# What Is Company Memory? Why Every AI Tool Will Need One

> What company memory is, how it differs from search, and why AI tools like Claude and ChatGPT need permissions on every fact.

Louvain, 2026-11-12. Source: https://louvain.ai/blog/what-is-company-memory

# What Is Company Memory? Why Every AI Tool Will Need One

**Company memory is a single, permissioned record of what your company knows: what was agreed, decided, promised and asked, drawn from your email, documents, meetings, chats and calendar.** It lets people and AI tools ask a question and get an answer traced back to the source. The important word is *permissioned*. Every AI tool your team uses will soon read company knowledge. Something has to decide what each of them is allowed to see.

## TL;DR

*   Company memory turns scattered email, files, meetings and chats into checked facts, each linked to where it came from.
    
*   It is different from search. Search finds documents. Memory answers the question and shows its evidence.
    
*   AI assistants like Claude and ChatGPT now connect to company data. Without fact-level permissions, they can surface things a person should never see.
    
*   Good company memory enforces permissions on every fact, records when something was true, and says plainly when it doesn't know.
    

## What company memory actually is

Most of what a company knows is never written down in one place. A price is agreed on a call. A deadline moves in an email thread. A client asks for something extra in Slack. The decision lives in someone's head, or in a message nobody can find three weeks later.

Company memory collects that knowledge as it happens. It reads the places work already takes place and records the facts inside them: who agreed to what, by when, and where it was said. Each fact keeps a link to its source, so anyone can check it.

The result is something you can ask. "What did we agree with this client about the March deliverable?" "Who owes us the signed contract?" "What changed on this project last week?" The answer comes back with the email, document or meeting it came from.

## How it differs from search, wikis and notes

**Search** returns a list of documents and leaves you to read them. It doesn't know that the second email overruled the first.

**Wikis and intranets** only hold what someone remembered to write up. They go stale the day after they are updated.

**Meeting note-takers** capture one meeting well. They don't connect that meeting to the email that followed or the contract signed a month earlier.

Company memory works across all of them. It keeps track of which statement is the latest, which was a draft, and which was replaced. When sources disagree, it should say so rather than pick one quietly.

## Why AI tools make permissions the hard part

AI assistants are now plugging straight into company data. Claude, ChatGPT and others can connect to your email, drives and chat through connectors. That is useful. It is also where most companies get stuck.

The problem is oversharing. A folder shared too widely years ago, an HR thread, a client's confidential figures. A person browsing might never stumble on them. An AI tool asked a broad question can pull them into an answer in seconds.

Document-level permissions are not enough. One document can contain facts that different people should see differently. A board pack might be fine for partners and not for the wider team. A client thread might be fine for the account team and nobody else.

That is why the next layer of company software is not another assistant. It is the layer that decides, fact by fact, what each reader may see, whether that reader is a person or an AI agent.

## What good company memory should do

1.  **Enforce permissions on every fact.** Each fact keeps the permissions of the place it was said. An answer is built only from what that reader can see.
    
2.  **Treat AI agents as users.** An assistant acting for a person should read exactly what that person can read, never more.
    
3.  **Cite everything.** Every figure, date and name in an answer should link to its source.
    
4.  **Know when things were true.** Prices, owners and deadlines change. Memory should record when each fact held, not just the latest version.
    
5.  **Say when it doesn't know.** A confident wrong answer is worse than "the sources don't say".
    
6.  **Track commitments.** Promises made in email and meetings should be visible until there is evidence they were kept.
    

## Who it is for

Large enterprises have had expensive search and knowledge tools for years. Growing teams have not. Yet a 20-person consultancy, a property management company or a fast-moving startup loses just as much to "where did we agree that?" as a bank does.

These teams also tend to adopt AI tools faster, with less IT oversight. That makes a permission layer more important, not less.

## How Louvain approaches it

[Louvain](https://louvain.ai) is permissioned company memory for growing teams. It connects your email, documents, meetings, chats and calendar, and turns them into checked facts, each cited to its source. Permissions are enforced on every individual fact. Teams get a daily brief, see where every project stands, and track what is owed, without chasing people for updates.

Louvain also works as a connector inside Claude, ChatGPT and other AI tools, so those assistants get safe, permission-aware access to company knowledge. The aim is simple: the first thing you open every morning when you get to work.

## FAQ

### Is company memory the same as enterprise search?

No. Enterprise search finds documents. Company memory answers the question, shows its sources, and tracks how facts change over time.

### Can't I just connect ChatGPT or Claude to my drive?

You can, but the assistant will see whatever the connector can see. Without fact-level permissions, people may get answers built from material they were never meant to read.

### Does company memory replace my existing tools?

No. It reads the tools you already use, such as email, documents, chat and meetings, and sits alongside them.

### Who should own it inside a company?

Usually operations or IT sets it up, but everyone uses it. The point is that nobody has to maintain it by hand.
