A company forgets in a specific way. The document survives. The decision survives, usually, as a line in a ticket or a slide. What disappears is the reasoning: why this option and not the other one, what evidence settled the argument, who approved the choice, and what would have to change for the company to choose differently. Six months later the question returns, and the company pays for the same argument twice.

A company brain exists to stop that loss. It is a system that keeps the company's decisions in one record, with the reason, the source and the approver attached to each one, and keeps that record current as the work happens. People can open it and read it. AI agents can work from it. When a decision changes, the record shows the change and keeps the earlier version readable, so the company can see not only what it believes now but what it believed then, and why it moved.

What a company brain keeps

The unit of a company brain is not the document. It is the decision, kept with the material that makes the decision intelligible later:

  • The decision itself, stated plainly: what was chosen, and what was not.
  • The reason: the argument that carried the choice, including the options that lost.
  • The sources: the messages, documents, tickets and meetings the reason rests on, linked so a reader can check them.
  • The approver: the named person who owned the call.
  • The date and the version: when the entry was made, and which earlier entry it replaces.

Documents still exist underneath. The brain does not replace them. It sits above them and keeps the layer that documents alone never held: the reasoning that connects them.

How it differs from enterprise search

Enterprise search answers a question at the moment you ask it. It looks across the company's connected apps, finds the passages that match, and assembles an answer from the fragments. Glean, the best known product in that category, describes its own work as connecting knowledge across every app, system and team a company runs. That is a real and useful job. It is a different job from remembering.

Three differences matter. First, search works on what was written down as documents and messages; the reasoning behind a decision often lives in a thread, a call or an argument that no single document states. Second, search assembles its answer fresh each time, so two people can receive two different accounts of the same decision, and neither account is kept. Third, search has no view of standing: it cannot tell you which of two conflicting documents records the decision the company actually made, or whether that decision was later reversed. A company brain keeps one reviewed record of the decision, so the answer to why we did this is the same on Monday and in March, and a reversal is an event in the record rather than a rumour.

The two also work together. Search finds things. A brain is where the found things, and the reasoning drawn from them, are kept in a form the company has agreed to stand behind. Our comparison of Jilu and Glean sets the two kinds of product side by side.

How it differs from a wiki

A wiki is a set of pages that people write and maintain by hand. It fails in a predictable place: writing the page is a second job, done after the work, and the work always wins. Pages go stale quietly, nobody knows which page is current, and the pages describe how things are supposed to work rather than what was decided and why.

A company brain is built from the work itself. The raw material is the chat, the documents, the tickets and the meetings the company already produces. Drafts of record entries are drawn from that material, and a person reviews and publishes them, so keeping the record is a review task measured in minutes rather than a writing task measured in evenings. The record is organised around decisions rather than topics, and a superseded entry stays visible as superseded instead of being silently overwritten, or left to compete with its replacement.

What a company brain is not

It is not a chatbot window over the company's files. An answer that cannot show its source is an opinion, however fluent it sounds, and a memory that lives only behind a chat box cannot be opened and read by the people it belongs to.

It is not a surveillance tool. A record that scores or ranks the people who feed it will be starved by them, and it will deserve to be. The record exists to serve the people who make it: to let a new hire learn in a week what used to take a season, and to let a team stop re-arguing a settled question.

And it is not a promise of perfect recall. Entries can be wrong. The test of a company brain is not that it never errs; it is that every entry carries its source, its date and its approver, so an error can be found, corrected and replaced in the open. The four beliefs behind that standard are set out in the Jilu manifesto.

How Y Combinator defines it

The category has a definition beyond this site. In Y Combinator's Summer 2026 Request for Startups, Tom Blomfield describes a company brain as a system that "pulls knowledge out of all these fragmented sources, structures it, keeps it current, and turns it into an executable skills file for AI". He adds: "This isn't a company-wide search or a chatbot over your documents. It's a living map of how a company works." The entry is published on Y Combinator's Requests for Startups page (checked 13 November 2026).

Jilu's record makes the same four moves. It pulls knowledge from the fragmented sources where the work already happens: chat, documents, tickets, meetings, email and customer records. It structures what it pulls into decisions, each one kept with its reason, its sources and its approver. It keeps the record current as the work goes on, showing each change and keeping the earlier version readable. And it makes the record usable by AI: people read it directly, and agents work from it and answer with its sources, so one governed record, rather than a separate summary inside each tool, is the source that people and agents share.

One difference is deliberate. The Request for Startups sketch imagines this layer feeding autonomous action, as an executable skills file that an agent runs. Jilu keeps a governed record instead of an auto-generated file. A person reviews and approves an entry before it stands on the record, and agents work from the record under the same permissions as the person asking, rather than executing a file that no one approved.

What to look for when you evaluate one

Products in this category describe themselves in similar words. The differences show up in the record itself, so ask to see the record, and check these eight things:

  1. Sources on every entry. Pick any claim in the record and ask to open the message, document or meeting behind it. If the trail stops at a summary, you are looking at an opinion store.
  2. Reasons and approvers, not outcomes alone. An entry that records what was decided without why, and without who, will not settle the argument when it returns.
  3. A review gate. Machines can draft. A person should approve before an entry joins the record the whole company relies on. Ask what publishes itself, and be careful with the answer that everything does.
  4. Visible corrections. When a fact changes, the new version should take effect and the old one should stay readable, marked as replaced. Silent overwrites destroy the history the record exists to keep.
  5. Permissions that follow the sources. People and agents should see only what they are allowed to see in the underlying material, and the record should inherit those limits rather than flatten them.
  6. A record you can read, not only ask. If the only door into the company's memory is a chat box, the memory belongs to the model. The record should stand open to the employees it describes.
  7. Export. The record is the company's asset. Ask how it leaves with you if you leave the product, and in what form.
  8. Plain dealing. Published pricing, and claims you can check against the vendor's own pages. A vendor whose claims cannot be checked will not tell you what its record cannot do either.

Jilu is one product in this category: a company brain that keeps decisions with their reasons, sources and approvers, built on the draft, review and publish discipline described above. How it works, step by step, is set out in How Jilu Works, and how a company brain and a search product divide the work is set out in Jilu vs Glean.