Questions worth answering carefully.

For most questions, ChatGPT and Claude are excellent. If you want to draft a letter, summarise a news article, or work through an unfamiliar concept, a general-purpose model is the right tool. We use them ourselves.

The trouble starts when the stakes of the answer require you to verify where it came from.

A general model answers from a memory of the public internet at the time it was trained. It can read a document you give it, but treats the document as ordinary text. Ask it about a Hadith, and it summarises the words; it does not, by default, surface the chain of narrators that determines whether the Hadith is sahih or da'if. Ask it about a regulation, and it paraphrases the text; it does not separate the binding clause from the recital, or note which jurisdiction the regulation governs.

The model is not wrong to do this. It was built for general use, and general use does not require these distinctions.

Cognoir is built for users whose work does require them. Every document you upload is parsed by its actual structure — surah and ayah for Quran, collection and book and hadith number and narrator for Hadith, case name and citation and paragraph for case law, author and section and page for academic papers. When you ask a question, the answer comes back with that structural citation attached, in the form your field expects.

The other practical difference: a general model does not remember the documents between sessions. You re-upload, re-explain, re-anchor every time. Cognoir keeps your library — a Hadith collection you assembled over months, a case file you built across a matter, a reading list for a dissertation — and queries it as one corpus.

ChatGPT is a brilliant generalist. Cognoir is a specialist for documents you cannot afford to misquote.

NotebookLM is the closest tool in shape to Cognoir — upload documents, ask questions, get citations. Google's free tier covers broad research well, and for many researchers it is the right tool.

Cognoir is different in three ways that matter for specialised work.

Cognoir parses each document by its actual structure and cites it in the form its discipline already uses. A general tool, asked about a Hadith, will summarise the words. Cognoir will name the collection, the book, the hadith number, and the chain of narration. The same precision applies to case law citations, academic citations, and the other fields Cognoir is built for.

Cognoir does not use your documents to train AI models — ours or anyone else's. EU hosting is available at signup. Workspaces are isolated per customer.

NotebookLM is better at audio and video overviews, and at the price of free for low-volume use.

If you need citation discipline a peer reviewer or a court will accept, Cognoir is built for that. If you don't, NotebookLM is a fine choice.

Your documents are yours. They are not used to train AI models — ours, or those of the providers we work with. Our contracts with those providers prohibit training on customer input, and we pass that protection through to you.

Each customer's workspace is isolated. The retrieval system queries only the documents in the workspace of the user asking the question. Your library never appears in another customer's results.

Documents are encrypted at rest and in transit. EU hosting is available at signup for users whose work requires data residency in the EU.

When you delete a document, the file is removed from storage and from the search index in the same operation. No soft deletes, no shadow copies, no retention beyond the operational window required to serve the current request. The only exception is an audit-log record that the deletion occurred — the record does not retain the document's content and exists for one year.

EU and UK users can request a full data export at any time, delivered within 24 hours. Account deletion is irreversible and removes everything except the deletion record itself.

Full details, including our subprocessor list and audit-log retention policy, are on our security page.

Fourteen document types and three input formats. Quran, Hadith, classical Islamic texts, case law, statutes, regulations, contracts, financial filings, medical literature, academic papers, technical documentation, educational materials, books, and news — plus web pages and OCR for scanned pages and images.

Each type is parsed and cited by the conventions of its discipline. Full detail on each: /source-types.

Yes, on the Team plan. Knowledge base sharing supports four roles — owner, admin, editor, viewer — with permissions that match how scholarly and legal work actually distributes across teams.

A knowledge base shared with you appears under “Shared with me” in your workspace. Your conversations, generated study materials, and progress are private to you, even within a shared knowledge base.

Single sign-on (Google, Microsoft) is included on the Team plan.

If you do not subscribe, your workspace becomes read-only for 30 days. You can still log in, read your documents, view past conversations, and export your data. You cannot upload new documents or generate new content.

After 30 days of read-only access, the workspace and its contents are deleted. We email you twice during this period — once when read-only begins, once 24 hours before deletion — so you have time to act.

Not yet. The Team plan covers most organisations with up to a few dozen users. If your needs exceed that — single-tenant deployment, custom contracts, dedicated support, on-premises hosting, or specific compliance certifications — email us. We are not actively selling at that tier, but we would like to know what would be required.

A general-purpose chatbot subscription gives you access to a model. Cognoir gives you a workspace built around documents you provide, with structural parsing for fourteen document types, retrieval that points to the passage that supports each answer, and citation formats your discipline already uses.

We charge for the work the workspace does — parsing, indexing, retrieval, citation formatting, generation — which scales with the size of your library. The trial is 14 days specifically so you can decide whether the work the workspace does is worth the price for your use case.

Not yet, as a formal program. If you are a graduate student, an independent scholar, or a non-profit researcher and the Pro plan is meaningfully out of reach, email us and we will figure something out. We would rather have you using Cognoir than not.