An accounting and advisory firm runs its legal document automation on premises.

A GB300-class appliance in production and a GB10-class appliance for development, installed at the firm, run the Understand Tech platform locally. The first application automates legal documents; the firm holds its source code.

SectorAccounting, legal services and business advisory · several thousand business clients

StatusIn production · Appliances contracted mid-2026 after a six-month proof of concept; in-service date not yet published

EvidenceCustomer interview · Contract scope

Story
Stack
Load
Measurements

“Sending it to the cloud is not acceptable for our clients or for us.”Managing partner · customer interview recorded by Understand Tech

THE SETTING

Who uses it, and where.

Accounting, legal services and business advisory · several thousand business clients. Environment: office.

Users. About ten administrators during the proof of concept; end users across the firm. Concurrency and request volumes: not yet published.

THE STACK

What runs on the machine.

Understand Tech platform on the appliances: application builder, agents and applications, sovereign language model executed locally.
Quarterly platform releases installed on the appliances.
The legal-documentation application, built during the proof of concept, delivered as source code that the firm maintains and evolves itself.
Model IDs and runtime versions: not yet published.

Network posture. On-premise processing: client data does not leave the firm’s infrastructure. A hosted preview instance is used only to try new models before each quarterly platform release.

THE WORKLOADS

What it does, day to day.

Filling legal templates from structured data: articles of association, minutes, annexes.
Text that adapts to company type and variables; conditional clauses; multi-scenario generation.
Case summaries and fact sheets; compliance checks with alerts and suggested clause adjustments.
House-style formatting, Word and PDF export, integration with the CRM and document management.
Assistants for accounting and advisory work.

OUTCOME

What changed.

Reported · not yet measured

Confidentiality: client data is processed on site. Cost predictability: no per-request metering on the appliances. Autonomy: the firm holds the application source code. Measured productivity or cost figures: not yet published.

EVIDENCE & LIMITS

How to read this record.

The firm is not named at its request. The interview was recorded by Understand Tech; the machine list and scope come from the contract. Testimony, not a benchmark.

Measured figures (latency, throughput, concurrency, cost) are published only from a documented run on an identified machine. None is attached to this record yet.

LESSONS & FAILURES

What the team learned.

Not yet published. Three lessons from the project team, including what did not work, would complete this record.

RELATED RECORDS

Go deeper.

Setup record Checklist: test the full offline path Checklist: measure shared inference

Author: Understand Tech · reviewed 2026-09-16. Customers are not named at their request.