Your company’s knowledge,
answered.
Naxis Assistant answers your team’s questions from your own documents — the handbook, the policies, the contracts. Every answer shows the exact passage it came from, and when the answer isn’t there, it says so instead of guessing.
✓ Cites its sources — or refuses ✓ Respects who may see what ✓ Single-tenant: yours alone
Production deployments are allowed Monday through Thursday. Fridays are reserved for critical hotfixes with engineering-lead approval.[1]
1 Engineering Handbook § DeploymentsBuilt for companies where a wrong answer costs money
Generic chatbots improvise. Naxis is engineered around three guarantees that make it safe to put in front of your whole team.
Cited — or silent
Every answer carries numbered citations to the exact passages behind it, so anyone can verify it in one click. If the answer isn’t in the documents, Naxis refuses instead of inventing one. There is no in-between.
Permission-aware
Access rules are enforced beneath the AI, not around it. A person only ever gets answers from documents their groups allow — finance figures for finance, HR files for HR. The assistant can’t be talked out of it.
Yours alone
Each client gets a dedicated, single-tenant deployment — one system, one database, no shared infrastructure. Document embeddings are computed inside your deployment and never leave it. Your data trains nothing.
Live in three steps
Connect your knowledge
Point Naxis at where your documents already live — Google Drive, SharePoint, Dropbox, S3, your website — or upload files directly. No migration, no re-organising.
Set who sees what
Create groups like finance, hr or engineering and assign them to sources and people. Naxis keeps everything in sync automatically — new and changed files are picked up on schedule.
Ask in plain language
Your team asks questions the way they’d ask a colleague, and gets answers with the source right underneath. Follow-up questions work — it remembers the thread.
Meets your documents where they are
Many sources, one knowledge base. Each source syncs incrementally — only new, changed or removed files are processed — and a temporary outage never touches what’s already indexed.
Same question. Different clearance.
Permissions are checked at retrieval, below the AI — so the assistant never even sees a document the asker isn’t allowed to read.
“What did we spend on travel in Q2?”
Source: Expense Report Q2 · Finance
“What did we spend on travel in Q2?”
The finance report exists — but Alex isn’t permitted to see it, so for him it doesn’t.
No prompt trick, jailbreak or clever phrasing can cross a group boundary — the restricted documents are excluded before the AI ever runs.
Ready for the questions your DPO will ask
Naxis was designed for European data-protection reality — not retrofitted for it. The compliance tooling ships in the product, not in a PDF.
Subject access & erasure
Export everything held about a person, or erase it — including their conversations — with built-in, auditable tooling. Users can export and delete their own data themselves.
Generated processing manifest
A processing-record manifest is generated from the live deployment configuration — data categories, retention, sub-processors, models — never hand-written, never out of date.
Hash-chained audit log
Every question, retrieval and admin action lands in an append-only audit log secured by a hash chain. One command verifies that nothing was ever altered.
Transparent by design
Every surface discloses that answers are AI-generated, every answer is stamped with the model that produced it, and high-risk uses are contractually excluded.
Choose how private “private” means
Both editions are single-tenant and identical in features — they differ only in where the language model runs.
Naxis Hosted
Managed by us, dedicated to you.
- ✓ Dedicated single-tenant deployment, managed and updated by Naxis Technologies
- ✓ Answers generated via the Anthropic API under zero-data-retention terms
- ✓ Document embeddings computed inside your deployment — your files never leave it
- ✓ Exactly one disclosed sub-processor, on the generated manifest
Naxis Residency
Everything inside your infrastructure.
- ✓ Runs entirely on your servers or private cloud — nothing crosses your boundary
- ✓ Open-weights language model served locally; embeddings local as always
- ✓ Zero sub-processors — the manifest section reads “None”
- ✓ For regulated industries and strict data-residency requirements
The questions everyone asks first
Is our data used to train AI models?
No. Your deployment is single-tenant, embeddings are computed locally inside it, and in the Hosted edition model calls run under zero-data-retention terms — nothing is stored or used for training. In the Residency edition, nothing leaves your infrastructure at all.
What happens when the answer isn’t in our documents?
Naxis says “I don’t have that in the knowledge base.” It never fills gaps with plausible-sounding text. This cite-or-refuse rule is enforced in the core of the system, not by a tunable confidence threshold.
Can employees see documents they shouldn’t?
No. Group permissions are applied when documents are retrieved from the database — before the AI runs. A document outside someone’s groups is invisible to their assistant, and no phrasing of a question changes that.
Which file formats are supported?
Word documents, text-layer PDFs, Markdown and plain text, plus Google Docs, Sheets and Slides via the Drive connector, and public web pages. Scanned image-only PDFs (requiring OCR) are on the roadmap.
How fresh are the answers when documents change?
Every source syncs on a schedule (and on demand). Syncs are incremental: only new, changed or removed files are re-processed, so an edit in your Drive shows up in answers after the next sync — typically minutes, not days.
How long does setup take?
A deployment is typically live the same day. Connecting a source is a short guided form (the in-product help walks through each provider step by step), and users are invited by email with one-time activation links — no passwords to distribute.
See it answer on your documents
A demo takes 30 minutes. Bring your handbook, your policies, your toughest question — and watch every answer arrive with its source attached.