Question 1

How is Leyoda different from ChatGPT Enterprise or Microsoft Copilot?

Accepted Answer

ChatGPT Enterprise and Copilot are cloud services — your documents are sent to Microsoft or OpenAI servers for processing. Leyoda installs an AI system inside your own office or data centre. Your prompts, documents and answers stay on the appliance — only anonymous system telemetry leaves (and even that can be disabled for fully air-gapped operation). They are complementary: use cloud AI for general work, Leyoda for the workflows that require strict data residency.

Question 2

What hardware do we need? Can you supply it?

Accepted Answer

A modern workstation or server with a consumer or professional GPU (RTX 3090 and up for most workflows). We can supply and configure the hardware as part of the setup, or install on hardware you already own. The hardware is always yours — not rented, not managed by us remotely.

Question 3

How long does installation take?

Accepted Answer

The proof-of-concept takes 7 business days from our scoping call to a live demo running on your data. A production deployment — full integration with your document management system, identity provider and backup infrastructure — typically takes 2–4 weeks depending on your environment.

Question 4

Can you connect to our existing document management system?

Accepted Answer

Yes. We integrate with file shares (Windows SMB, NFS), SharePoint On-Premises, SharePoint Online via Graph API (read-only), common DMS platforms, and SQL databases. We map your current systems on the scoping call and agree on connectors before we start building.

Question 5

Who owns the data and models?

Accepted Answer

You own everything — the hardware, the model weights, the embeddings, the logs and all output. We do not retain copies and we do not train on your data. Our control plane only sees anonymous system telemetry from the appliance (uptime, hardware health, model version) — never prompts, documents or model responses. We have no remote shell or admin access to your system unless you explicitly grant it for a support session.

Question 6

How is privacy ensured?

Accepted Answer

By architecture, not policy. The system runs on hardware inside your perimeter. Model weights are on encrypted local disk. Every query and document access is logged in an append-only audit trail you control. GDPR compliance follows from the on-premise design — prompts and documents never cross a jurisdictional boundary. Outbound internet is enabled by default (it's what lets our control plane receive system telemetry); you can disable it any time for fully air-gapped operation, which also stops telemetry to Leyoda.

Question 7

Does Leyoda see our prompts or data through the control plane?

Accepted Answer

No. Our control plane receives only anonymous system telemetry — appliance uptime, CPU/GPU load, disk health, model version, error counts. It does not see, log, or transmit prompts, documents, queries or model responses. Those stay on the appliance in your network. The control plane is read-only and observability-scoped, and you can disable even the telemetry channel for fully air-gapped deployments.

Understand on-prem AI, in nine short videos.

Intro

Ready to apply this to your team?