How it works

How TonsleyAI works

From understanding your AI opportunity to a fully deployed private AI platform — here’s how we get you there. We identify the highest-value use cases, assess readiness, and design a solution that delivers measurable ROI, before anything goes live.

The path

From first conversation to production — without wasted spend

Step 1 · Free

AI Reality Check

A free 30-minute call — whether AI is worth your time, and where to start.

Step 2

AI Opportunity Workshop

A half-day Sprint or full-day Strategy Day — a costed plan you could deploy, with the fee credited toward your build.

Step 3

Proof-of-Value

A working AI pilot in your own environment, on real data with real users.

Step 4

Deploy & scale

Production deployment inside your environment, with ongoing support.

How we work

Choose your path

AI Opportunity Workshop

Half-day Sprint or full-day Strategy Day. A ranked shortlist of opportunities, a plain-English safe-vs-private data map, and your best first use case scoped. We credit the full fee toward your build.

Proof-of-Value

A working AI pilot deployed in your own environment — real data, real users — so you prove the value before committing to production. Deliverable: a working pilot with documented ROI and a production roadmap.

Full platform deployment

From pilot to production, deployed in your secure environment — solution design, integration, and ongoing workshops, development and optimisation support. You’re never on your own.

Book a free AI Reality Check

Wondering what you would actually end up with? Two shapes, and the questions each one answers →

The landscape

How AI deployment modes compare on data exposure

A working framework for buyers — the six ways AI gets deployed today, and what each one costs you in data control.

Encryption covers data at rest and data in transit. AI adds a third state: data in use. A model cannot summarise a contract or check a report without processing what is inside it — so the question is not only where your data is stored, but who processes it and where. That is what the vendor-data-access column below is measuring.

AI deployment

Vendor data access

Best use case

Data exposure risk

Free public AIMeta AI, Grok, Claude, DeepSeek, Kimi, ChatGPT

Vendor data accessYes
(incl. model training)

Best use casePublic data & market research

Exposure riskExtreme

Paid public AIOpenAI ChatGPT, Google Gemini, Anthropic Claude

Vendor data accessYes
(model training opt-out)

Best use casePublic data & market research

Exposure riskVery high

Product AI (SaaS)Microsoft Copilot, Salesforce Agentforce, Atlassian Rovo, HubSpot Breeze

Vendor data accessYes

Best use caseNon-sensitive operational data

Exposure riskHigh

Enterprise AIOpenAI ChatGPT Enterprise, Claude Enterprise, Gemini Enterprise, Mistral Enterprise

Vendor data accessYes

Best use caseNon-sensitive operational data

Exposure riskMedium

Semi-private AIAzure AI Foundry, AWS Bedrock, Google Vertex AI, Open Telekom Cloud, Scaleway AI, STACKIT AI

Vendor data accessMaybe (see contract)

Best use caseNon-sensitive and internal data

Exposure riskLow

Private AITonsleyAI — your cloud or on-premises

Vendor data accessNo

Best use caseSensitive & proprietary data

Exposure riskNone

Vendor data-access behaviour as at July 2026, based on each vendor’s published terms. Terms change — check the current terms before relying on this for procurement.

A worked example. Microsoft publishes this openly, which is what makes it the clearest illustration available. Copilot’s Flex Routing allows large language model inference to run outside the EU Data Boundary at times of peak demand — in the United States, Canada or Australia — while data at rest stays inside it. For new EU and EFTA tenants created after 25 March 2026 it is on by default. That is a considered trade-off to keep a very large service responsive, not a lapse. The point is about defaults rather than vendors: choosing a major, convenient platform does not by itself settle where your data gets processed — someone still has to know, open the setting, and decide. The mirror-image question for an Australian organisation is whether its own vendor agreements carry a comparable exception, and whether anyone has checked. Microsoft’s documentation → (As at August 2026.)

Not sure which row is you? Answer five questions and find out → It runs in your browser — there is nothing to submit.

Deployment

Deploy your way — start low-cost, scale to air-gapped

Semi-Private

Your cloud tenant, with frontier models called through the hyperscaler.

UpfrontNo hardware

OngoingUsage-based

Most start here

AI-in-a-Box

Your premises, on an appliance that arrives pre-configured.

UpfrontThe appliance

OngoingNo usage fees

Private Cloud

Your cloud tenant, with the model running inside it.

UpfrontLow or none

OngoingCloud compute

On-Premises

Your premises, your hardware, air-gapped where required.

UpfrontYour hardware

OngoingNo usage fees

There is no single cheapest tier, which is why these are not ranked. The cloud tiers start cheaply and keep billing for as long as you run them; owned hardware costs more to stand up and then very little to run, so past enough volume and enough time it becomes the cheaper option. Which one wins depends on the horizon you are actually planning for — and the tier your data allows comes first either way.

Private Cloud

Your AWS, Azure or GCP tenant. Region-resident by default, full audit logging. Low or no upfront cost.

Semi-Private

Your cloud plus hyperscaler frontier models (Azure OpenAI, Bedrock, Vertex) — no upfront hardware and no model-size ceiling. The practical first step for most SMEs.

On-Premises

Air-gapped-capable, on your own hardware behind your firewall — model size scales with the hardware you provide.

AI-in-a-Box

A pre-configured appliance, up to 200B parameters (120B on a single node), operational in a day — on-prem without the setup.

No per-seat licensing. No platform usage fees for self-hosted deployments. One platform, your infrastructure — every deployment includes ongoing support.

Why TonsleyAI

Why organisations choose TonsleyAI

Complete data control

Your data never leaves your infrastructure — no third-party access, no cloud vendor lock-in. Built to be compliant and deployed inside your own environment, so control stays with you.

Integrated, not single-domain

Most private AI is single-domain — a chat window, a document search, one connected tool — and leaves the joining-up to you. TonsleyAI runs the work end to end on one workflow engine: controlled, auditable agents that read your documents, carry out multi-step jobs and write results back into the systems you already use, across 500+ integrations — all inside your own environment.

Built for growth

Start with a single use case; scale to hundreds of AI workflows. Every deployment includes ongoing workshops, development, and optimisation. You’re never on your own.

We map which data can go where

Knowing which data can safely reach which model is the hard part, and it is where most projects come unstuck. Every engagement starts with a plain-English map: what can use a public model, what belongs in your own tenant, and what should never leave your infrastructure. Then we build to that map.

Ready to put AI to work?

Start with a free 30-minute AI Reality Check — honest advice on whether private AI is worth your time, and where to begin.