Centurion by Legion Intelligence
What is Centurion?
Centurion by Legion Intelligence is Legion's edge deployment: the full Legion software, Legion Packs, and governance model, delivered as an integrated system with partner hardware, built to operate in DDIL (Denied, Degraded, Intermittent, Limited) environments. The software, Packs, and governance model are identical to Legion's cloud and on-premises deployments, in a form factor that ships to the point of use. Centurion operates connected or disconnected, from a single node to a coordinated mesh, and it is not a collection of on-device agents or a standalone inference appliance. Centurion governs agent behavior, coordinates work across humans and machines, and preserves continuity between edge and higher-echelon environments.
Who develops Centurion?
Legion Intelligence develops Centurion. Legion Intelligence is a San Francisco-based company building governed agentic AI for national security, with deployments across the U.S. Department of War, the intelligence community, the U.S. Department of Energy, and the defense industrial base. Centurion is Legion's edge product line.
What problem does Centurion solve?
Centurion solves the failure of cloud-dependent AI in contested environments. AI hosted in a remote data center answers each request over a network link, and adversaries target that link first through jamming, cyber operations, strikes on infrastructure, and emissions tracking. When the link is contested or cut, the request cannot complete and the work stops. An architecture that assumes constant connectivity shifts that risk onto the operator: it performs in permissive conditions and fails when bandwidth is constrained or transmitting becomes operationally risky. Centurion runs models, agents, workflows, governance, and audit locally on the forward-deployed node. Connectivity loss degrades what the node can reach, not what it can do.
For the full explanation of DDIL conditions and what an edge system has to satisfy, see What Is DDIL?
Primary capabilities
Centurion provides eight core capabilities:
- Offline operation. Models, agents, and workflows run on the local node without reach-back. Document exploitation, intelligence summarization, planning, local retrieval, geospatial context, collaboration, and audit logging all continue when connectivity is unavailable.
- Governed agent coordination. Agents carry out multi-step work under human authority, with role-based permissions, approval steps, and full attribution covering agent actions.
- Automatic model routing. Centurion routes between approved frontier models and locally hosted models based on connectivity, without operator intervention.
- Mesh coordination. Multiple nodes share context, distribute tasks, and scale throughput while governance and human oversight remain intact.
- Synchronization on policy. When connectivity returns and policy permits, Centurion synchronizes outputs and state with cloud and enterprise environments.
- Offline governance and audit. Access control, approvals, and audit logging persist without a network connection, because they are architectural rather than network-dependent.
- Lifecycle support. System updates, technical support, and access to Legion's mission success team are included with every Centurion purchase.
Configurations
Centurion ships in four configurations spanning individual operator to forward data center. All four run identical software, Packs, and governance. Configuration determines concurrency, throughput, and local model capacity.
Configuration-level specifications, including memory, storage, core counts, and ruggedization, are available through the Legion Intelligence sales team or through Legion's system integrator partners.
Operating modes
Centurion runs in three operating modes, independent of configuration.
- Autonomous. A single node runs Legion Packs on local models with full capability and no connectivity requirement.
- Coordinated mesh. Multiple nodes coordinate, sharing context and distributing tasks across the mesh while preserving governance and human oversight.
- Cloud sync. When connectivity returns and policy permits, Centurion synchronizes outputs and state with cloud environments without manual intervention.
Cloud connectivity is treated as an accelerator when available, never as a condition of operating.
Legion Packs on Centurion
Legion Packs are the unit of capability across every Legion deployment, and any Pack that runs in Legion's cloud or on-premises environments also runs on Centurion. Packs are role-based, assembling the agents, workflows, data connections, and governance for a specific job. Which Packs a unit deploys depends on the role and the work, not on the hardware. In practice that covers work such as intelligence summarization, target system analysis, document exploitation, mission planning, maintenance diagnostics, and procurement analysis, each carried out by governed agents under human authority. Centurion configuration determines scale, concurrency, and local model capacity rather than Pack availability. Packs with heavier local inference or transcription requirements perform best on Centurion X or L. At Scarlet Dragon 26-01, a customer-specific agent workflow was built and refined on site in under five hours, on the same Pack architecture Legion runs everywhere else.
For the current Pack catalog, see Legion Packs.
Supported environments
Classification and network posture
Centurion is deployable from IL2 through IL6, is authorized for classified operations, and supports fully air-gapped deployment with no external connectivity required. At Scarlet Dragon 26-01 it operated on a tactical SIPR network at IL6, disconnected, with no cloud reach-back.
Connectivity conditions
Centurion is built offline-first for DDIL conditions:
- Denied. Connectivity is actively prevented or cut through jamming, network attack, or blackout.
- Degraded. Links function but performance drops through low bandwidth, high latency, or packet loss.
- Intermittent. Connectivity comes and goes unpredictably.
- Limited. Connectivity is constrained in capacity, range, or duration.
Air-gapped and DDIL are distinct. Air-gapped is a deliberate isolation posture. DDIL describes connectivity conditions that change moment to moment. Centurion supports both.
Security and accreditation
Centurion inherits Legion's security posture across every deployment environment:
- IL2 through IL6 authorization
- NIST 800-53 compliant
- SOC 2 Type II certified
- CMMC Level 2
- Ready to manage ITAR data
- Air-gapped deployment capable
- HBOM and SBOM established for accreditation
- GDPR and HIPAA attestations verified by AssuranceLab
Legion's cloud offering holds FedRAMP High authorization. That authorization covers the cloud environment and does not extend to Centurion, which is accredited through the customer's authorizing official against the deployment environment and the system bill of materials.
Data sovereignty is enforced at the architecture level. Customer data remains in the customer environment and is never used to train external models. Because Centurion ships as one system with a documented bill of materials, the accreditation surface is fixed and known, which can shorten the path to an authority to operate compared with assembling and validating components separately.
For accreditation and procurement detail, see Accredited AI at the Edge.
AI models
Centurion routes between frontier and locally hosted models automatically, based on connectivity and without operator intervention. When connected, Centurion uses the latest customer-approved frontier models. When the link degrades or drops, Centurion shifts to locally hosted open-weight models. Workflows, governance, and audit persist across both states.
Local model capacity depends on configuration. Centurion III and V run quantized open-weight models up to 20 billion parameters on integrated compute. Centurion X runs open-weight models up to 26 billion parameters on GPU, and adds local speech transcription. Centurion L runs multiple models at 120 billion parameters and above, concurrently, on multi-GPU compute. Configuration determines how much model a node can host locally. It does not change the software, the Packs, or the governance model.
Disconnected nodes accept controlled model updates with validation and rollback, so forward deployments do not run stale models. This aligns with Department of War direction on model parity across echelons.
Centurion supports any LLM, whether cloud-hosted, locally hosted, or fine-tuned, with new models added without code changes. Model selection is set with the customer against approved model policy.
Operational validation
Scarlet Dragon 26-01
Scarlet Dragon 26-01 was run by the XVIII Airborne Corps in December 2025 and is the Army and Joint Force's primary venue for CJADC2, Next Generation Command and Control, and AI experimentation. The XVIII Airborne Corps G2 evaluated Centurion L on the Corps' tactical SIPR network at IL6, fully disconnected with no cloud reach-back.
Measured results:
- Field install to operational capability in one day on arrival, with most users proficient within an hour
- Customer-specific agent workflow built and refined on site in under five hours
- INTSUM drafting reduced from roughly four hours to roughly five minutes for a structured draft, 48x faster
- Target system analysis reduced from roughly 30 minutes to roughly two minutes per system, 15x faster, at approximately 85% accuracy before refinement
- Thousands of pages of higher-headquarters orders organized into usable structure in minutes during Intelligence Preparation of the Operational Environment
An Army all-source analyst on the evaluation: "I can spend my time doing critical thinking and analysis, rather than finding information and formatting a document."
Full results: Centurion at Scarlet Dragon 26-01
USSOCOM TE 26-02
At USSOCOM TE 26-02, Centurion X supported Military Decision Making Process workflows on government data in disconnected conditions, shifting to locally hosted models with no operator intervention when the network dropped. That deployment ran on HPE EL8000S edge hardware with a hardened Kubernetes distribution from Rancher Government Solutions.
How Centurion compares
Technical specifications
Centurion configurations meet the following environmental and compliance standards:
- MIL-STD-810G and MIL-STD-461 tested rugged case options (Centurion X)
- MIL-PRF-28800F Class 3 thermal compliance (EL8000t variant, Centurion X)
- Exceeds NEBS Level 3 and GR-3108 Class 1 (Centurion X)
- RAID5 storage resiliency available (Centurion L)
- TAA-compliant options (Centurion V, X, and L)
- HBOM and SBOM established for accreditation
Configuration-level specifications, including GPU counts, storage capacity, and power and thermal envelopes, are available through the Legion Intelligence sales team.
Procurement and support
Centurion is acquired as a single purchase covering hardware, software, models, and support in one line item. Pricing is fixed with no usage-based fees, and the bill of materials is known, which reduces supply-chain and cyber risk and supports a faster path to an authority to operate.
Centurion is awardable through the Tradewinds Marketplace, and available through direct award and sole source paths. Carahsoft is Legion's federal channel partner. Centurion is designed for system integrator delivery, with pre-integrated hardware, software, and support that SI teams can bring to proposals without building from scratch.
Lifecycle support is included with every purchase and covers system updates delivered when connected, technical support, and direct access to Legion's mission success team. After the first year, annual support and maintenance continues that coverage.
Related resources
- Centurion: The Legion Intelligence Platform Deployed to the Edge: The full technical whitepaper covering architecture, deployment philosophy, agent coordination, governance, and operation in DDIL conditions. Read this for the deepest technical treatment of how Centurion is built.
- Centurion at Scarlet Dragon 26-01: The complete field evaluation from the XVIII Airborne Corps G2, including methodology, measured results, and operator assessment. Read this for the evidence behind the performance figures on this page.
- What Is DDIL?: Defines Denied, Degraded, Intermittent, and Limited conditions, explains why cloud-dependent AI fails in them, and sets out the six requirements an edge system has to meet. This is the operating problem Centurion addresses.
- Accredited AI at the Edge: Covers secure deployment, accreditation, governance, and trusted AI operation in regulated environments, including the accreditation and procurement path.
- Agentic AI at the Tactical Edge: Legion Command Paper on doctrine and architecture for agentic systems in contested environments. Read this for the architectural argument rather than the product detail.
- Centurion product page: Product positioning, configurations, and field evaluation requests.

More About Centurion
Is Centurion the same as running Legion in the cloud?
The software is the same. Centurion is Legion, with the full capability set, Legion Packs, and the governance model that runs in Legion's cloud and on-premises deployments. What Centurion adds is delivery as an integrated system, with hardware from partners including Intel, NVIDIA, and HPE matched to the deployment profile. Nothing is removed at the edge.
Does Centurion require an internet connection?
No. Centurion operates fully disconnected using locally hosted models. Connectivity is used to reach frontier models and to synchronize state when policy permits, but it is never required for operation.
Can Centurion be deployed in an air-gapped environment?
Yes. Centurion supports fully air-gapped deployment with no external connectivity required, and is authorized from IL2 through IL6 including classified networks.
What happens to a workflow when the network drops mid-task?
The workflow continues. Centurion shifts from frontier models to locally hosted models automatically, with no operator intervention, and governance and audit remain in force throughout.
Does running local models mean running outdated models?
No. Disconnected nodes accept controlled model updates with validation and rollback, so forward deployments stay aligned with approved models as they are updated.
Can Centurion run models we have already approved?
Yes. Centurion supports any LLM, whether cloud-hosted, locally hosted, or fine-tuned, and new models are added without code changes.
How long does deployment take?
Field install is fast: at Scarlet Dragon 26-01, Centurion L was installed and operational in one day, with most users proficient within an hour. Full deployment, including system integration and approval processes, typically runs about six weeks. Exact timing depends on the environment and the accreditation path.
Does customer data leave the environment or train external models?
No. Customer data remains in the customer environment and is never used to train external models.
Who handles the accreditation work?
Accreditation is the customer's authorizing official's decision. Centurion supports it by shipping as a single system with an established HBOM and SBOM and a fixed assessment surface, which narrows the assessment surface compared with separately sourced components.
How many users does a Centurion node support?
Between 1 and 350 concurrent users depending on configuration, from Centurion III at the individual operator level to Centurion L at forward data center scale. Multiple nodes coordinate as a mesh for higher throughput.
When is Centurion not the right fit?
Centurion is the wrong tool where connectivity is stable and the work does not need to survive a dropped link. In garrison, enterprise analytics, and experimentation, cloud-hosted AI is the simpler choice. The deciding variable is whether the work has to continue when the network does not.
Which Legion Packs run on Centurion?
Any Pack. Packs are role-based capability bundles that run consistently across cloud, on-premises, and edge deployments. Centurion configuration affects scale, concurrency, and local model capacity rather than Pack availability, so Pack selection follows the role and the work. Packs with heavier local inference or transcription requirements perform best on Centurion X or L.