About

Why we’re building bolde.

Sensitive teams are being told to use AI, then told not to move the data. Bolde is built for that gap: AI that runs inside your network.

The story

Building it for ourselves first.

We have deployed AI on networks where guessing is not acceptable. Answers need sources. Actions need records. Data does not leave just because a tool is convenient.

When we looked for AI tools for our own work, the good ones all wanted sensitive data to move through someone else’s infrastructure.

So we are building the version we needed: AI that runs inside the network, on hardware the customer controls, with sourced answers and a reviewable record.

Other regulated teams have the same problem. Bolde is the product form of the tool we wanted for ourselves.

The gap

Data control exists for real reasons.

Cloud AI often means your data leaves your control. For finance, legal, defense, and regulated teams, that can be enough to say no.

Private inference is now practical. That changes what these teams can safely use.

The fix

Private AI, sourced answers, clean records.

Dedicated infrastructure. No external model calls. Every answer links to source material. Every important action is logged.

The question was not whether this could work. It was whether anyone would build it for the teams that need it most.

The principles

How we build.

Clarity over polish

If a detail does not help the user decide, we cut it.

Private by default

The product keeps work inside the line. The company follows the same rule.

Less, done better

We would rather ship the useful thing than explain ten things that almost work.

The founders

Depth is the moat. Not the credential.

The lead is not that we hold the right certifications. The lead is that we have operated AI on the highest-stakes networks in existence — the ones where a wrong answer is not a bad experience, it is a consequential failure. That background is what Bolde is being built from.

Jason Wareham — co-founder

Former U.S. Marine Corps Judge Advocate (LtCol, USMC). Georgetown Law LL.M. in digital evidence and cybercrime. CEO of Mojave Research Inc., the company The Washington Post first reported was selected by the Office of the Director of National Intelligence to deploy AI on intelligence-community data — coverage subsequently followed by Reuters, CNN, and NBC News. A career spent in rooms where “we’ll fix it in the next sprint” is not an available answer — where evidence, procedure, and consequence are the operating constraints, not the marketing copy.

Manbir Gulati — co-founder

Machine learning research scientist. Years of applied work inside U.S. defense and intelligence-community AI programs — the systems where “send it to ChatGPT” has never been a sentence anyone is allowed to say. Co-architect of the AI capability behind the same Mojave Research engagement reported by The Washington Post. Published research on synthetic data and privacy-preserving inference. Co-developer of the open-source RF-ML toolkit the signal-processing community runs on. Princeton Computer Science.

Request early access.

Bring the sensitive workflow you want to test first.

Request Early Access