For UK policing, justice and regulated organisations
Work on documents with AI in the Word your people already use. HEX 165 decides what information the AI may see and which approved models may process it, and records every decision for audit.
HEX 165 is the governance platform that TOPAS for Word runs on. It runs in your own UK cloud subscription or in a dedicated Grace Blackwell environment in a UK data centre.
The Sovereign Edge IDN (Inference Distribution Network) is our air-gapped hosting for the most sensitive AI workloads, now in progress.
The governance platform for AI. Four engines cover policy, compliance, token cost and runtime security, which is in progress. Every AI request is checked against policy, sent only to an approved model and priced, with a hash-chained audit log generated by design.
HEX 165 moves AI governance below the application layer and into the execution path itself. It controls what agents are allowed to do, what context reaches the model, how tokens are routed and optimised, when human approval is required, and what evidence is retained for audit, compliance and operational review.
Explore HEX 165A planned sovereign hosting environment with the Inference Distribution Network (IDN) at its core. Five platform layers are designed to operate inside an approved UK / EU region, from sovereign connected environments to fully air-gapped operation for the most sensitive workloads.
The design keeps data, models, prompts, logs, telemetry and encryption keys in region. In air-gap mode, inbound and outbound connectivity is disabled, with ingress and egress controlled through approved transfer procedures. It is not available yet.
Explore the Sovereign Edge IDNHEX 165 governs more than agent behaviour. It governs the policy they operate under, the compliance obligations they must satisfy, the runtime environment they execute in, and the token cost they generate.
The four engines work as one control plane. Compliance assures and STPM protects, and both inform Governance, which decides. TOPAS executes each decision on every AI request. Each engine is designed to be deterministic, auditable and repeatable, so the same input produces the same authorisation, finding, risk signal or routing decision.
Policy-as-code for agentic AI. Every agent call is checked before it runs, against identity, scope, clearance band, model permissions, budget and approval rules.
It enforces policy at machine speed, pins model versions to signed manifests, enables rollback when behaviour changes, routes human approvals where required, and records every decision to the hash-chained audit log.
Deterministic compliance assessment for agentic AI. One scan evaluates agent code, architecture and documentation across all 16 supported regulatory, security and assurance frameworks, including the EU AI Act, DORA, GDPR and NIST AI RMF.
It surfaces regulatory gaps, cross-framework obligations and agentic risk flags that standard legal checklists often miss. Every finding is traceable to the relevant framework, article, control, paragraph or evidence requirement, producing audit-ready evidence without probabilistic interpretation.
Sovereign Threat Protection and Monitoring. Kernel-level eBPF telemetry for sensitive AI and data environments, minimised and anonymised at source before export.
The STPM Engine is designed to detect abnormal runtime behaviour inside the approved UK / EU region, protect sensitive file paths, identities, commands and network context, and route high-impact response or evidence reveal through human-authorised controls.
It is designed to align with NCSC guidance and uses open standards including OpenTelemetry, OCSF, Sigma and OPA/Rego, so runtime protection does not turn telemetry into a second data leak.
Token, Optimisation, Policy, Auditability and Security. TOPAS applies governance to every AI request: it protects sensitive data, routes to approved models, controls tokens and records the evidence. TOPAS for Word brings it into Microsoft Word.
TOPAS measures cost per completed task, not just cost per token. It applies sovereignty-aware routing, optimises context layout, reduces repeated or structured payloads through GOT (Group of Tokens), and supports reversible compression with restore and validation controls.
Explore TOPAS for WordFrom deployment to governance, token optimisation, runtime protection and reporting, the Grace Blackwell Platform creates a single source of truth for agentic AI operations.
Deploy HEX 165 in your own UK cloud subscription or in a dedicated Grace Blackwell environment in a UK data centre. Each agent is registered with a defined identity, scope, clearance band and budget from the start, creating the control baseline for everything that follows.
The Governance Engine evaluates every agent call against active policy before it runs. Requests that fall outside the approved scope, clearance, model permissions, or budget fail-safe by design. Where policy requires human oversight, approvals are routed to authorised reviewers and recorded in the hash-chained audit log.
The STPM Engine will capture minimised and anonymised runtime telemetry from sensitive AI and data workloads. Autonomous detection will identify abnormal behaviour inside the approved UK / EU region, while high-impact response, evidence reveal or de-anonymisation is routed through human-authorised controls.
The Compliance Engine reads the audit log, scans agent code, architecture and documentation, and evaluates the AI estate against the active regulatory and assurance frameworks. Findings are produced deterministically, mapped to the relevant framework obligations, and supported by evidence from the same hash-chained audit log used for governance, reporting and assurance.
Reports per engine produce real-time dashboards and audit log exports, providing stakeholders with a single source-of-truth evidence base. Boards, regulators, internal audit, and external assurance teams can review decisions from a hash-chained audit log created during the platform's runtime rather than reconstructed after an incident.
HEX 165 runs in your own UK cloud subscription or in a dedicated Grace Blackwell environment in a UK data centre, under UK law. Requests go only to approved models in approved regions. TOPAS for Word works inside your Microsoft 365, under your Microsoft terms. Air-gapped hosting through the Sovereign Edge IDN is in progress.
Policy and regulatory rules run on deterministic engines. No AI interprets law or internal policy. The same input always produces the same authorisation or finding, auditable rule by rule.
Every policy decision, approval and model call is recorded in a hash-chained audit log as part of normal operation, and the chain can be checked by an external auditor. Anchoring the chain to write-once storage is in progress.
OPA and Rego for policy and signed manifests for models, with OTLP, OCSF and Sigma planned for runtime security as the STPM Engine arrives. Open at the protocol layer, sealed at the jurisdictional boundary.