Why InternalWiki exists

Before InternalWiki, I worked at a property family office. The office managed a portfolio of commercial leases, residential properties, and joint ventures across London. The documents were spread across shared drives, email attachments, and a filing system that predated most of the current team.

One afternoon, a colleague needed to confirm the break clause date on a commercial lease. The lease was a PDF attached to an email from 2019, sent between two people who had both since left the company. She searched the shared drive — nothing. She searched her email — nothing. She asked the office manager, who forwarded the request to the legal team, who found the document three days later.

Three days. For a date that was written in a document the company already owned.

This wasn't a one-off. The office managed 47 commercial leases, 23 residential properties, and 8 joint ventures. Documents were spread across a shared G: drive (some files dating back to 2012), personal email attachments, a SharePoint site that nobody maintained, and a physical filing cabinet in the conference room. Every week, someone spent hours tracking down a document the company already owned.

The freshness problem

That wasn't the worst part. The worst part was what happened next. A different colleague found what he thought was the current lease in a "Leases" folder on the shared drive. But it was the original 2017 version — before the 2019 renewal that extended the term by five years. He used the wrong expiry date in a board presentation. Nobody caught the mistake until the tenant's solicitor pointed it out two weeks later.

The document's age wasn't the problem — plenty of old documents in property are still valid. The problem was that there was no way to tell whether this specific document had been superseded. The shared drive treated all files the same. Old didn't mean outdated, but there was no signal either way.

The moment it clicked

I watched a senior associate prepare for a tenant meeting. She spent 45 minutes finding the right lease, verifying it was current, and cross-referencing the break clause with board minutes from 2019. Then she presented her findings to the partners — with sources, with dates, with confidence in her conclusion.

That's how a skilled analyst works. What if the AI worked the same way? Not just answering, but showing its sources. Not just retrieving, but verifying freshness. Not just confident, but accountable. An AI that could do what she did — in seconds instead of 45 minutes — would change how the office worked.

Why existing AI tools didn't help

By 2025, AI assistants were everywhere. I tried several for this exact problem. They all had the same issue: they'd give you an answer, but you couldn't tell where it came from. Ask about the lease expiry and you'd get a confident response — but was it from the 2017 version or the 2019 renewal? The AI didn't say. It couldn't say, because it didn't track which document supported which claim.

For personal use, this is fine. For a family office making decisions about $2M lease renewals, it's unacceptable. You need to see the source. You need to know it's current. You need proof.

The three principles

InternalWiki was built around three principles that came directly from this experience:

Cite everything. Every factual claim in every answer must trace back to a specific passage in a specific document. If the AI can't cite it, it shouldn't say it. Because enterprise decisions have consequences. When someone acts on an answer about lease terms, parental leave, or compliance obligations, they're making commitments. Wrong answers cost real money. Citations let people verify before they act.

Enforce permissions. In a family office, not everyone should see every document. The investment committee papers are separate from the HR files. Permissions from the source systems should be respected, not bypassed. Because “mostly secure” isn't secure. A probabilistic output filter that catches 99% of leaks still exposes sensitive data 1% of the time. For a company processing 1,000 queries a day, that's 10 leaks per day. At enterprise scale, probabilistic filtering is a liability, not a feature.

Understand freshness. A lease agreement from 2019 might be the current lease. An email from yesterday might already be wrong. Freshness depends on what the document is, not when it was written. Because the problem that started this company — a colleague using an outdated lease in a board presentation — came from treating document age as staleness. A five-year-old contract isn't stale. Yesterday's Slack message might be. The distinction matters, and no existing tool made it.

We chose to go deep on one thing — trust — rather than broad on everything. InternalWiki doesn't manage projects. It doesn't run workflows. It answers questions from your existing documents, with proof. That focus exists because trust isn't a feature you bolt on. It's an architecture you build from day one.

That's why InternalWiki exists. Not because the world needed another AI tool, but because the specific problem of finding trusted, cited, current answers from existing enterprise documents wasn't solved. It is now.