The answer lives in one person’s head
A new sales rep wants to know whether a customer can get extended payment terms. The answer exists somewhere: in last year’s policy, in an email from the CEO, in a contract with a similar client. It’s faster to ask Anna, who’s been here six years. Anna is on vacation. The rep guesses, and a month later it turns out the rule changed back in the spring.
A generic AI chatbot would make this worse. It would answer confidently even if no rule existed at all. MEMUS works the other way around: it answers only when it can show you the document and the clause, and when it can’t, it tells you.
How MEMUS connects to the rest of PRANUS
MEMUS is the shared memory of the whole system. Other services write to it, and everyone with the right access can get answers from it.
- Documents. A policy approved in DOCUS document management goes into the knowledge base with one click. From then on, people and agents answer based on it, with a link to the clause.
- AI agents. AI agents built in AGENTUS check MEMUS before they say anything to a customer. MEMUS replies in parts: “found, with a quote,” “partially,” or “not found.”
- Questions for humans. When there’s no answer, the question goes to experts on the Questions desk in the CHATUS team messenger. Once approved, the expert’s reply becomes a saved answer, so the agent can handle it alone next time.
- Org chart. Access to knowledge bases follows your org chart in TEAMUS. When someone moves to another department, their access changes with them.
- Personal assistant. Assistentus searches MEMUS on behalf of a specific person, with that person’s permissions, by voice or text.
What the AI does
It understands document structure. Every passage remembers its section and page, so a citation points to the right clause, not to “somewhere in a 300-page PDF.” You can define a template for each document type: what to extract and what to call it.
It searches several ways at once. MEMUS picks a fast, standard, or deep mode depending on the question. Frequent questions get pre-approved answers instantly, without calling a large model.
It declines when it isn’t sure. The refusal check runs on several signals before the model is called. It’s cheaper than generating an answer, and it’s auditable.
It learns from your experts. An expert’s reply becomes a correction, and outdated knowledge gets an end date and stops showing up. Agents answer according to the rule that applied on the relevant date.
It resists manipulation. Instructions hidden in third-party documents are never treated as commands.
Getting started
Loading your knowledge is part of the standard PRANUS rollout. The AI implementation agent creates knowledge bases by department, takes your files, shows you how it parsed them before loading, connects the base to your AI agents, and checks it with control questions. From your chat history, it drafts playbooks for sales and support: services and pricing, common questions, objections, and what not to promise. Learn how your chat history becomes a knowledge base and how implementation in 14 days works.
If company knowledge must never leave your premises, PRANUS can be deployed on your own server or in an isolated environment.
Who it’s for
- Owners and managers tired of being the only source of answers.
- Sales and support teams where a new hire has to give customers the right answer from day one.
- Legal, HR, and finance teams with policies, contracts, and procedures that keep changing.
- Companies rolling out AI agents that need those agents to answer from documents, not from imagination.
Want to see MEMUS cite its sources — and decline when it should? Book a demo and we’ll show you on a live system.