How to choose a company brain platform,
and test it properly.
What to insist a company brain actually does, who is selling one in 2026, the questions that separate a product from a demo, how to run a trial on your own awkward material, and how to settle the one choice that decides most of the rest.
Brief
A company brain holds what a business knows, keeps it current from the tools the business already runs, and answers questions from it with the evidence attached. It differs from a wiki because nobody maintains it by hand, and from a chatbot because the knowledge accumulates between conversations.
Insist on four properties when you evaluate anything in this category: it keeps itself current, it holds context over time, it holds a position rather than only documents, and it shows its evidence. Holding a position is the property that makes an answer something your business can act on.
The decision that settles everything else is broad or deep. Most platforms are broad: connect all your data, serve it to agents, stay neutral about what any of it means. That is right when the problem is finding things across a large estate of tools. A deep one covers a single domain properly. Work out which you are shopping for before you compare features, because the two do not compete on the same axis.
Run a trial on your own awkward material. Every demo in this category looks extraordinary, because a curated document set and a good model always do. A pilot designed well settles it in 30 days.
What to insist a company brain actually does.
The term is new enough that vendors use it loosely. These are the properties that separate the thing from a folder with a search box, and each one is worth writing into your requirements.
It keeps itself current
Knowledge arrives from the tools you already run, on its own schedule. This is the property that decided the fate of wikis: keeping one accurate was somebody's unpaid second job, and the job always lost.
AskWhat happens when a connector breaks, and who notices first, you or them?
It holds context over time
Evidence accumulates and is weighted, so something seen once in March and again in July is understood as a pattern rather than two unrelated notes. The answers get better the longer you run it.
AskShow me an answer that draws on evidence from more than one quarter.
It holds your current position
The strongest version knows what the company has decided: who it serves, what it charges, why it wins. That is what lets it tell you when new evidence sits awkwardly against something you agreed six months ago.
AskWhere does our current position live in this, and what happens when it changes?
It shows its evidence
Every answer comes with what it was drawn from, and when. Without that you have a confident narrator, and a confident narrator is worse than no answer at all when a real decision rides on it.
AskLet me click through to the source, and let me find one where the citation is thin.
A fuller treatment, including how a brain differs from a wiki, an assistant and enterprise search, is on what is a company brain.
Why companies started buying this year.
Assistants got good enough to expose the real problem. Once a team can ask a capable model anything, the limit stops being the model and starts being what it knows about your business. A brilliant assistant with no context gives brilliant, confident, generic answers.
The returns question got asked out loud. Boards that funded AI enthusiastically in 2025 started asking what came back. The uncomfortable finding across the market is that individual productivity rose while company results largely did not, because the knowledge stayed inside individual sessions instead of accumulating anywhere the business could use.
Retrieval became cheap. Vector search, embeddings and connectors that once needed a specialist team are now commodity infrastructure. That collapsed the cost of building the plumbing and moved the competition to what sits on top of it.
The plumbing is solved. What a company brain is for is the open question, and that is where the market is dividing.
A new name for an old ambition.
Useful context when you are being sold to, because each previous era failed in a specific way and you can ask a vendor which of those they have actually fixed.
The knowledge management era
Intranets, wikis and document management promised the same outcome and mostly failed for the same reason: they depended on people writing things down and keeping them accurate, forever, with no reward for doing it.
Ask a vendorWhat does my team have to remember to do for this to stay accurate?
The enterprise search era
Index everything, make it findable. Genuinely useful, and it took the burden off the author. What it could not do was understand: it returned matching documents and left the reader to work out which one was still true.
Ask a vendorHow does this tell the current answer from the closest one?
The company brain era
Language models can read, summarise, connect and explain, and connectors keep the input flowing. The term crystallised in 2026, was named in Y Combinator’s summer request for startups, and the funding and the incumbents followed within months.
Ask a vendorWho decides what the evidence means, and is that person yours or mine?
Each era removed a human bottleneck and revealed the next one, first authoring, then finding, then reading. The bottleneck now is judgement.
Who is selling a company brain.
Publicly documented positioning as of August 2026, grouped by what each is shaped like rather than scored. The category moves monthly, so treat this as dated rather than settled.
| Who | Shape | Best suited to | Worth knowing |
|---|---|---|---|
| Glean | Broad. Search and assistants across everything the company holds | Larger organisations with a big estate of tools and an internal team to run it | The category incumbent, now positioning around an enterprise brain |
| Microsoft | Broad. Knowledge and agents wired into the productivity suite | Companies already committed to the Microsoft tenancy | Arrives by default wherever the tenancy already is, which makes it the value option |
| Gyld | Broad. A context layer for AI agents, built around MCP | Teams that want their own assistants to reach company data | Connect your data once, then reach it from ChatGPT, Claude or Cursor |
| Colrows | Broad and technical. A semantic graph over company data | Data teams comfortable modelling their own knowledge | Prices on semantic assets, on an annual committed figure |
| Falconer | Engineering-leaning knowledge platform | Technical organisations where the knowledge is already written down | Publishes its pricing, and its own guides on building a company brain |
| Brain Co. | Broad, institutional | Early to say; the positioning is still forming | Backed by Kushner and Elad Gil at Series A |
| nBrain, Ability.ai, Klikflo, Webair | Various, from small-business to technical | Depends heavily on the individual product | Named across category coverage, with thinner public detail to evaluate |
| Liffey | Deep. One domain: how a company wins and grows customers | Companies of roughly €1M to €100M+ in revenue | A structured method underneath, and a senior commercial architect on top |
Two things this table deliberately leaves out. There is no scoring, because the products are shaped for different jobs and a single score would only flatter whoever wrote it. And there are no feature ticks, because at this stage of a category every vendor's roadmap is longer than their product. If we have described something of yours inaccurately, tell us and we will correct it.
The questions to put to every vendor.
Vendor-neutral on purpose. Ask these of everyone you shortlist, us included. Each one comes with what a good answer sounds like, because the tell is usually how quickly it arrives rather than what it contains.
What happens in month six?
Anything in this category impresses in week one. Ask who notices when a connector silently stops, who decides an answer has drifted, and what the vendor does about either.
Good answerA named monitoring and review process, and a straight admission of what still falls to you.
Can it handle knowledge that is not a document?
Bring a call recording, a stalled deal and a support thread. A great deal of what decides revenue was never written as a file, and a platform that only ingests files will still answer confidently about the part it happens to hold.
Good answerThey ingest it in the trial, in front of you, rather than describing the roadmap.
What does it do when it is wrong?
Feed it a position you superseded months ago. Watch whether it says so, hedges, or repeats it back with conviction. This is the most revealing test and the one demos are least prepared for.
Good answerIt flags the conflict, or it tells you it cannot and explains what would.
Can I follow the evidence?
Every answer should trace to a dated source you can open. Look for citations that are vague, absent, or occasionally point at something that does not say what the answer claimed.
Good answerClick through works every time, and stale sources are marked as stale.
What happens the day someone leaves?
Ask how access is inherited, how it is revoked, and what becomes of everything that person's account could reach. This rarely comes up until somebody leaves, and it is far easier to settle now than to fix later.
Good answerPermissions mirror the source tools and revoke with them, demonstrably.
Who owns this inside our business?
Every one of these needs a named person on your side, whatever the sales conversation implies. Work out who before you sign, because a platform without an owner becomes another place where answers go stale.
Good answerThe vendor tells you plainly how much of the ownership stays with you.
On data, ask where content is processed, where it is stored, how long it is kept, and whether it is used to train anything. In Europe those are compliance questions with real answers, so expect a clear one from every vendor you shortlist.
A trial that tells you something.
The default trial is a curated document set and a friendly demo, and it will always go well. If you want to learn anything, run it this way instead. It takes a fortnight and it is the same shape whichever vendor you are testing.
- Write ten questions first, before you see any product. Real ones your business needs answered, agreed with the people who will use it. Everything else follows from these, and writing them afterwards means writing them to fit what you were shown.
- Answer them yourself, on paper. You need a known-good baseline or you have no way to judge what comes back. This step is the one people skip and the one that makes the trial worth running.
- Give it your awkward material, not your tidy material. The messy folder, the call recordings, the deal that did not close. Tidy inputs test nothing.
- Include something you know is out of date. A superseded pricing sheet or an old buyer definition. Then ask a question it would answer wrongly.
- Have three different people ask the same question. Different words, same intent. Consistency across phrasings tells you more than any single impressive answer.
- Check every citation on the answers you liked. Especially those. A confident answer with a citation that does not support it is the failure mode you are shopping to avoid.
- Ask the same ten questions again two weeks later, after new material has flowed in, and see whether the answers moved in the right direction.
Judge a platform on the questions you brought, not the ones it answered well.
Broad or deep is the real decision.
A broad platform is designed to hold everything, which means it has to stay neutral about what any of it means. That neutrality is exactly what makes it work across engineering, legal, support and finance at once, and it is a genuine strength.
What a commercial team needs is not the same as what an engineering team needs. Go wide enough and HR policy, internal admin and process documentation flow into the same pool as pricing, pipeline and the read on which segment actually converts. Every answer then has to be filtered by somebody who already knows which part of the company they are asking about. The system tries to do too much, and gets shallower as it gets broader.
We would rather be excellent at one job than adequate at all of them.
So Liffey covers one domain: how a company wins and grows customers. Underneath it is a method that says how the parts of a commercial strategy depend on each other, and alongside it is a senior operator whose week includes judging what new evidence means. Neither is something a neutral context layer sets out to provide.
Depth is not exclusion. Where a signal from elsewhere in the business changes what your team can sell, it belongs: product releases and feature-level detail pulled from GitHub, for example, which is how several of our clients run it, so the sales team knows what shipped and the marketing team knows what is worth saying about it. The test is always the same. Does this change a commercial decision?
Which one you are shopping for.
Work this out before you compare features. The two shapes do not compete on the same axis, and a shortlist mixing them will produce a scorecard that means nothing.
Buy a broad platform when
- You want one layer across engineering, legal, support, finance and commercial
- The core problem is finding things across a large estate of tools
- You have internal people with time to decide what the knowledge means
- What you need is retrieval for agents rather than a commercial position
- Company-wide search is already a named priority with a budget behind it
- You are large enough that a rollout programme is a normal piece of work
Buy a deep one when
- The problem is commercial: who you serve, what you charge, why you win
- You want a structure that holds positions, so an answer can be current rather than merely close
- What decides your number sits in calls and deals as much as in files
- Nobody internally has time to turn evidence into decisions every week
- You want senior operating experience attached rather than left for you to staff
- You would rather start in days than run a procurement programme
These sit comfortably together, and plenty of companies end up with both. An organisation running Glean or Microsoft across the whole business can still want something dedicated to the commercial side, in the same way it runs something dedicated to finance.
Your company’s commercial brain
and a senior operator to maintain it.
A senior commercial architect, the method they work from, and the brain that holds what your company knows about winning.
The architect
A senior commercial operator, twenty years from scale-up to PLC, working alongside your team on the strategy itself. They judge what the evidence means before anyone acts on it, which is the property no neutral platform sets out to provide.
- From 20 hours a month
- Has carried a number, set pricing, built a motion
- How an engagement runs
The method
A structure that says how the parts of a commercial strategy relate: the value you deliver, the buyers it is worth most to, how you reach them, and what happens after the win. It is what makes the answers hang together into one commercial picture.
- Built from work with companies across Europe
- Your version of it is yours to export
- The method in full
The platform
It reads the systems you already run, holds what matters about how you win, and answers in plain English with the evidence attached. It watches patterns as they build and snapshots the pipeline weekly, so this week reads against last week.
- Calls, deals and threads, as well as documents
- Ask from Slack, Claude, ChatGPT, Gemini, Copilot or Mistral
- An MCP endpoint, a versioned REST API and a CLI
Run the same trial against us as well. There is a recorded walkthrough of a working commercial brain, and a five-minute diagnostic that names the weakest part of your commercial picture without you talking to anybody.
Questions people ask.
What is a company brain?
A system that holds what a company knows, keeps it current from the tools the company already runs, and answers questions from it with the evidence attached. It differs from a wiki because nobody has to maintain it by hand, and from a chat assistant because the knowledge accumulates between conversations rather than disappearing with the session.
How do we shortlist company brain platforms without wasting a quarter?
A workable sequence:
- Decide broad or deep first. One question, and it removes most of the market either way.
- Write your ten questions and answer them yourself, before you take a single demo.
- Shortlist on shape, not features. Anything at this stage of a category demos well and ships differently.
- Run one trial properly rather than three loosely. Depth beats breadth here.
- Ask every vendor who owns it internally, and hold their answer against your own capacity.
Who are the main company brain platforms in 2026?
Glean and Microsoft are the broad incumbents. Gyld, Colrows, Falconer, Webair, nBrain, Ability.ai, Klikflo and Brain Co. are among the newer entrants, mostly building horizontal context layers. Liffey is the one built for a single domain, commercial strategy, with a method and a senior architect included rather than left for you to staff.
How is Liffey different from Glean?
Glean indexes the whole company and is very good at finding things across a large estate of tools. Liffey covers one domain, holds a position rather than a document set, and comes with a senior commercial architect who judges what new evidence means. Different jobs, and a large organisation can reasonably want both.
Should we wait for Microsoft to ship a company brain?
If the requirement is company-wide search inside the productivity suite, waiting is defensible and the price will be good. If the requirement is a commercial strategy that stays current and is applied by someone senior every week, that is a different product with a different owner, and it is unlikely to arrive as a feature of a tenancy.
What should we budget for a company brain platform beyond the licence?
An internal owner, whichever product you buy. Somebody has to decide what belongs in there, remove what has been superseded, and read the evaluation results. Vendors rarely lead with this, and it decides whether the platform stays useful or stops being opened at all.
Does Liffey connect to AI agents the way the context layer platforms do?
Yes. Liffey is reachable from Slack, Claude, ChatGPT, Gemini, Copilot and Mistral through an MCP endpoint, alongside a versioned REST API and a CLI. The connection is the commodity part. What your assistant reaches when it gets there is what decides whether the answer is any good.
What should we ask every company brain vendor?
Six questions separate the shortlist quickly, and each has a right answer the vendor can give in a sentence.
- Where does our content live and get processed? Database, application, inference and embeddings are four separate answers.
- Who are the named sub-processors, and where are they?
- Are our inputs used to train models, and what is retained once an answer is returned?
- How do permissions work, and what does a leaver trigger?
- What can we export, in what form, and how quickly is the rest deleted?
- Who is accountable for what is in there being current?
Accountability for currency predicts satisfaction more reliably than any feature comparison. Liffey answers the first five in published documents, and the sixth is a named senior commercial architect who works in your business.
How do we run a company brain pilot that tells us something?
Write down the 10 questions the business genuinely needs answered, and answer them yourself first, so you have something to measure against. Connect real evidence rather than a curated sample, because a clean corpus proves nothing about an ordinary Tuesday. Give it 30 days and a named owner.
Then measure one thing: whether anybody acted on an answer without going back to check the source. Usage counts and satisfaction scores look healthy long before that happens, and that is what you are buying.
Do company brain platforms use our data to train models?
Worth asking as two questions rather than one: whether the vendor trains on your content, and what the underlying model provider retains. The second is where the real answer usually sits, since most platforms in this category build on somebody else’s model and inherit its terms.
Liffey has enterprise terms with Anthropic and AWS that exclude your inputs from model training, and the AI layer runs with zero data retention, so prompts and responses are processed in memory. Every session starts from your own knowledge base.
What happens to our knowledge if we leave a company brain platform?
Buy on the basis that your business keeps what it builds. Ask for the export format before you sign, confirm it carries the reasoning rather than only the source files, and get the deletion window in writing.
With Liffey you export the full structure at any time, and we delete or return the underlying content within 30 days at your choice, under the published data processing agreement. What you decided, and why, stays written down in your business.
The company brain category is young. How do we buy well into it?
Buy so that the value holds through any change of supplier. That means the thinking is written down in a form you can carry, the positions read as a document a person can pick up rather than as rows only the product understands, and you have tested the export yourself.
Liffey is built to that standard deliberately, because the same property is what makes the work survive a change of commercial leadership inside your own business.
What does Liffey cost compared to these platforms?
There is a free tier. From €2,995 a month, a senior commercial architect operates fractionally in your business, evolving and being accountable for the strategy that wins and grows your customers. Licence against licence is the wrong comparison, because that architect’s time is included here where a platform leaves you to staff it. Full detail is on the pricing page.
“We’ve grown by nearly 50% over the past year. In that time, Liffey has helped us better articulate our unique value proposition, evolve our pricing model, reset our go-to-market strategy and build scalable commercial systems. All of which has enabled us to make better decisions on a day-to-day basis and grow with more confidence.”
Nick Comer, Founder & CEO
Twenty minutes on your commercial strategy.
Where the strategy is thin, what belongs in a commercial brain, and whether a broad platform would serve you better. If it would, we will say so.