ConfidentialMind
ConfidentialMind is a self-hosted AI platform that runs large language models, RAG services and agents inside your own infrastructure, whether on-premises, air-gapped or private cloud, exposing them through OpenAI-compatible endpoints so no data leaves your environment.
What is ConfidentialMind?
ConfidentialMind is a self-hosted generative AI platform published by Confidentialmind Oy, a Finnish company based in Espoo and founded in 2023. It is a Kubernetes-based set of containerised services that runs large language models and AI systems entirely inside an organisation's own infrastructure, whether on-premises, air-gapped, in a private cloud or in a VPC. Standard Kubernetes components are wired together by a proprietary layer that adds the authentication, security controls and management production AI needs. The documented release at the time of review is version 3.1.0.
From a single portal, administrators deploy models, connect external providers and stand up AI systems over their own data. A Model Gateway presents everything behind one OpenAI-compatible base URL, so existing clients work by changing an endpoint and passing a platform API key, and aliases let a public model name be repointed later without touching client code. Retrieval-augmented generation is a deployable service with its own API, a managed vector database and optional completion, reranker and OCR models; it reads documents straight from S3 or Azure Blob Storage on a synchronisation schedule. Agents, tool calls, MCP integrations and coding agents sit alongside it.
Governance is treated as a first-class concern. Multitenancy separates users into tenants and sub-groups, with view or admin permissions granted per service across three levels of user. A GPU dashboard shows VRAM per node and per card and which models occupy them, with both NVIDIA and AMD supported. Every platform action is logged in an OpenTelemetry-compatible format and usage metrics are kept for compliance, next to an observability layer for debugging agent behaviour.
Four audiences are addressed explicitly: enterprises, public sector bodies, universities and hosting providers, the last of which can resell a multi-tenant inference cloud with billing integrations and revenue-share licensing. Named references include Elastx, Centria University of Applied Sciences and RAIN.global. Microsoft Azure and AWS are supported today, with GCP and IBM Cloud announced. The pitch is sovereignty rather than raw model performance: no data leaves your environment.
What it does
- Deploy open-source large language models from a self-service portal
- Expose OpenAI-compatible chat completion, embedding and responses endpoints
- Build and deploy a RAG service over your own documents from a template
- Run agents, tool calls and MCP integrations inside your own perimeter
- Separate teams or customers into tenants with per-service access rights
- Connect existing data sources such as S3, Azure Blob Storage and databases
- Monitor and allocate every GPU in the organisation from a single interface
When to use ConfidentialMind / When not to
A quick filter to help you decide if ConfidentialMind is the right fit.
When to use ConfidentialMind
- Infrastructure and platform teams that must serve internal AI requests without sending data to a public cloud API
- Public sector bodies handling citizen, tax or judicial records under national data security rules
- Universities giving researchers, faculty and students governed access to generative AI over sensitive research data
- Hosting providers who want to sell a multi-tenant AI platform on top of the GPUs they already operate
- Organisations running air-gapped or multi-location deployments that need one platform across every environment
When not to use ConfidentialMind
- Individuals and small teams: the cheapest plan is 500 EUR per month and deploying in your own environment starts at 2,000 EUR
- Anyone looking for a ready-to-use AI assistant rather than a platform to install on Kubernetes with GPUs
- Users who need a mobile application: no iOS or Android app exists
- Teams that need an interface or documentation in a language other than English
- Buyers who require published terms of service or a security certification such as ISO 27001 before signing
How to use ConfidentialMind
A typical end-to-end flow, from setup to results.
- Request a demo or contact sales: there is no self-service sign-up, every entry point goes through the vendor
- Optionally start on CM Test, an isolated tenant in ConfidentialMind's own private test cloud, before committing to your own environment
- Check the hardware prerequisites, in particular NVIDIA or AMD GPUs and the documented VRAM requirements
- Prepare or provision the Kubernetes environment: the installer can create the cluster itself on existing virtual machines or bare metal
- Install the platform with the single installation tool, following the K3s, OpenShift/OKD or air-gapped guide as appropriate
- Take the onboarding sessions provided for platform administrators and for the developers who will build the AI systems
- From the admin portal, deploy models and services, then create tenants, sub-groups and per-service permissions
- Deploy a RAG service from a template and attach its data sources, such as an S3 bucket or an Azure Blob container
- Create an API key in the portal and point an existing OpenAI client at the gateway base URL to start calling models
- Manage your own upgrade schedule, staying within the two most recent major versions as the vendor requires
Pros & Cons
Pros
- Data stays inside your own environment, including in fully air-gapped deployments
- OpenAI compatibility means existing clients and code work by changing a base URL
- Covers the whole chain, inference, RAG, agents, vector database, observability and multitenancy, instead of assembling separate tools
- Native multitenancy lets one platform serve several teams, departments or paying customers on shared GPUs
- Both NVIDIA and AMD GPUs are supported, which avoids locking into a single hardware vendor
- Entry prices are published, which is uncommon in this segment, and the upgrade schedule stays in the customer's hands
- Publicly named references across hosting, higher education and telecommunications
Cons
- No terms of service are published anywhere on the site, which is unusual for a subscription of this size
- No security certification is claimed, neither ISO 27001 nor SOC 2, although security is the central promise
- No free plan and no announced free trial: the cheapest tier is the vendor-hosted test cloud at 500 EUR per month
- Published prices are starting points only, with the real invoice depending on the global deployment footprint
- Running the platform requires Kubernetes and GPUs, so the operating cost and skills are far from negligible
- No subprocessor list is published and the privacy policy carries no effective date or version number
- Interface and documentation are English only, and there is no mobile application
Pricing & Plans
There is no free plan and no announced free trial. Three subscriptions are published, each quoted as a starting price. The entry point, CM Test, costs 500 EUR per month and gives access to ConfidentialMind's own private test cloud rather than to your infrastructure. Deploying the platform in your own environment starts at 2,000 EUR per month, as does the CM Hosting plan intended for service providers. Additional clusters are added to the infrastructure total and billed on the global deployment footprint, while the fully managed service and professional services are charged separately. Every plan is contracted through the sales team.
- access to the private test cloud
- an isolated tenant
- the full platform experience on modern GPU infrastructure and one business day support
- the full platform in your own IT environment
- product updates
- features and bug fixes as they ship
- same day support via Slack
- Teams or email
- and a fully managed service available for an extra fee
- the full platform to run your own inference neocloud
- multi-tenant separation of customers on shared GPU infrastructure
- billing integrations into your own system
- same day support and revenue share licensing available
Data, GDPR & hosting
A consolidated view of how ConfidentialMind handles your data.
GDPR overview
GDPR implementation is concrete on the vendor's own processing. The privacy policy maps an Article 6 legal basis to each purpose, consent, contractual necessity, legal obligation and legitimate interest, and lists eight data subject rights: access, rectification, erasure, objection, restriction, withdrawal of consent, portability and complaint to a supervisory authority. The data protection contact is info@confidentialmind.com. No data protection officer is named and no Article 27 representative is designated, which is consistent with an establishment in Finland. The public sector pages state the platform helps agencies meet GDPR and national data security regulations. Two gaps deserve mention: the privacy policy carries no effective date or version number and may be amended at any time, and no security certification such as ISO 27001 or SOC 2 is claimed anywhere on the site.
Who owns the data?
The platform is installed in your own environment, so the data it processes never reaches ConfidentialMind. The privacy policy says so explicitly: because the software operates on your premises or in your private cloud, you are the processor for everything handled through it. ConfidentialMind acts as data controller only for the personal data whose purposes it determines itself, essentially contact, order, billing and browsing data gathered through the website and the commercial relationship; where another party sets the purposes, it acts as a processor on that party's instructions. No terms of service are published, so no separate contractual clause on ownership of customer content could be examined.
Reuse rights
Because the software runs inside your infrastructure, ConfidentialMind never receives the content you process through it and places no reuse restriction on it. What the privacy policy does describe is the vendor's own processing: contact details, order and customer records, financial data, account credentials, browsing and device data, used to perform contracts, send product information and marketing, and meet tax, commercial and export obligations. The legal bases cited are consent, contractual necessity, legal obligation and legitimate interest. Service providers are bound by contracts including data processing agreements, although no subprocessor list is published. Transfers outside the EEA are possible and described only as compliant with EU law, with no mechanism named. Training AI models on customer data is never discussed.
Data retention & training
Hosting summary
The question works differently here than for a normal SaaS tool. ConfidentialMind is self-hosted: the platform runs inside the customer's own infrastructure, on-premises, air-gapped, in a private cloud or a VPC, so the vendor hosts none of the data processed through it and no hosting jurisdiction can be attributed to it. Microsoft Azure and AWS are supported as installation targets, with GCP and IBM Cloud announced, and the installer can also provision Kubernetes on existing virtual machines or bare metal. One exception matters: the CM Test plan runs in ConfidentialMind's own private test cloud, whose location is not published. For its own corporate and website data, the vendor states that personal data may be transferred outside the EEA in compliance with EU law, without naming a country or a transfer mechanism. The marketing website itself is served through a content delivery network, which says nothing about where the platform or its data live.
Things to keep in mind
Risks and trade-offs to weigh before adopting ConfidentialMind.
- No terms of service are published: an engagement starting at 2,000 EUR per month would be signed without any publicly available contractual conditions, so ask for them in writing
- Security is the central promise but no certification backs it: no ISO 27001, no SOC 2, no audit report, and no security or trust page exists on the site
- The published prices are starting points only, and 500 EUR per month buys the vendor's hosted test cloud rather than a deployment in your own environment, which starts four times higher
- Self-hosting moves the responsibility to you: once the platform runs in your environment, model governance, access control, logging and incident handling become your team's duty, not the vendor's
- No subprocessor list is published and the privacy policy carries no effective date or version, so what the vendor processes about your staff can change without a visible trace
- Because everything runs internally, users may treat model output as pre-approved by the organisation; keeping data in-house says nothing about whether an answer is correct, and human review of AI output remains necessary
- Small signs of an unfinished public presence deserve attention: the case studies page still carries its template instructions and the Twitter link on the contact page leads to a dead account
Setup & Integrations
Technical difficulty
High for the team operating the platform, low for the developers consuming it. Installation targets a Kubernetes environment with NVIDIA or AMD GPUs; a single installation tool can create the cluster itself on existing virtual machines or bare metal, with dedicated guides for K3s, OpenShift and OKD prerequisites and air-gapped deployment. Onboarding is provided for administrators and developers, and a fully managed service is available for an extra fee. Application integration, by contrast, is trivial: create an API key and point an existing OpenAI client at the gateway URL.
Deployment
Integrations
Behind ConfidentialMind
Social
Resources
All the official URLs gathered for verification and reference.
Frequently asked questions
Where does the data processed by ConfidentialMind actually go?
Can I try ConfidentialMind for free?
How much does it cost to run it in my own environment?
Does ConfidentialMind provide an API?
Which infrastructure and hardware does it need?
Which clouds are supported?
Which models can I run on it?
Is a mobile application available?
Are terms of service and security certifications published?
Who is behind the product?
Should you pick ConfidentialMind?
ConfidentialMind answers a narrow but real question: how does an organisation get a modern generative AI stack without letting its data leave the building? The answer it proposes is complete rather than clever. Inference, retrieval-augmented generation, agents, a managed vector database, multitenancy, GPU allocation and OpenTelemetry-compatible logging all arrive in one Kubernetes-based package, and the OpenAI-compatible gateway means existing code moves over by changing a base URL. For an infrastructure team that would otherwise assemble half a dozen projects and maintain them, that consolidation is the product.
The technical evidence is solid. The documentation is detailed and versioned, covering air-gapped installation, OpenShift prerequisites, VRAM requirements and tenant governance, and the customer references are named rather than anonymous: a Swedish hosting provider, a Finnish university of applied sciences and a telecommunications company. Supporting both NVIDIA and AMD, and letting the customer own the upgrade schedule, are the choices of a vendor that expects to sell to people who run their own infrastructure.
The commercial and legal side is thinner. No terms of service are published anywhere on the site, and no security certification is claimed, which sits awkwardly beside a product sold on security. The privacy policy has no effective date, no subprocessor list accompanies it, and the case studies page is still an unedited template. Prices are published, which is welcome, but they are starting points: 500 EUR per month buys the vendor's test cloud, not your own deployment, which begins at 2,000 EUR and scales with your global footprint.
For a buyer whose data cannot go to a public API, this is a serious candidate. Ask for the contract terms, the subprocessor list and any audit evidence before signing, since none of it is public today.
- Choosing a selection results in a full page refresh.
- Opens in a new window.