The commander interface, in plain language.
Ask the battlespace anything. CommanderNL™ translates intent into orchestrated agent action and returns ranked, traceable options that are auditable via hash-chained, append-only decision records, ML-DSA-signed in assured deployments. In configured air-gapped deployments, operational data and inference remain within the accredited customer boundary.
Talk to the Battlespace.
A classification-aware, sovereign-trained natural-language interface. Commanders speak in plain operational language; agents respond with ranked, traceable options. In configured air-gapped deployments, operational data and inference remain within the accredited customer boundary; update and support paths are documented separately.
- Classification-aware · honours caveats and need-to-know
- Sovereign-trained · weights you own and audit
- Air-gapped deployments keep operational data and inference inside your accredited boundary
- Human-in-command · guardrails on every action
Capabilities
Natural-language tasking
Multi-lingual operational language understanding tuned for military doctrine.
Sovereign-trained models
Weights you own, train, and audit, running in your accredited environment.
Classification-aware
Policy-based output filtering and release controls, tested against semantic and indirect information leakage.
Hard guardrails
Governed by a fail-closed model register (an unapproved model tag is refused, never inferred), a five-role RBAC capability matrix with hash-only token storage, and a supervisor whose quarantine powers always carry an actor and a justification. Oversight is enforced by the engine, not by a prompt: bounded, reversible automation runs human-on-the-loop, and every explicit approval gate is human-in-the-loop.
Traceable answers
Every recommendation is traceable to its source data, model and tool versions, policy gates, confidence outputs and operator actions.
Coalition releasability
Per-allied-node release controls embedded in the response itself.
Sovereign training & weights
Models are trained and fine-tuned on customer-controlled corpora in customer-controlled environments. Source records, model weights, prompts and local inference traces remain at the originating node. In federated deployments, policy-authorised queries, derived outputs and required provenance metadata may cross the federation boundary, and embeddings may cross only where expressly approved and documented in the deployment release policy. No telemetry, no opaque model calls, no shared inference pools.
Guardrails & operator authority
CommanderNL™ never executes lethal or irreversible actions on its own. Authority gates are defined in policy, enforced through signed policy artefacts and execution gates, and bound to qualified human operators. Every recommendation is auditable via hash-chained, append-only decision records, ML-DSA-signed in assured deployments.