Yurts

Yurts, which presents publicly as Legion Intelligence, develops a generative and agentic AI integration platform aimed at defense, government, and large enterprise organizations. The platform connects large language models to a customer's existing systems of record and document repositories, supporting agentic search, question answering in plain language, drafting of emails, reports and proposals, and multi-step workflow automation. Every agent output is traced back to source documents, actions are auditable end to end, and access is governed by the customer's existing permission policies, with human-in-the-loop approval at defined steps.

The product is designed to run in the environments its customers operate in: cloud, on-premise, air-gapped, classified, and edge deployments, and is described as model-agnostic to avoid vendor lock-in. Marketing materials cite deployment at impact level IL6 and more than 100 integrations. The company states it does not train on customer proprietary data. Earlier positioning described the platform as converging documents from systems such as SharePoint, Teams, Slack, and Google Docs into a searchable repository, with departmental applications spanning R&D, finance, HR, sales, customer support, and legal teams.

Yurts delivered what it describes as the first AI platform for the U.S. Department of Defense on a secret-level network. Reported customers and references include U.S. Special Operations Command, XVIII Airborne Corps, the U.S. Air Force, the U.S. Army, and the Department of Energy.

Founding story

Ben Van Roo co-founded Yurts in August 2022 with former Meta engineer Jason Schnitzer and research scientist Guruprasad Raghavan. Van Roo grew up in an Air Force family and had intended to fly but was disqualified by his eyesight; he instead worked at the RAND Corporation on military supply chain research with deployments to Iraq, Afghanistan, and Kuwait, then on machine learning at Chegg, followed by a period at defense AI company Primer.ai. Around 2017 he concluded that the main opportunity lay not in building foundation models but in rolling them out securely to tens of thousands of users inside existing enterprise software estates. The company originated from a strategic collaboration with the Department of Defense.

Business model

Yurts sells an AI integration and agent platform to organizations, deployed within the customer's own environment (cloud, private data centers, on-premise, air-gapped, classified, or edge) and connected to their existing systems and applications. Revenue is associated with government contracts and enterprise deployments; the company has disclosed contracts with the U.S. Army, U.S. Air Force, Department of Energy, and a $16 million contract with U.S. Special Operations Command, alongside commercial enterprise deployments.

Contract- and deployment-based sales to government agencies and enterprises, including multi-million-dollar federal contracts such as a $16 million agreement with U.S. Special Operations Command.

Traction

By December 2024 the company had contracts with the U.S. Army, U.S. Air Force, and Department of Energy, plus a $16 million contract with U.S. Special Operations Command, and had delivered the first AI platform for the Department of Defense on a secret-level network. It had about 50 employees, roughly a quarter of whom held security clearances. Company materials cite customer references from U.S. Special Operations Command, XVIII Airborne Corps (Scarlet Dragon 26-01), Civil Affairs training cadre, and a Department of Energy research scientist.

Latest developments

The company now markets itself as Legion Intelligence, focused on agentic AI for national security, promoting Legion Multiplayer as a shared work surface and Centurion as a deployable edge system for denied, degraded, intermittent, and limited environments without cloud connectivity. Compliance coverage is cited across IL2–IL6, FedRAMP High, NIST 800-53, SOC 2 Type II, ITAR, and CMMC, with reported deployment outcomes including 94% faster intelligence summary generation and 216 staff hours saved per week.