Governed AI inside Microsoft Word, for UK policing, justice and regulated organisations.
TOPAS for Word is an add-in for Microsoft Word. Your people draft, review and analyse documents in the Word they already use, and every request for AI help goes through HEX 165 first.
HEX 165 decides what information the AI may receive, which approved models may process it and when a person must approve the result. Each decision is recorded for audit.
Draft, review or analyse a document in your usual workflow, using the TOPAS panel inside Word.
HEX 165 decides what the AI may receive and which approved models may process it, before any model is called.
The controls applied and the decisions made are recorded, so you can show how AI was used on each document.
An illustrative task: reviewing a case summary with personal identifiers removed before AI processing.
Fictional details. The officer keeps authorised access to the original text, and the AI receives only what policy permits.
What a user does in Word, and what HEX 165 does at each step.
Work on the original text as usual. TOPAS marks sensitive details such as names and contact numbers so they are redacted before anything goes to AI.
Choose a task such as a review, a summary or a draft. The request goes to HEX 165 with your Microsoft sign-in, never straight to a model.
HEX 165 applies your policy, checks the budget and screens the document for hidden instructions before anything is sent.
The request goes only to an approved model in an approved region. If the destination does not match policy, nothing is sent.
Results come back as a preview with citations. Nothing goes into the document until someone has checked it, and changes that need approval need a second named person.
Decisions, approvals, model calls and costs go into a hash-chained audit log. Seeing redacted text needs a request and an approval, and the access expires.
Finds and protects sensitive information before any AI processing.
Limits processing to approved models in approved regions.
Sets what the AI may receive and do, using your roles and approval rules.
Keeps a traceable record of policy decisions and AI activity.
Cuts unnecessary tokens and measures what each AI task costs.
Each organisation gets its own HEX 165 environment, built from the same signed release. Your people, documents and identity stay in your Microsoft 365.
HEX 165 runs in your own cloud subscription in a UK region. Your team runs it, with support from Grace Blackwell.
A dedicated environment for your organisation in a UK data centre, under UK law, operated by Grace Blackwell staff.
The 14-day trial runs in Grace Blackwell's evaluation environment with synthetic or public content only. Nothing is installed on your devices: your IT team approves browser access to Word and the add-in. Sovereign Edge hosting for the most sensitive workloads is in progress.
TOPAS for Word runs on the TOPAS engine, Engine 04 of HEX 165. The Governance Engine decides and TOPAS executes: it applies each decision to the request, routes it to an approved model, controls tokens and cost, and records the evidence.
Select the lowest-cost compliant model path after policy, residency, SLA and accuracy checks.
Optimise repeated and structured context through cache alignment and Group of Tokens (GOT).
Validate compressed outputs against full-context reference results, with original source material retained in the Restore Vault.
Compare AI execution cost with equivalent human effort, using approved role, skill and rate benchmarks.
Record every material route, cost, compression, restore, approval and value decision to the hash-chained audit log.
A sidecar for applications, agents and delivery pipelines, bringing TOPAS policy, routing, budgets, GOT optimisation and audit telemetry into AI-enabled software through the HEX 165 API.
The Compliance Engine assures and the STPM Engine (in progress) protects. Both inform the Governance Engine, which decides. TOPAS executes each decision on every AI request.