Private AI that does your firm's real work.
Installed in your office. Owned by you.
We supply and install the hardware in your office, then set up private AI for legal, accounting, logistics and public-sector teams across Europe. It works on your real files, from contracts to client records, and the AI itself never calls out to anyone else's model. The system is yours to own or lease.
Starts with a call. Discovery is a separate, paid step.
Working with
Ask your documents. Watch the answer arrive.
A scripted preview of a real deployment: pick a document type, ask one of the prepared questions, and watch the answer come back with its source shown. Every document here is a real public record, open the original and check any answer yourself.
The work you can't put in the cloud is the work AI is now good at.
Most firms have kept AI away from client files, and they were right to. The Dutch and Belgian bars advise against it, and under the CLOUD Act a US provider can't promise your data stays in Europe. That's exactly the work where AI now saves the most time. This is how you do it without the risk.
Your best work is off-limits to public AI
Contracts, client records, deal files: the work AI would help with most is the work you can't paste into ChatGPT. The Dutch and Belgian bars now say so directly (NOvA, 2025; OVB and Avocats.be, 2025).
So the highest-value work stayed off AI
Public tools are useful for general drafting. They are the wrong place for privileged material, so the work that would benefit most has stayed off AI.
There is now a compliant way in
Run the model inside your own network and the bar's closed-environment condition is met, so privileged work is finally in scope. Firms already doing this report about 11 hours a week back per lawyer (RSGI, 2026).
Installed in your office, anywhere in Europe. Hardware we supply, yours to own or lease. Your network.
Which tool for which work?
General AI tools are genuinely good at general work. The question is what to do with the work that involves client files. That line decides which tool you use, and you don't have to pick a side.
Work that can leave the building
- Marketing text and blog posts
- Research on public sources
- Emails without client data
- Internal notes and brainstorming
ChatGPT, Copilot and Gemini handle this well. Keep using them.
Work that stays inside
- Contracts
- Client records
- Case files
- Tax returns and financials
- Deal documents
This is the work Leyoda does. It runs on a machine in your office, and the AI never calls out to anyone else's model.
You can verify all three.
- No connection to us or any outside AI provider, your files never leave your network.
- No US company holds your files, so no US court can demand them.
- Your data is never used to train any model, ours or anyone else's.
We didn't draw this line. The Dutch and Belgian bars did: no client data in public AI tools, confidential work only in a closed environment (NOvA, 2025; OVB and Avocats.be, 2025). We built for the inside of it. Most firms end up running both.
For firms that held AI back because the work was too sensitive to risk.
You kept client files out of public AI for good reasons. This is the way in that respects those reasons: the system runs in your office, on your network, and the AI never sends your work to anyone else's model.
Legal
Contracts, case files and privileged correspondence can't go to a US cloud, and the bars now say so.
Accounting
Client returns, audit files and financial records carry duties of confidentiality you can't outsource.
Logistics
Customs filings, supplier terms and trade data carry commercial risk if they leave your control.
Government
Citizen records, procurement and classified workflows can't sit on someone else's servers.
Five layers. All inside your building.
A server sits in your office and your documents stay on it. Here is what each of the five layers does, from your team's browser down to the machine itself.
Your team
Your staff open a browser inside your office and log in with their existing account. The interface is straightforward, nothing new to learn.
The moment you paste or upload, a copy exists on infrastructure you do not control, governed by terms you did not write.
- A copy now exists outside your network
- Retention follows their policy, and policies change
- Deletion is a promise you cannot audit
The file is read by models running on a machine you can physically point at. Pull the network cable and it still works.
- No copy ever exists outside your network
- Safe for privileged and regulated work
- Every action is logged, and you can audit the log
We walk through this with your team on the first call. If something does not fit your setup, we adjust before installation.
From a first call to a system running your work.
No prototype theatre. We look at how the work runs today, write down what it would take, and only then talk about hardware and price.
- Stage 1 of 4·Step 1
A call to see if it's a fit
A short, informal call. Tell us what's eating your week, and we'll talk through whether on-premise AI is worth exploring for you. No charge for this, and no obligation to go further.
NDA firstA mutual NDA is signed before you share any process detail. - Stage 2 of 4·Step 2
Discovery, and your AI readiness report
This is the paid part, fixed in scope and price before it starts. We sit with the people who actually do the work, map up to five candidate workflows, check which systems and data each one needs, and size the hardware the first one would take. It ends in a written report.
The report is yoursYou keep the report and can act on it with us or with anyone else. Finishing discovery obliges neither side to continue. - Stage 3 of 4·Step 3
Your machine, and what it costs
Discovery tells us which of the three system sizes you need: Starter, Team or Enterprise. You can spread that cost across your monthly rate, or pay it upfront and pay less each month. Whichever you choose is agreed in writing before anything is installed.
Priced from evidenceHardware is sized against your real document volumes and concurrency, not a guess. - Stage 4 of 4·Step 4
Installation and your first term
We install the machine in your office and put the first workflow into daily use. First contracts run three, six or twelve months, whichever suits how fast you want to move.
See if it's a fitRuns in your buildingThe system runs on hardware in your office, and the AI never calls out to anyone else's model.
What these workflows are worth, by sector.
Three worked examples, projected from representative workflows in each sector.
Projected estimates based on representative workflows, not measured results. Actual figures vary by team size, data volume and task complexity.
A 40-person law firm reviewing 300 NDAs a month by hand spends about 45 minutes per agreement, split across four associates.
On this workflow, private AI brings review time to roughly 19 minutes per document, about 7 hours a week back for the team.
A regional accounting office handling 180 tax returns a quarter is bottlenecked on data entry, where transcription errors creep in.
Automated extraction is projected to cut data-entry time by about 50% and ease the seasonal hiring need.
A logistics coordinator spending 2.5 hours a day on customs documentation covers 60–80 shipments manually.
With the documents handled on-premise, the same coordinator could clear roughly double the volume in under 30 minutes, freeing time for supplier work.
We are taking a small founding cohort of firms in legal, accounting and logistics. Early deployments receive founding pricing in exchange for product input.
Why firms are moving now
What's changed, and why it's worth your attention now.
Most firms have kept AI away from client files, and they were right to. The Dutch and Belgian bars advise against it, and under the CLOUD Act a US provider can't promise your data stays in Europe. But that's exactly the work where AI now saves the most time.
Compliance
€290m
Public AI is closed to your client work.
The Dutch and Belgian bars now tell members to keep client data out of public models, and the Dutch regulator fined Uber €290m for moving EU data to the US. For privileged work, the public US tools are off the table.
Capability
11 hrs/wk
AI has moved from drafting text to doing the work.
Firms that found a compliant way in are saving real time: legal teams report about 11 hours a week back per lawyer (RSGI, 2026), and most finance leaders now let AI handle part of the routine workload (AICPA, 2026).
Ownership
0 bytes
It runs in your office. Never calls out.
The AI sits on hardware in your building, so your files stay on your network, no US cloud, no provider that can be made to hand them over. It's the one setup the bars actually permit, and the reason a cautious firm can start now.
You don't have to be first. You do have to start before this becomes the baseline your clients expect.
See if it's a fitSee what it's worth to your firm.
Pick the work and how often you do it. We supply the time it saves, no email gate.
36h/week × €120 × 46 weeks
An estimate, not a quote. The value is real when that time goes back into billable or growth work, and we work out your real numbers during discovery.
You own the whole system.
We install it in your office and you run it. Buy the hardware outright or lease it, whichever suits you. The models and your data are yours either way, and the system keeps working whether or not we exist.
No subscription and no per-use bill. If you buy, the machine is yours outright. If you lease, you can buy it out or return it at the end of the term.
The hardware sits in your office
Installed like any other server. Buy it outright, or lease it and buy it out later. Either way, it runs inside your walls.
You own the models
Open-weight models run locally on your hardware. No licence that expires, no cloud dependency.
No format lock-in
Standard file formats, documented setup, the models and your data are yours. What keeps firms with us is the ongoing engineering, not a system you can't leave.
No usage pricing
Flat monthly fee for support. Run it as much as you need. No metering.
Three system sizes.
Start with the tier that fits your current workflows. You can expand later.
The figures below are what the system itself costs to buy and install. They are indicative, not a final quote. That cost is either folded into your monthly rate, or paid upfront in exchange for a lower monthly rate. Discovery decides which machine you actually need and which of the two suits you. We don't publish monthly figures yet, they aren't fixed.
Multiple workflows across a team, with role-based access and integration into your existing systems. For practices ready to move past one workflow and build a real internal capability.
- Mid-sized law firms
- Logistics companies
- Manufacturing SMEs
- Multi-office organisations
- Multi-document analysis
- Internal knowledge assistants
- Workflow routing and classification
- Department-wide AI search
- Dedicated AI workstation or server
- Multiple workflow automations
- Multi-user deployment
- Role-based access control
Not sure which option fits?
We size the installation based on your workflows, document volumes, and security requirements.
Move fast, and still pass procurement.
We package what early-stage teams actually need: a working AI prototype on your data, a compliance posture that survives enterprise security reviews, and pricing that respects your runway.
Prototype sprint
Fixed-scope 2–4 week engagement to ship a working AI prototype on your real data. One outcome, one price, no surprise scope.
Enterprise-ready from day one
GDPR and EU AI Act baseline, data-residency choices, model cards and DPIA templates, so your first enterprise customer’s procurement team says yes.
Startup-friendly pricing
Setup scoped to your size and stage, agreed directly with you rather than a fixed enterprise minimum. Clear path to add more workflows as you grow.
The five things firms check before they start.
Anything we haven't covered here, you can ask us directly on the call.
Cloud AI asks you to send your files out and trust whoever receives them. A firm that runs on confidential work can't do that, and shouldn't have to. So we built it the other way round: the AI moves into your office, and your files never move at all.
Tell us which work we should look at.
A few details so the first call is about your actual work, not a generic demo.
What's your industry?
Optional, helps us prepare.
Private AI, installed in your office, owned by you.
It starts with a short call to see if it's a fit, no charge, no obligation. If it makes sense, we move into a paid discovery phase: we sit with the people who do the work, map the candidate workflows, and hand you a written AI readiness report you keep either way. Only then do we talk hardware, monthly price and your first contract.