The short answer
Most companies already have their knowledge written down. It is spread across Slack threads, Google Drive folders, Notion pages and people's heads, and no single place holds the current version of all of it.
A company brain is the layer that brings it together. It reads what the company writes in the tools it already uses, keeps that up to date, and answers questions from it, with the source of every claim. The same brain serves a person asking in a chat and an AI agent asking over an API.
It is not one more place to write documentation by hand. Most of what it holds still lives in the tool where it was written. The brain keeps a current, searchable copy and knows where each part came from.
Where the term comes from
The idea is older than the name. Tiago Forte's Building a Second Brain made it familiar for one person's notes. In 2026 it was applied to whole companies: Andrej Karpathy described wikis that an AI keeps up to date as it works, and Y Combinator listed the company brain as a category in its Request for Startups.
What is new is the pressure behind it. AI agents now do real work inside companies, and they need the company's context to do it well.
Why AI agents need one
A person who finds an outdated page usually notices. It looks old, or it contradicts something they heard last week, so they ask a colleague. An agent has none of that judgement. It treats whatever it finds as true, whether the page is a year old or the decision behind it was reversed.
Agents also start from nothing. Each one knows only what it was given: a coding agent knows the repository, a chat assistant knows the conversation, and neither knows why the team chose its database or which price was agreed last month. When each agent keeps its own memory, the same question gets a different answer in each tool.
A company brain gives every agent the same source. They read the same facts and follow the same written procedures, and each answer cites where it came from, so a person can check it.
| Agent memory | A company brain | |
|---|---|---|
| Belongs to | One agent, or one person's account | The whole company |
| Holds | What that agent was told or saw | What the company has written, in every connected tool |
| When something changes | Each memory is updated on its own, or not at all | Updated once, read by every agent |
| Shows its sources | Depends on the agent | With every answer |
What a company brain does
However it is built, a company brain does four jobs. The sections below take each one in turn and say how Noetix does it today.
- Take in what the company writes, from the tools where it is written.
- Store it so it can be found by meaning, not only by keyword.
- Keep it current as things change, without someone tending it by hand.
- Answer people and agents, with sources.
1. Taking in what the company writes
A brain that waits for people to feed it falls behind. Knowledge has to arrive as part of the work itself: the document someone already wrote, the channel where the decision was made.
In Noetix, an admin connects Slack, Google Drive and Notion and chooses what to import. Anyone in the workspace can upload files or write pages directly, and AI agents connected over MCP can create and edit pages as they work.
Different sources need different handling. A document is replaced when it changes. A Slack channel is a conversation, so a model reads the new messages and keeps the facts worth keeping as pages, each one traced back to the messages it came from. That is why watching Slack is something an admin switches on, while Drive and Notion are watched as soon as they are connected.
2. Storing it so it can be found
A keyword search misses the page that says the same thing in other words. From everything it holds, Noetix builds a vector index, which finds text by meaning, and a knowledge graph of the people, systems and decisions the text mentions and how they relate. A question uses both.
What the brain holds comes in two kinds, split by where the real copy lives. Synced items from Slack, Drive and Notion belong to their tool, so they are never edited in Noetix: the next sync would overwrite the change. Pages (uploads, text written in Noetix, pages agents create) live only in the brain, so they are edited in place and every version is kept.
A one-off fact goes on a page called Notes as a dated line, so even a single fact has a place where someone can read and correct it.
3. Keeping it current
A brain that only grows turns into the stale wiki it was meant to replace. Keeping it current is the hard part, and most of the design goes there.
- Synced sources are re-read on a schedule. When a Drive file or Notion page changes, the old copy is replaced, not kept beside the new one.
- Editing a page replaces the old version in what agents read. The old version stays in the page's history and can be restored.
- When two sources disagree, the brain records the contradiction. An answer that rests on a disputed source says there are two versions, with their dates, instead of quietly picking one.
- A source that says when it took effect, such as a decision date, is shown with that date, so an answer can tell which version is newer.
- Decisions get their own pages: what was decided, why, what else was considered and when to revisit it. A reversed decision is not deleted. Its page says what replaced it, because why the company changed its mind is part of what it knows.
- Anyone can ask what changed since a date. The answer starts with what the brain now believes differently and the contradictions it has not settled, then lists the pages edited and the sources synced.
- A connection that keeps failing pauses itself and is marked as needing attention, with the reason, so a broken source is visible instead of silently going stale.
4. Answering people and agents
People ask in the Noetix app. Agents ask over MCP, the open protocol AI tools use to reach outside services, or over the REST API. Claude Code, Claude Desktop, Cursor and Codex connect with one line and sign in with a Noetix account.
Every answer lists the sources it used, with their dates and the passage that matched. A conclusion drawn by combining sources is presented as a conclusion, not as something a source said.
When the brain has nothing on a question, saying so is the useful answer. The question can be flagged as a knowledge gap. Gaps are listed with the most asked first, so the team knows what to write down next.
Procedures, not only facts
Knowing a fact is not the same as knowing how the company does something. A brain should also hold the written procedures: how to ship a release, how to handle a refund request, how to record a decision.
In Noetix these are skills: versioned documents an agent follows when a task matches one. Every connected agent follows a published skill on its next task, so the rules are stricter than for pages. Any member can draft a skill. Only an admin can publish one, and agents only ever see published skills. A skill can name the pages it relies on.
Who gets to write
An AI can read a long discussion about a possible change and record it as the decision. Company knowledge is easy to get confidently wrong, so a brain needs clear rules about what an agent may change on its own. In Noetix:
- A conversation with the assistant is not saved to the brain by itself. Something enters the brain only when a person or agent writes a page, or a person saves a fact the assistant suggested.
- A file attached to a chat stays with that chat. Adding it to the brain is a separate, explicit step.
- Agents may edit pages, because every version is kept and can be restored.
- Deleting something, and creating or changing a skill, waits for an admin's approval.
- Facts taken from Slack must trace back to the messages they came from. Whoever connected Slack chooses whether they are saved directly or wait for approval.
Sources, access and your data
A company brain holds the most complete picture of a company in one place. Who can read it, and what it depends on, matter as much as how good its answers are.
- One workspace is one company, with one brain. Each workspace's data is stored separately from every other workspace's.
- Members ask, add knowledge, edit pages and draft skills. Admins also connect sources, delete, invite people, publish skills and decide approvals.
- The brain is its own layer, not a feature of one chat app or one suite of tools. Any MCP client, or any program with an API key, reads the same brain.
- During early access, a new workspace gets some AI usage from Noetix to try it. A workspace can also bring its own key, from OpenRouter, OpenAI or any OpenAI-compatible provider, so model costs go to its own account at the provider's prices.
- Noetix does not train models on your content. Requests made with the workspace's own key run under its account with that provider, and its data settings.
What Noetix does not do yet
The limits today, so they are not a surprise later:
- Everything synced is visible to every member of the workspace. Permissions per source are not built yet.
- Deleting a file in Drive or Notion does not remove it from the brain. Remove it in Noetix too.
- Old knowledge does not fade on its own. What is out of date has to be replaced by a newer source or edited.
- The brain only knows what was written down somewhere. Knowledge that lives only in people's heads has to be written first, and the knowledge gaps list shows where to start.
- Jira, Linear, CRMs, Microsoft 365 and meeting transcripts are not connected yet.
Getting started
Noetix is the company brain for AI agents, now in early access: leave your email and we'll send an invite, or write to us with questions.