


Software that changes when the work does
A chatbot answers questions, and someone still copies the answer into the report.
Traditional software does one job, and changing it means a new contract.
A team's systems
Agents do routine steps
People make the calls
Finished work, back in place
That's the Command Layer.
People have always been the layer connecting a team's systems. Now agents can do that job.
When a new need shows up, the team can add it in days, without a new contract. Each new team runs on what’s already secured.
Department of War · U.S. Army · U.S. Air Force · U.S. Special Operations Command · U.S. Department of Energy · Missile Defense Agency · SLAC National Accelerator Laboratory
Three bets we made early
We founded Legion in 2022 on three bets about how national security would adopt AI. The market has been catching up to all three ever since.
Security comes first
We run inside the customer's boundary. We've worked in classified environments since 2023 and were the first generative AI platform accredited at IL6.
Models will be a commodity
They keep improving and look alike, and open models are close behind. So we connect any model, including ones a customer hosts, to the team's data, tools and approvals.
Cost decides what gets adopted
AI that costs more the more people use it doesn't spread. Ours is built so cost falls as usage grows, and customers see what they spend.
We started in national security on purpose. A team there may work on a classified network one day and with no connection the next, and the software has to keep up with both. What holds up there holds up anywhere the stakes are high.
Our Investors






What we value
Velocity
We close the gap between what a team decides and what gets done, starting with software in people’s hands early.
Resiliency
We plan for the network to drop and the requirements to shift, because in this work both happen.
Control
Authority belongs to the people accountable for the outcome, and everything we ship is built to keep it there.

Build What's Next
Our team has shipped software into hard conditions, used it from the user side, or left commercial AI for problems with real stakes.
If that's the work you want to do, we'd like to hear from you.


