Marvin is the AI assistant built into NIZU Cloud WorkSpace. It lets staff, clients and leads ask plain-language questions about their own business data — for example “tell me about invoice #586” or “which clients have gone dormant?” — get a clear conversational answer, and increasingly get work done: creating a project, drafting an invoice, adding a task. It also answers visitors on your public website through an embeddable chat widget.
Most AI assistants hand your data to a language model and let the model decide what to do. Marvin does the opposite. NIZU identifies who is asking, checks that they are allowed to see the requested record, and fetches that record live from the workspace database. Only then is a language model asked to phrase the answer. The model never authorizes access and never fetches data on its own.
Marvin also ships a white-label chat widget you drop onto any website with a single
<script> tag. Visitors are anonymous and untrusted, so that widget runs a
separate, sandboxed pipeline: it answers only from public knowledge you
configure — a knowledge-base text, your published help articles, and what it learns
from your own site's sitemap.xml and llms.txt. It can never reach
client records, invoices or any private data. When it cannot help, it hands the conversation
to a human on your team.
Staff use Marvin in the main app; clients and leads reach it through the portal; website visitors meet it through the widget. Each person sees only their own data and their own conversation history, while administrators can review conversations for quality.
Marvin turns NIZU into something you can simply talk to. Instead of navigating menus and filters, you ask a question — or give an instruction — and it happens, without exposing your data to a system that should not see it.