AI Solution
Claude with EU data residency and full control
Your people want Claude. Your security team wants to know where the data goes. Claude Desktop in third-party (3P) mode answers both: Chat, Cowork and Code run on your laptops, and the model runs at the cloud provider and in the region you choose, including EU regions on Amazon Bedrock and Google Cloud. We design and roll out that setup so that you decide the region, the exit point, who can sign in and what gets logged. That control is what we mean by sovereignty.
What is Claude on 3P?
Claude on 3P is Claude Desktop in third-party mode. Chat, Cowork and Code work as usual, but every model request goes to an inference provider your organization chooses, such as Amazon Bedrock, Google Cloud or your own gateway, instead of Anthropic’s API. Conversations stay on the user’s device.
Anthropic built it for organizations whose security, regulatory or contractual rules prevent them from sending data through claude.ai or the Claude API. Pricing is token-based and billed by your cloud provider, with no seat licence. Settings are usually pushed through your device management tool, such as Intune or Jamf.
Where the model runs depends on the provider you pick.
Amazon Bedrock. AWS runs the model and processes the data in the region you select. EU regions are available.
Google Cloud. Google runs the model and processes the data in the region you select. EU regions are available. This is the service many people know as Vertex AI; Anthropic’s documentation calls it Agent Platform.
Microsoft Foundry. Anthropic operates the models, on Azure infrastructure or its own, and acts as an independent processor. Anthropic documents only global and US deployment types, so Foundry is not a route to EU residency today.
Your own gateway. A gateway you operate, such as LiteLLM, sits in front of one of the providers above. Data handling follows whatever it routes to.
Anthropic’s API. It works in 3P mode, but Anthropic then processes the data under your agreement with it, and inference cannot be pinned to the EU.
The app sends prompts, responses, files and tool outputs only to the endpoint you configure. It still contacts Anthropic to download its sandbox components when a session starts. Crash reports, usage analytics, update checks and the model catalog are separate, carry no prompt or response content, and can each be switched off in the managed settings.
Can we get EU data residency for Claude?
Yes, if Claude runs through Amazon Bedrock or Google Cloud with an EU region selected. Anthropic’s own API cannot do this: its inference location setting accepts only global or US, and Anthropic stores commercial data in the US. EU residency therefore means using a cloud provider as the route to the model.
Three settings decide where your data actually lives.
The region. On Bedrock, use an EU regional inference profile and avoid the global one. A global profile can send requests to any region that has capacity. A regional profile keeps them in the geography, and it is currently priced about 10 percent above global routing. On Google Cloud, choose an EU region for the endpoint.
The device. Conversation history is stored on the user’s laptop, so residency follows the device as well as the cloud region. A laptop that leaves the EU takes its history with it. The app does not encrypt the session record, which makes full disk encryption on the laptop part of the design.
What still reaches Anthropic. With telemetry, update checks and the model catalog switched off, the only traffic left to Anthropic is the download of the sandbox components. Tool servers you connect are separate destinations, and each one needs its own review.
Where have we done this?
A European energy company wanted Claude Desktop for its business teams. Its rules were strict: conversation data stays in the EU, none of it goes to the model vendor, and every request passes through the company’s own gateway.
We designed and delivered the rollout on the company’s managed laptops. Claude Desktop runs in 3P mode and reaches Claude, hosted by a major cloud provider in an EU region, through that gateway. People sign in with their company account and receive short-lived access. Membership of one access group decides who gets the app and its locked settings. Budgets per person keep spending predictable, and usage shows up in the monitoring platform the company already had.
The gateway behind it serves more than 500 people across more than 100 business units. The design went through the company’s internal security and data protection reviews before go-live.
We use the same pattern inside Docktape for our own sensitive work, so we run day to day what we build for clients.
Pain points
We can help you if you face:
- People already using Claude on personal accounts, with company data
- A rule that conversation data has to stay in the EU
- A security team asking where prompts go, and nobody able to answer
- AI spend spread across credit cards and teams, with no limits
- A managed laptop fleet where every app must be packaged and locked down
- A Claude pilot that works for ten people and has no path to five hundred
Our offering
Our offering includes
Want Claude on your laptops with the data staying in the EU? Book a free consultation and we will map your route, region and rollout.
FAQ
Your questions answered
Arnold Gálovics
CEO
On Amazon Bedrock and Google Cloud, no conversation data does. Anthropic’s documentation states that Claude Desktop on 3P does not send user data, prompts or completions to Anthropic on those two providers. The app does contact Anthropic to download its sandbox components. By default it also sends crash reports and usage analytics, which contain no prompt or response content and can be switched off. Microsoft Foundry is different, because Anthropic operates the models there.
No setup makes a tool GDPR compliant by itself, because compliance depends on what you use it for. What this setup changes is the set of facts your data protection officer works with: processing in an EU region, your cloud provider as the processor under your agreement with it, history on your own devices and logs you control. We document those facts for the assessment. The legal judgment stays with your team.
Most of it can. Sign-in with Entra ID, device rollout with Intune and the gateway all fit a Microsoft environment. The exception is the model. Claude on Microsoft Foundry is operated by Anthropic, which documents only global and US deployment options for it. For EU residency we send the model traffic to Amazon Bedrock or Google Cloud, and nothing else has to move there.
Laptops can connect directly, and for a small pilot that is fine. Past the pilot we recommend a gateway. It gives you one exit point, sign-in through your identity provider, budgets that are enforced and one place to log. Without it, those controls sit on each laptop or in each cloud account.
There is no seat licence. You pay your cloud provider for the tokens people use. Someone who asks a few questions a day costs little, and someone running long Cowork or Code sessions costs more. On Bedrock, keeping requests inside the EU is currently priced about 10 percent above global routing. We set budgets per person and per team so the monthly total has a ceiling, and we size them with you from pilot usage.
Mobile and web access, sharing conversations between users, voice mode and sessions that run in Anthropic’s cloud are not available. Connectors that Anthropic hosts itself are not available either. You connect your own. Web search on Amazon Bedrock needs a search provider that you configure. Chat, Cowork, Code, projects, artifacts, memory, scheduled tasks, skills and plugins all work.
Windows 10 version 2004 or later, or Windows 11, on x64 or Arm64. Cowork does its work inside a small virtual machine, so the laptop needs hardware virtualization and the Virtual Machine Platform feature switched on, and the app has to be installed from the MSIX package. Anthropic publishes a readiness check that tests a device before rollout. On virtual desktops, the hosting platform has to allow nested virtualization. Macs need macOS 14 or later.
The app, its settings and the user’s access come back by themselves, because they are assigned by group. Conversation history lives on the device, so it has to be moved. Anthropic supports an export and import that you can switch on, or an administrator can copy the data folder. The export file is not encrypted. Treat it like any other sensitive file. If the old laptop is lost and nothing was exported, that history is gone.
AWS, Google Cloud, Azure and Oracle Cloud. The gateway, sign-in and monitoring can run on any of them, and we hold certifications on AWS, Google Cloud and Oracle Cloud. The Claude models themselves are hosted by Amazon Bedrock, Google Cloud and Microsoft Foundry, so on other clouds the gateway forwards requests to one of those.
Yes. Claude Desktop on 3P supports connectors built on the Model Context Protocol (MCP), which is how Claude calls a tool, and administrators decide which ones are available. We go a step further and publish your internal APIs as tools behind the same gateway. Each call is made as the signed-in person, so the system that owns the data still decides what that person may see.
ARC, our Private AI Platform, is a browser-based AI workspace for the whole company, with several model vendors behind one interface. Claude on 3P is the Claude desktop app, for people who need an agent that works with their files and tools. Both use the same governed gateway, and at the energy company above they run side by side.
Suggested studies
How we work with others
Energy
Claude Desktop with EU data residency for a European Energy Company
We rolled out Claude Desktop to the business teams of a European energy company, with the model in an EU region, every request passing through the company's own gateway and no conversation data going to the model vendor.
Show more
Energy
Private AI Platform for a European Energy Company
We built ARC, a private AI platform that gives the employees of a European energy company governed access to modern AI models, with no data retention by third-party providers and centralized cost visibility.
Show more
Finance
Cloud migration of a Bank’s data platform
Show more
Finance
Security Assessment for German fintech
We helped a fast-growing German fintech to identify potential vulnerabilities in their cloud environment. This assessment strengthened their security posture, ensuring the safety of their sensitive data.
Show more
Testimonials
See what our clients say
-
Check all reviews on
Clutch
- eCommerce
"They always find a solution to my problem, there is no request that they can't develop."
-
"The Docktape team excels at agile development and execution on requirements, always putting the customer first and foremost, never sacrificing on quality, being transparent about risks to ensure timely delivery and high satisfaction."
-
"I am grateful for the guidance they provided on which AWS services to use to lower the storage cost of our huge dataset."
Contact