About

Companies know their people. They cannot find the right one.

Keegi is a way to ask a company who has already been through the thing you are about to attempt, and to get names back with the work behind them.

The knowledge is there. The index is not.

Every company past roughly two hundred people knows less than the sum of its people, and the gap widens with every hire. Somebody spends three days on an integration a colleague two floors away shipped last year. A new joiner puts a question into a channel of four hundred and hears nothing, because the person who could answer is in a different channel. A manager staffs a project from the six names they happen to know.

None of that is a knowledge problem. The knowledge exists. What is missing is any way to reach it that does not run through somebody’s memory.

A title tells you what somebody is responsible for.

It does not tell you what they worked out along the way. A profile says Senior Product Manager. The part you needed was somewhere else:

Ran the migration off the old billing system.
Launched in three markets and can tell you which one was hard.
Has solved this exact integration twice, once badly.
Sat through an enterprise rollout that failed and knows why.

None of that is on an org chart, and close to none of it is in the skills field somebody filled in during onboarding and has not opened since. The most useful thing to know about a colleague is the specific work they have already been through.

So Keegi reads the work instead.

It reads what a company has already produced, in the systems it already runs. Jira and Linear for what was built and who carried it. Slack for the questions people answered. Notion for what somebody took the trouble to write down. GitHub for what was reviewed and merged. The HRIS supplies names and reporting lines and nothing more, because a self-reported skills matrix is a claim rather than a record.

Out of that it writes one card per person: what they have worked on, what they shipped, what they have repeatedly been the one to fix. Then somebody asks a question in an ordinary sentence, something like who can help me with the Lithuanian tax filing, and up to three colleagues come back.

What this looks like without it.

Ask the channel.: Four hundred people in it, and the person who knows is in a different one.
Search.: Forty-seven results. None of them answer the question.
Read titles.: VP of Engineering. So, probably?
Ask the two people you always ask.: They are the company's real index. It does not survive their holiday, let alone their resignation.

Every one of those is somebody doing retrieval by hand, badly, because there is nothing better in the building.

Every name arrives with the reason for it.

A name on its own is a guess you are being asked to trust. Keegi puts one line under each person and that line points at something real: the ticket they closed, the thread they answered, the change they reviewed. You can open it and read it.

That is the difference between a system that has read what people actually did and one that matched your words against a paragraph they wrote about themselves.

When the evidence is thin, Keegi says so. It will return one name, or none, in preference to three plausible ones, and it will tell a company its corpus is too weak to buy this yet.

What we believe

Experience is an asset and most companies treat it as exhaust.

What somebody worked out by doing a hard thing should not become unreachable the moment the project closes and the channel is archived.

Nobody should have to be well connected to get help.

The informal network is real and it works, and it works only if you already know people. New joiners, people in the smaller office, and anyone who arrived after the founding group get the worst version of it.

The right person to ask is often somebody you have never met.

Finding people is not the same as watching them.

Reading what a company produced in order to work out who can help is one thing. Reporting on how much any individual works is another, and Keegi does not do it. There is no productivity score and no per-person dashboard for a manager, at any price, for any customer. It is not a setting left switched off; it is absent, because the moment it exists as a feature flag it exists in a works council’s threat model. The security page carries the whole boundary.

The name

Keegi is Estonian for someone. It comes from ma tean kedagi, I know someone, which is the sentence a colleague says right before they solve your problem for you.

We would like more companies to be able to say it on purpose rather than by luck.

Where this goes

The version of this worth building is one where asking for help inside a company costs about what asking a question costs. Every project leaves the organisation slightly easier to search. Experience stops being stranded when a team reorganises or somebody moves on. And who you can reach stops depending on who you happen to sit near.

None of that arrives at once, and the docs say plainly what is built today and what is not.

Where the data sits

Keegi runs in the EU, with the compute pinned to the same region as the database rather than to whichever one was quickest to provision. Model calls are the exception, under zero data retention: we tried pinning them to Europe and it broke the engine, so that is written down instead of glossed over. The security page has the detail and the sub-processor list.

Talk to us

Twenty minutes. No demo, no pitch. I want to know whether this problem is real where you work.

Five questions first Back to the demo