What people ask us most

Yes, always. LMbox runs on hardware you own, inside your own infrastructure. No request, no document, no conversation is ever sent to an external service. You can even operate fully disconnected from the internet (air-gap mode).

Yes for the AI, with an honest caveat on storage. LMbox provides sovereignty on 3 layers out of 4: indexing (local RAG), inference model (on the box), and audit log (verifiable SHA-256 chain). The 4th layer - where the source documents live - depends on the SaaS vendor you chose before LMbox. If your documents are on SharePoint Online, they stay in M365 (a perimeter you already accepted), and LMbox reads them locally without moving anything. For maximum sovereignty, pick the self-hosted variants of hybrid connectors (SharePoint Server, Confluence Data Center, GitLab self-managed, etc.) - LMbox supports them natively.

For business use cases - drafting, contract analysis, document search, summarisation - yes. LMbox runs the latest open models, which now rival the best cloud offerings on these tasks. And because LMbox knows your internal documents, procedures and knowledge base, its answers are often more accurate and more useful than a generic service that doesn't know you.

Licence-wise, it's unlimited - no per-user fee. Capacity depends on the form factor you pick. The LMbox Rack comfortably handles 50 to 150 users with dozens of simultaneous sessions.

LMbox connects to your directory (Active Directory, Azure AD, Okta, Google Workspace) for single sign-on. We index your knowledge sources - SharePoint, Google Drive, Confluence, intranet, document management - without ever moving your files.

Three days on site for installation and commissioning, plus one to two weeks to index your document sources. Your teams can already use LMbox for general purposes from day one.

All LLMs hallucinate - Mistral-7B as much as GPT-4. The honest answer isn't « our model is better », it's architectural. LMbox wires 4 deterministic layers on every agent: (A) a citation verifier calling Légifrance/EUR-Lex live on every reference emitted; (B) a runtime guard that cancels generation in 200 ms when a hallucination is detected; (C) a JSON Schema-validated output + repair loop; (D) a grounding layer that makes an invented source structurally impossible. See the detailed architecture.

We do. Updates, backups, monitoring, security: all handled inside the support contract. Your IT team has a single point of contact and a dashboard to monitor box health. No specific AI skills required in-house.

The box belongs to you. You keep the hardware, your data, your independence. No lock-in, no penalty. We believe trust must be re-earned every year.

Yes. Air-gap mode is available: no outbound connections, updates delivered physically, hardened audit logs. We support organisations in defence, hospital systems and sensitive public services.

A 30-minute discovery call to understand your context and priority use cases. If LMbox fits, we offer a paid 30-day on-site PoC with your real data - fully credited towards the contract if you go ahead.