Platform

The whole platform, on your hardware.

One deployable unit carries the agents, the orchestration, the models, the data connections and the audit trail — a single binary that runs inside your perimeter. Cloud is not required; it is available only when you decide.

Architecture, end to end

Everything below the dashed boundary runs on infrastructure you control. The hyperscale cloud sits outside it — no egress by default, reached only on your routing decision.

Runtime & hardware
Rust-native single binary · memory-safe · standard 16-CPU / 128 GB server · no GPU farm · scales to a full cluster
Isolation
Air-gap compatible · zero internet dependency
Access control
RBAC for people · PBAC for agents · least privilege by default
Security
Zero-trust (mTLS) · guardrails · human-in-the-loop gates

Common questions

Can it run fully air-gapped, with no internet?

Yes. The platform is a single self-contained binary with zero internet dependency — no telemetry, no callbacks, no egress paths. The hyperscale cloud sits outside your boundary and is reached only if you explicitly route to it.

Which models can we use — and can we bring our own?

Open-weight, national or cloud models, switchable at any time, plus an embedded small LLM for fully local inference. MLOps lets you upload and serve your own ML models. There is no provider lock-in and no model menu set by a vendor.

What does deployment actually require?

A standard 16-CPU / 128 GB server — no GPU farm. It installs as one binary inside your network and scales to a full cluster as you grow. Most first deployments start as a proof of value on a single machine.

How is access controlled — for people and for agents?

Two layers work together. RBAC (role-based access control) scopes what each person may see and do. PBAC (policy-based access control) governs what each agent may do, read and call — through policies evaluated as it runs, because an autonomous agent needs its own least-privilege envelope, not a borrowed human role. Every user, agent and dataset is scoped to least privilege and segregated across domains.

How is it secured and audited?

Service-to-service traffic is zero-trust: every call is authenticated and encrypted with mutual TLS, so nothing inside the deployment is trusted implicitly. And every prompt, retrieval, tool call and decision is logged and replayable for review or compliance evidence.

Start a free trial

Runs on your hardware or your data centre · cloud-ready into your own tenancy.