t authorized to read is not retrieved into your conversation — it is never loaded and then hidden.

For questions about PRQT, Paraqyt is designed to ground its answers in research material it actually opened in your conversation, and to return citations you can inspect.

Citations to PRQT works are validated against what was opened, and identify the work — published paper, book or current manuscript — so its provenance stays visible. When public-web research is used, web citations are listed separately from PRQT citations.

Grounding reduces error; it does not eliminate it. Treat answers as a starting point for your own reading, and verify anything important against the cited source.

Paraqyt currently uses xAI models as its AI reasoning provider. PRQT Labs selected xAI in part because of its emphasis on technical and STEM-oriented reasoning.

That is a considered choice for this product, not a claim that any one provider is best at everything. If the provider changes, this page will change with it.

The product is designed around a small number of deliberate controls on how research content reaches the AI provider:

Paraqyt can use public-web search when it helps — for example, to set PRQT beside a conventional treatment. It decides whether web research is useful; our servers decide whether it is permitted.

Web search is offered only while a conversation carries no protected PRQT content, no private attachment content, and nothing derived from either. Once protected or private material is in play, web search is not available. It is never used to retrieve the corpus or your attachments.

You retain ownership of the research and technical materials you upload — papers, calculations, drafts, notes or internal research. PRQT Labs does not acquire ownership of them.

Paraqyt may store, process and transmit relevant attachment content to service providers only as necessary to provide the research functionality you ask for, under the privacy controls described above. When Paraqyt opens an attachment to reason over it, relevant content from that attachment is sent to our AI provider under those controls.

These boundaries matter as much to a private R&D team working with proprietary material as to an academic researcher with an unpublished draft: once attachment content is in a conversation, public-web search is no longer offered there, and that content reaches the AI provider only under the controls above.

Systematic scraping, automated extraction, reconstruction of protected works, and attempts to circumvent access restrictions are prohibited.

The platform includes technical controls intended to detect and restrict abusive patterns. Abuse can lead to restriction or permanent termination of access.

Self-service account deletion is a required pre-beta capability being built into Paraqyt. It is not live yet, and this page will say so plainly when it is. Once live, it will work as follows: