Can we use only European or self-hosted AI models?

Yes. You decide which providers are configured. With self-hosted AI models via Ollama or LM Studio, no model call leaves your network.

Does our data leave the company?

Only if you decide so. Operaide runs on your own infrastructure, and for each step of an app you choose the AI model: a local model on your own hardware for data that must not leave, or a frontier model in your cloud tenant. An app can also use a local model to remove personal data before anything reaches a cloud model. Every run is traced, so you can see which data went where.

Does Operaide need an internet connection?

No. Operaide also runs air-gapped, as captive AI without any internet connection: with self-hosted AI models, and the aggregated usage metrics sent as a file instead of over a connection. In connected setups, Operaide sends these metrics to the Operaide audit service once a day.

How do we get updates?

New versions are released as Docker images. You pull them from the registry and decide when to install them. Signed images, so you can verify each one before installing, are available on request. In air-gapped networks, you bring the image in like any other software.

Who at Operaide can access our installation?

Nobody, unless you grant access. Operaide runs on your infrastructure under your administration.

What happens if we stop using Operaide?

Your apps are TypeScript code in your own repository, and your data sits in standard databases you can export. Nothing is locked in a vendor cloud.

Do you provide documentation for our compliance review?

Yes. We provide architecture and compliance documentation, including an orientation against the BSI AIC4 criteria, on request.

Sovereign AI.On your terms.

Operaide is software you run, not a cloud service you send data to. You decide where it runs, which AI models it uses and where your data lives, with no dependency on a US hyperscaler.

Request a demo For IT: installation and operation

Trusted by leading enterprises

Your infrastructure

You choose where it runs.

One Docker container that runs wherever containers run, even without internet access.

Your data centre
On-premises
In your own data centre, under your own operations team.
Your VPC, any provider
Cloud
In your VPC with any provider, or with a hosting partner of your choice.
No internet connection
Air-gapped
Captive AI with no internet connection at all, running on self-hosted AI models.

Your AI models

Use the AI models that fit your rules.

Operaide doesn't come with its own AI model. You connect the providers you trust, and every step of an app can use its own model, chosen for its task.

European providers

Use AI models from European providers such as Mistral.

Self-hosted AI models

Run open-source AI models on your own hardware via Ollama or LM Studio, with no data leaving your network.

Commercial providers

Use OpenAI, Anthropic, Azure, Google, Amazon Bedrock or OpenRouter where your policies allow it.

Recommended partnerOne EU-hosted gateway to 40+ AI models. Personal data is anonymised before it reaches any model.
MistralEuropean
OllamaSelf-hosted
LM StudioSelf-hosted
OpenAICommercial
AnthropicCommercial
Azure OpenAICommercial
Azure AICommercial
Google Generative AICommercial
Google Vertex AICommercial
Amazon BedrockCommercial
GroqCommercial
CerebrasCommercial
OpenRouterCommercial
CohereCommercial
DeepSeekCommercial

Your data

Your data stays where you put it.

We do not receive your business data. Here is exactly what leaves your installation, and what is stored only there.

Stored prompts, AI model responses and tracesStays
Documents and filesStays
Agent inputs, outputs and intermediate resultsStays
Per-event user activityStays
Your credentialsStays
Calls to the AI providers you configure, with the prompt and the data it contains, or none at all if you use self-hosted AI modelsLeaves
Calls to the external systems you connectLeaves
Aggregated usage metrics, once a day, to the Operaide audit service: user and run counts, token totals per AI model, storage size, platform version. Air-gapped, you send them as a file insteadLeaves

Controlled data flows

Every way out is on one list.

Admins set up every connection once. Apps built on the Reaktor framework use only these, so agents can't open channels of their own.

Add connection

Connections6 configured · managed by adminsApps use only these

AllAI providersAPIsMCPSign-on

NameStatus

MistralAI provider · 6 appsActive
Azure OpenAIAI provider · 3 appsActive
OllamaAI provider · on-premises · 4 appsActive
ERP systemREST API · 2 appsActive
Microsoft Entra IDSingle sign-on · All appsActive
Document archiveMCP server · 1 appActive

No lock-in

You can always take it with you.

Sovereignty also means being able to change course. Nothing in Operaide traps your apps or your data.

Apps are your code

TypeScript in your own Git repository, on standard technology: a Docker container, npm packages and SQLite. No proprietary runtime.

Your data, exportable

Every app keeps its data in its own database, and full-instance backups restore on any host.

No AI model lock-in

Providers are configured, not built in. Switching AI models doesn't mean rewriting apps.

Regulatory context

How Operaide fits your compliance framework.

Operaide supports you in meeting your compliance obligations. Here is what each regulation means for an Operaide deployment, and where the platform helps.

Shared responsibility

Who takes care of what.

A compliant deployment involves three parties: the platform, the infrastructure operator and you.

Responsibility	Operaide platform	Infrastructure operator	Your organisation
Sign-in, roles and permissions
Sign-in, roles and permissions		Not responsible	Assigns roles
Separation of organisations and apps
Separation of organisations and apps		Not responsible	Not responsible
Traceability and usage records
Traceability and usage records		Not responsible	Not responsible
Controlled external connections
Controlled external connections		Not responsible	Configures them
TLS, network, firewalls
TLS, network, firewalls	Not responsible		Not responsible
Disk and database encryption
Disk and database encryption	Not responsible		Not responsible
Backups and disaster recovery
Backups and disaster recovery	Tools included		Not responsible
Platform updates
Platform updates	Releases new versions, signed on request	Installs them	Decides when
Infrastructure certifications (ISO 27001, C5)
Infrastructure certifications (ISO 27001, C5)	Not responsible		Not responsible
Use-case classification (EU AI Act)
Use-case classification (EU AI Act)	Audits	Not responsible
Human oversight in workflows
Human oversight in workflows	Building blocks	Not responsible	Designs it
Data retention and long-term archiving
Data retention and long-term archiving	Export	Archive	Defines policy

FAQ

Sovereignty questions

Get started with Operaide.

We'll show you how Operaide fits your infrastructure, your AI model strategy and your compliance requirements.

Request a demo