Search your documents with AI and check every answer

DOCNUS turns a folder of contracts, policies, reports and letters into a knowledge base your team can question in plain English. You upload documents, ask a question, and get an answer that names the file and page it relied on. Click the citation and the source opens at that page, so you can read the original wording before you act on it.

How citations work

An AI answer is only useful if you can check it. Every DOCNUS answer cites the document and page each point came from. Clicking a citation opens the source document at that page.

This changes how you use the answer. Instead of trusting a summary, you treat it as a pointer: "the notice period is 90 days, see Supplier Agreement, page 7". You then read page 7 yourself. If the answer has misread something, you will see it straight away.

Answers are written by Anthropic Claude models, based on the passages DOCNUS retrieves from your own library.

What you can upload

DOCNUS accepts:

• PDF, including scanned PDFs
• Word documents (.docx and .doc), which are converted to PDF
• Plain text (.txt)

Scanned PDFs are read with OCR (Docling with RapidOCR), and that includes tables. An old signed contract that exists only as a scan, or a table of figures in a scanned report, becomes searchable in the same way as a native PDF.

Hybrid retrieval: meaning plus keywords

Plain keyword search finds documents that contain the words you typed. It struggles when the document uses different words for the same idea. Pure meaning-based (vector) search has the opposite problem: it understands paraphrase but can blur exact strings such as reference numbers or defined terms.

DOCNUS runs both and combines them:

1. Vector search finds passages that mean the same thing as your question. Embeddings are produced with OpenAI models, and each organisation has its own index namespace.
2. Keyword search (BM25-style full-text search) finds passages containing your exact terms.
3. The two result sets are fused, then reranked so the most relevant passages come first.
4. If the first pass looks weak, DOCNUS searches again before answering.

A concrete example

Say you ask: "Which suppliers can terminate for convenience?"

A keyword search for "terminate for convenience" only finds contracts that use that phrase. It misses a clause that reads "Either party may end this Agreement at any time on 90 days' written notice without giving a reason." That clause answers your question, but shares almost none of its words.

Meaning-based search recognises that clause as relevant. Keyword search, meanwhile, makes sure that if you ask about "Schedule 4" or invoice "INV-20417", the passages containing that exact text are not lost among loosely related ones. Combining the two gives you both kinds of match, and the citation lets you confirm which clause the answer came from.

Beyond search

The same library feeds other parts of DOCNUS:

• Knowledge graph. People, organisations, agreements, locations and other entities are extracted from your documents. Each entity has a profile listing the documents and passages that mention it.
• Obligations. Invoice due dates and payment terms on every plan. On Professional and above, renewals, notice periods, termination rights and other obligations are read from contracts, with CSV export.
• Privacy assessment. On Professional and Enterprise, each document can be checked for personal data against seven privacy laws. See privacy assessment.

What it does not do

We would rather you know the limits before you sign up.

• File types are limited. Only PDF, Word and plain text. Spreadsheets, slide decks, emails, images, audio and web pages are not supported. General assistants accept far more, for example ChatGPT's file uploads and NotebookLM's supported sources.
• No integrations. DOCNUS does not connect to Microsoft 365, SharePoint, Google Drive or email. You upload files directly.
• No public API. There are no API keys for programmatic access.
• It works from your library. Answers are built from passages retrieved from the documents you have uploaded, so if a document is not in DOCNUS, it cannot be cited.
• You still need to check. Citations make checking quick, but an answer is not a substitute for reading the source on decisions that matter.

If you are weighing DOCNUS against a general assistant, our comparisons with ChatGPT, NotebookLM and Microsoft Copilot set out where each one is the better fit.

Plan limits

Prices are in US dollars per month, excluding tax. Every account starts on Free and can upgrade afterwards.

Pricing is per team, not per user. Full details are on the pricing section.

Where your documents go

Each organisation's data is isolated, with its own search index namespace. Documents are encrypted in transit. Answers and embeddings are produced by subprocessors that process data in the US, listed in our privacy policy. Deleting a document removes its file, text, search index entries and graph data. Read more on our security page.

Using it for compliance work

Search is often the slowest part of compliance tasks. Under the UK GDPR, a subject access response must cover what an organisation can find through a "reasonable and proportionate search", and being able to ask "which documents mention this person?" and get cited pages helps with that step. Our DSAR guide explains where DOCNUS helps and where you will need other tools.