It answers from your documents. And cites the page.
nerix.ai is an AI assistant for the people and businesses that live in contracts. Ask a question, get the answer with the exact page it came from — and when your documents don't hold the answer, it says so instead of inventing one.
The problem
The answer exists. It's on page 11 of a PDF nobody can find.
Folders no search can reach
Contracts live in shared drives and inboxes, filed once and found never. The knowledge exists; the retrieval doesn't.
An afternoon per question
"What did we agree to?" means opening files one by one until something looks right. The answer costs hours and still arrives unverified.
Confident AI, citing nothing
Generic assistants answer fluently about documents they have never read. Sounding right is not the same as being right — and nothing links the claim to a page.
The toolkit
One library. A growing set of tools.
Everything works from the same indexed library, so each tool answers from what you already filed.
How it works
Three steps, and the third is the one nobody else does.
Upload what you already have
Drag in a folder of contracts. Every file is indexed, summarised and filed against the client, supplier or colleague it belongs to.
Ask in the language you think in
English or Arabic, across the whole library at once. Keyword and AI search run side by side so you can see what each one found.
Get the page, or get a straight no
Every claim resolves to a document and a page you can open. When your documents don't hold the answer, it says so instead of inventing one.
Cited answers
No citation, no claim.
Every statement resolves to a document and a page you can open. Turn the grounding switch on and the assistant must cite a passage for every claim — or tell you it can't answer, rather than filling the gap from a model's training data.
Most AI tools can't say "I don't know". This one is built to.
Library
A library that has actually read your files.
Upload anything — PDFs, spreadsheets, scans, code. Every document gets an AI summary and topics, a visible index status, and a place in a folder tree built around your clients, staff and suppliers.
Dual search
Two searches, side by side. Not blended — on purpose.
Keyword search finds exact words, even in files still being indexed. AI search finds meaning, even in different words. Blend them and you can't tell why a result ranked; keep them apart and you can trust both.
The Sheet
Drafts with a history you can rewind.
Documents, code, HTML, spreadsheets and diagrams land in the Sheet. Every edit is a version, and restoring an old version never erases a newer one — the history only grows. Export when it's ready.
Agents
Not one model. A team with a plan.
A Coordinator breaks your question down and delegates to specialists — each with a declared list of tools and a model class matched to its job. You can watch who did what, and what it cost.
Control
Your policy, enforced by the server — not by a checkbox.
Write rules once and they apply to every conversation. Skills are named procedures the assistant reaches for when a question matches. And any tool that would send data outside your organisation asks first, by default.
Providers & residency
Choose the model. We print the caveats.
Mistral, hosted in the EU, is the default. OpenAI is there when you want it. DeepSeek is the cheapest — and the picker itself tells you its first-party API processes data in China. Honesty you can hold us to.
Your data stays in the region you choose
Bilingual
English and Arabic. Both native, neither an afterthought.
Ask in Arabic about an Arabic contract and get an Arabic answer — with the same page-level citations. The interface, the library, the search and the answers all work right-to-left, and a mixed English–Arabic library is normal, not a special case.
Pricing
Start free. Pay when your team does more.
For an individual: one seat and 200 credits a month, free. For teams and companies: from €49, with seats, roles and storage to match.
Instead of testimonials
No testimonials yet. Facts you can check instead.
Your documents never train a model. It isn't a toggle you could miss — the setting reads "Never" and cannot be changed.
The audit log is append-only. Not even an owner can edit or delete an entry.
Your credit balance is the sum of an append-only ledger — nothing is ever edited or removed from it.
DeepSeek's China-processing caution is printed inside the product, on the picker itself.
Erasure takes effect within 30 days — including in backups.
With grounding on, the server refuses ungrounded answers. The refusal is a feature.
When we have customers who'll vouch for us in public, they'll be here — named, with permission. Until then: claims you can verify.
Questions people actually ask.
What happens when my documents don't contain the answer?
With "Answer only from my documents" on, the assistant must cite a passage for every claim — and says it can't answer rather than filling the gap from a model's training data. The refusal is deliberate: a confident guess about a contract is worse than no answer.
How do citations work?
Every claim resolves to a specific document and page — "page 11 of 34" — with the quoted passage, so you can open the source and check it yourself.
Is my data used to train AI models?
No. Never, by us or by any provider. It is not a setting you could misconfigure — in the product it reads "Never" and cannot be changed.
Where is my data stored?
In the region your organisation selects: European Union (Frankfurt), United Kingdom (London) or United States (Virginia). Changing region moves your documents and embeddings, and the move is recorded in the audit log.
Which AI providers can I use?
Mistral (EU-hosted, the default), OpenAI, and DeepSeek — with DeepSeek's cautions printed in the product, including that its first-party API processes data in China. Anthropic and Google are coming soon. Administrators can restrict the picker to providers hosted in your data region.
Does it work in Arabic?
Fully. Right-to-left interface, Arabic documents, Arabic search and Arabic answers with the same page-level citations. A mixed English–Arabic library is normal.
What does a message cost?
A typical message is estimated at about 12 credits, and the exact cost is settled after each message — never before. Indexing a document and exporting a Sheet also spend credits. Every entry lands in an append-only credit history.
How do I start?
Open the app and sign in. The free plan is one seat and 200 credits a month — enough to index a stack of contracts and start asking.
Ask your documents something hard.
Sign in, upload a contract, and ask the question that usually costs an afternoon.