Everyone sees only theirs.
Two people ask the same question and get two truthful answers, each scoped to the documents that person may see. Access is read from your sources and enforced inside every query, never bolted onto a surface.
Group filter inside every query, no surface above it
Enforced inside retrieval itself.
Group filters live inside the search, in the WHERE clause of every retrieval and read the assistant makes, on the web console, on every messaging channel, on the API. There is no unfiltered path to the index, so there is no surface to misconfigure and no channel that "forgot" to check.
That is why a guest on WhatsApp and an administrator on the web console are the same problem, solved once: each query carries its asker, and the asker carries their groups.
Mirrored from the sharing your drives and systems already declare
Groups, assign people, point sources at them, done
Per-source, per-folder, per-file, with per-file overrides when a rule is too blunt
Access changes apply immediately, without re-indexing content
Channel guests get nothing until a group is explicitly opened
It carries your name, and your rules.
Your team signs into "<Your company> Assistant", your logo on the sign-in page, the header and the chat widget. Accounts are invitation-based; administrators never set or see a password.
The assistant answers in the asker's language, remembers each person's conversations, and is tuned per deployment: which sources exist, which groups see them, how channels treat guests and group chats.
Company name and logo across sign-in, console and widget
People recognised across web and every messaging channel
Administrators see exactly who can read what, per person, per file
Console in English, Greek, German, French, Spanish; answers follow the asker
Every answer and every permission change on a hash-chained log
Access control for AI answers, explained
Can AI respect our document permissions?
Yes, if the filter lives in the right place. In Naxis Assistant, access control sits inside the search itself, on every surface alike, so there is no unfiltered path to your documents and no channel that "forgot" to check. The full mechanism is written up for your reviewers.
Where do the permissions come from?
From you, in the vocabulary you already use: groups. Sources can mirror the sharing your drives and systems declare, administrators assign people to groups, and a per-file access map shows exactly who can read what.
What happens when someone loses access?
Access changes apply immediately, without re-indexing anything. The next question is already scoped to the new reality.
Can two people really get different answers to the same question?
Yes, and both answers are true. Each person is answered only from documents their groups may read, so scope differs while honesty does not. The sales lead sees the deal note; support sees the tickets; neither sees the other's material in their answer.
Can we put our own name and logo on it?
Every deployment carries the client's name: your logo on the sign-in page, the header and the chat widget, your company name in the assistant's identity, your language defaults. Accounts are invitation-based, administrators never set or see a password.
Is there a record of who asked what?
Yes, a complete one. Every answer and every permission change lands on an audit record that can prove it has not been tampered with. Deployments that must not retain content can log metadata only; the details are on the security page.
Ask as two people. Compare.
The demo ships with real roles inside, ask the same question from two of them and watch the scope change.