Your plant shouldn't stop when the internet does.
Mentis deploys our own AI models inside your four walls and trains them on your data — so your operations keep running when connectivity drops, or when a frontier AI lab changes its terms or removes support.
You don't want to stop when the internet stops working — or when a frontier lab removes support.Why we built Mentis on-premise first
Three ways to run Mentis. One will fit you.
Pick by company size and how sensitive your data is. Not sure? Our team will help you decide in one call.
Mentis cloud
- Fully managed, monitored and updated by our team
- No servers or IT effort on your side
- Secure, isolated environment for your data
- Up and running in the shortest time
Choose this ifYou have a lean IT team, one or a few sites, and want results quickly.
Talk to us →Private cloud
- Deployed in your own cloud account
- Inside your network boundary and security policies
- Scales across plants, lines and regions
- Integrates with your enterprise IT, ERP and SSO
Choose this ifYou run many plants, already have a cloud estate and central IT governance.
Talk to us →On-premise
- Runs entirely on your own servers, inside your walls
- Works fully offline and in air-gapped networks
- Our own AI models, trained on your data, on your hardware
- No dependence on the internet or third-party AI labs
Choose this ifYour IP, regulation or critical infrastructure means data can never leave the site.
Talk to us →We deploy our own models — and train them on your data.
Generic AI doesn't know your machines, your procedures or your control code. Mentis models are trained on your operation, run where you choose, and never depend on someone else's API.
Deployed on your infrastructure
Models run on your servers or in your cloud account — not on a third-party AI service.
Trained on your data
Fine-tuned on your SOPs, machine signals, PLC code and expert know-how, so answers fit your plant.
Your data stays yours
Training data and trained models stay on your infrastructure, and are never used for other customers.
Upgrades on your schedule
No surprise model changes or shutdowns. You decide when to update.
The controls IT and OT expect.
Security and governance are part of the platform, not an add-on.
Single sign-on
SAML and OIDC with your existing identity provider.
Role-based permissions
Granular access by site, line, workflow and data type.
Immutable history
Every run, edit, approval and agent action is recorded.
Encryption everywhere
Data encrypted in transit and at rest.
Human approval gates
Configurable approvals for any agent action that changes a process.
Read-first connectivity
Plant-floor integrations start read-only; write access is explicit and scoped.
Enterprise questions, answered.
Our team can share detailed documentation as part of your security review.
Which deployment should we choose?
As a rule of thumb: Mentis cloud for firms of roughly 100–300 employees that want to move fast; private cloud for large, multi-site manufacturers with an existing cloud estate; on-premise when data security comes first — typically semiconductors, energy, defence and pharma. We'll confirm the right fit with you on a first call.
Can Mentis run without internet access?
Yes. On-premise deployments run fully offline, including in air-gapped networks. Your agents keep working even if your connection goes down.
What happens if a frontier AI lab changes or drops its models?
Nothing changes for you. Mentis runs our own models, so your agents don't depend on third-party AI services, their pricing or their support decisions.
Is our data used to train AI models?
Only your own. We train Mentis models on your data so they fit your operation. That data and the trained models stay on your infrastructure, and are never used to train models for other customers.
How do agents access plant-floor systems?
Through scoped connectors. Integrations start read-only, and any write access must be explicitly granted and approved.
Can we start in the cloud and move on-premise later?
Yes. Workflows, agents and trained models can move with you as your requirements change.
Put your first workflow on autopilot in weeks, not months.
Bring one process. We'll digitise it, deploy an agent on it, and show you the results on your own floor.