Instant Connection for Pixel Streaming

— New Feature Automated Setup

Comparison

Windows 365 for Agents Explained: GPU Desktop Alternatives

Learn what Windows 365 for Agents does and compare GPU desktop alternatives. Run AI assistants alongside Blender and other production apps on Vagon.

On this page

Windows 365 for Agents gives AI agents a managed Cloud PC where they can operate applications and complete tasks under enterprise policies. If you want an assistant to work in GPU desktop software, the next question is practical: where will the application, project files, and rendering workload run?

A Blender task makes that clear. Asking Claude Code to adjust a scene is one step. You also need Blender installed, a working application connection, enough GPU capacity for the render, and the saved project available when you return to make revisions.

At Vagon, we bring those workstation requirements together. You can create a cloud computer with Blender preinstalled, add your assistant, and keep the project in a persistent workspace. Change performance between sessions as the work demands, while your everyday computer stays available for other tasks.

We'll explain how Windows 365 for Agents works, distinguish it from Microsoft's GPU desktop offerings, and help you choose an environment for producing, reviewing, and continuing real application work.

What Is Windows 365 for Agents?

Windows 365 for Agents provides managed cloud computers that agents can use to carry out work in desktop applications and websites. Microsoft handles the Cloud PC infrastructure, while your organization determines which applications, identities, and policies the agent works with.

The service gives an agent somewhere to act. You still need the agent itself, its instructions, and the tools that connect it to the task. For example, an agent could navigate an internal portal, retrieve information, and save a report using a computer configured for that workflow.

How the Agent Gets a Computer

Microsoft organizes these machines into Cloud PC agent pools. A task follows a checkout/check-in cycle:

  1. The agent acquires a Cloud PC from an available pool.

  2. It performs the work through the configured tools and application access.

  3. It saves the required outputs and releases the computer for reuse.

Microsoft describes these agent computers as resetting after use. If a task creates a report, exported file, or modified project, the workflow needs to save that deliverable somewhere it remains available after the session ends.

The platform's MCP tools and session APIs cover computer access, desktop interaction, browser automation, and command execution. Microsoft also provides screen-sharing and control capabilities that an integrating application can expose for human supervision. You can therefore design a workflow that lets someone step in when the agent needs help.

Copilot Studio Machines settings showing Hosted browser, Cloud PC pool, and Bring-your-own machine options.

Source: Microsoft

Where Agent 365 and Copilot Studio Fit

The similar product names describe different responsibilities. Agent 365 provides agent governance and management. Copilot Studio provides tools for building agents, including a documented integration with Cloud PC pools. Windows 365 for Agents supplies the computer environment those agents use. Microsoft sells the runtime separately from Agent 365 and Copilot Studio.

Microsoft architecture diagram showing Windows 365 for Agents beneath agent platforms, Agent 365, and Microsoft IQ.

Source: Microsoft

This approach addresses enterprise automation with managed sessions and organizational controls. For GPU desktop applications, the purchasing decision adds another layer: the hardware and workspace must support the production software, its saved project, and the revisions you want to make after the agent finishes.

Does Windows 365 for Agents Include a GPU Desktop?

You should confirm GPU availability for the specific Windows 365 for Agents deployment you are considering. The public Agents provisioning guide covers pool configuration, images, and application preparation, but does not specify GPU tiers you can select for those pools. It does not establish the GPU configuration your production application would receive.

Microsoft does offer GPU-enabled Cloud PCs through Windows 365 Enterprise and Flex. Its published GPU Select, Standard, Super, and Max tiers provide minimum video memory allocations of 4 GB, 8 GB, 12 GB, and 16 GB respectively. Those desktop offerings have their own purchasing and availability details. Their existence does not establish the same options for the Agents service.

Windows Task Manager showing an NVIDIA A10-12Q GPU and dedicated GPU memory on a GPU Cloud PC.

Source: Microsoft Learn

Check the Application Workload Before Choosing the Computer

For an agent-assisted Blender job, the GPU requirement comes from what Blender must do. Preparing a Python script and rendering a complex Cycles scene place different demands on the machine. Before committing to an environment, establish:

  • The render device and memory available to the application. Confirm the GPU and VRAM allocation, then check whether your scene and assets fit the intended configuration.

  • The supported rendering path. Check that your application version, rendering engine, and driver can use the GPU for the operation you need.

  • The assistant's runtime requirements. Identify the operating system and dependencies required by its application connection. Microsoft's GPU Cloud PCs do not support nested virtualization, including WSL, so a setup that depends on WSL needs another supported path.

For the first trial, keep the scope small: open a representative scene, confirm the intended GPU is available in Blender, and render one frame. Inspect both the image and the saved project before expanding to a sequence.

That gives you a useful purchasing criterion: the computer can execute your actual application workload and retain the work needed for the next revision. Agent access, GPU capacity, and a usable production workspace each need to be established.

What Changes When the Agent Works in a GPU App?

Let's carry the Blender example through a production task: adjust the scene's lighting, render a camera sequence, and keep the project ready for another revision. The assistant needs access to the scene, Blender needs the resources to execute the work, and you need a desktop where you can inspect and finish it.

Start with the Application, Then Connect the Assistant

On Vagon, you can select Blender in our Computer Composer preinstall catalog. We handle that application installation when creating your computer. You can then add your preferred tools, project assets, and any version-specific dependencies inside the workspace.

The assistant's connection is the next layer. For example, this published Blender workflow uses a Python bridge inside the open application to receive scripts from Claude Code. Scene changes appear in Blender, while longer renders can run in a background Blender process from the saved project.

Rendered FPV drone with cyan and magenta lights above a city street at night.

Source: mromidev

In that arrangement, run Claude Code and the bridge on the Vagon computer alongside Blender. Commands, project access, and application execution then happen in the cloud workspace. Preinstallation gets the production app ready; you still configure the bridge and sign in to the assistant with your own account.

Keep the Working Environment Through Revisions

After the first render, you might want a softer shadow or a different camera angle. Those changes need the editable scene and its assets, rather than only a downloaded image. Our cloud computer retains your installed applications, settings, and saved files between sessions while your subscription remains active.

If the next render needs a different hardware allocation, save your work and change performance between sessions. Your application setup and project remain in place when you restart on the selected tier.

This is why we recommend a dedicated Vagon workspace for ongoing GPU application work. You can delegate the operation, enter the same desktop to make creative decisions, and continue the project without moving its execution back onto your everyday computer.

GPU Desktop Alternatives: Which Setup Fits the Job?

For GPU application work, choose around the environment you need to operate. A creator preparing a Blender render and an IT team deploying desktops across an organization may need different services, even when both require a GPU.

These alternatives provide GPU desktop environments. Reproducing Windows 365 for Agents' managed pool lifecycle would be a separate integration project.

Option

Strongest fit

What you prepare

Vagon Cloud Computer

A production workspace with preinstalled apps and performance you can change between sessions

Your project, assistant account, and application connection

Windows 365 Enterprise/Flex GPU

GPU desktops within an existing Windows 365 deployment

The GPU desktop offering, application setup, and assistant tools

Azure Virtual Desktop

An IT-managed deployment with control over GPU session hosts

Hosts, images, drivers, policies, storage, and application tools

Amazon WorkSpaces Personal graphics bundles

GPU desktops managed within AWS

A compatible bundle, access configuration, software, and assistant tools

Vagon Cloud Computer: Get the Application Workspace Ready

For an individual creator or a small production team, we recommend Vagon when the immediate goal is to get an assistant working with GPU software. You can begin with the preinstalled application and retained workspace described above, without assembling your own virtual desktop infrastructure.

We also manage GPU drivers, so driver maintenance is handled as part of the platform. You can connect through a browser or our desktop app, then move between assistant-led operations and direct work in the application.

Project transfer is built into the workflow. With Vagon Files, you can upload assets from the dashboard while the computer is stopped. Sync the project onto the cloud computer's internal disk before opening it in Blender. Keep active editing and rendering on that disk, then transfer the completed files back through Vagon Files or download them.

That gives you a practical path from project input to editable output, with the assistant and application operating in one workspace.

Windows 365 Enterprise/Flex GPU: Extend Your Microsoft Desktop Setup

If your organization already operates Windows 365, its GPU desktop offerings are worth evaluating through the same IT team. Confirm the appropriate tier and regional availability with your Microsoft account team; GPU offerings are purchased through that channel rather than the web direct store.

This route lets you evaluate GPU application work within your existing desktop-management approach. The agent tools and application connection still need configuration. Keep the purchase distinct from Windows 365 for Agents: a GPU desktop subscription does not by itself establish the managed agent-pool workflow discussed earlier.

Azure Virtual Desktop: Configure Around Your IT Requirements

Azure Virtual Desktop supports GPU-backed session hosts, with documented settings for application rendering and remote frame encoding. Your team selects supported hosts and drivers and configures the graphics policies required by the deployment.

Microsoft Intune settings picker showing Remote Session Environment policies for hardware graphics adapters and H.264 encoding.

Source: Microsoft Learn

Choose this route when control over the infrastructure is part of the requirement. Account for the work involved in maintaining the host environment, application image, and storage alongside the assistant integration. A GPU VM alone does not complete the desktop delivery setup.

Amazon WorkSpaces Personal: Build on Your AWS Environment

For organizations using AWS, WorkSpaces Personal graphics bundles provide another desktop route. AWS offers fractional and full GPU configurations, with compatibility depending on the operating system, protocol, and region.

Amazon WorkSpaces Select Bundle screen showing GPU-enabled Graphics.g4dn, GraphicsPro, and GraphicsPro.g4dn options.

Source: AWS

Evaluate the exact bundle with your production software and project. Use documentation for WorkSpaces Personal when you need a desktop; WorkSpaces Applications is a separate application-streaming product. Include the assistant and its application dependencies in your deployment plan.

If your priority is starting the GPU application job rather than managing a desktop deployment, start with our Vagon workspace. The first useful milestone is straightforward: your project opens, the assistant can perform the intended operation, and you can inspect and continue the result.

Compare the Cost of a Finished Task

An hourly rate becomes useful once you know what the machine must complete. For our Blender example, count the time spent preparing the project, checking the assistant's changes, rendering previews, producing the sequence, and reviewing the files. Include any corrections that require another run.

Microsoft bills Windows 365 for Agents on consumption, with an additional monthly fee for always-available Cloud PCs. Rates vary by geography, and its usage calculation applies hourly-window rounding. Compare that against the deployment you actually need; the Agents runtime rate does not establish the cost of a GPU desktop suitable for your scene.

Microsoft 365 admin center Cost management page with the classic Billing and usage link highlighted.

Source: Microsoft Learn

On Vagon, our performance rates are billed by the minute, with storage charged separately. You can match the selected resources to each stage instead of keeping one hardware configuration throughout production. Save and stop the computer before changing tiers, then continue with the same installed tools and project files.

Build your estimate from three parts:

  • Machine usage: preparation, rendering, review, and retries on the selected performance tiers.

  • Ongoing access: project storage, the assistant subscription or API usage, and any paid application licenses.

  • Setup and maintenance: time spent installing tools, connecting the assistant, and keeping the environment usable.

Preinstalled applications and managed GPU drivers reduce the setup work you take on with Vagon. Retaining the configured workspace also means that the next revision starts from your existing environment.

Run one representative task, record its billable duration, and inspect the saved output before projecting monthly usage. That gives you a budget tied to usable production work and a clearer reason to choose the workspace supporting it.

Start Your First GPU Application Task on Vagon

Start with a scene you already understand and one change you can judge. A lighting revision is a useful first task: you can compare the preview with the original, inspect the editable scene, and decide whether to expand the workflow to a full camera sequence.

Create your Vagon Cloud Computer and prepare the job in four steps:

  1. Choose Blender during computer creation. Select it in Computer Composer so we can install it for you. Pick a nearby region, allocate space for your project and outputs, and choose performance appropriate to the scene.

  2. Bring in a working copy. Upload the scene and its assets through Vagon Files, then sync them onto the cloud computer's internal disk. Open the project in Blender and check that the textures and other dependencies are available.

  3. Add Claude Code and the application connection. Install the assistant inside the Vagon computer using Anthropic's setup instructions, then sign in with your account. For the Python bridge approach introduced earlier, prepare the bridge in Blender and its sending script. Confirm that the assistant can read the scene's file path and camera list.

  4. Save and inspect one revision. Agree on the change, save a new scene version, and render a preview to a project folder on the internal disk. Keep the original available for comparison.

Once the connection works, try a prompt such as:

Inspect this scene and report its active camera and lighting setup. Propose a lighting adjustment to soften the main shadow. Wait for my approval, then save a new scene version and render one preview frame. Keep the original file unchanged.

Review the result in Blender before asking for the sequence. Check the shadow, camera framing, and saved scene, then give the assistant a specific correction if needed.

For a longer render, keep the computer and required processes running, and check automatic shutdown settings. When the work is complete, transfer the deliverables through Vagon Files or download them.

You now have the application, assistant connection, and project together for the next task. Let that production work use your Vagon computer while your everyday machine stays available, and return to the cloud desktop whenever you need to inspect or finish the result.

FAQs

Can I Use My Own Agent with Windows 365 for Agents?

Microsoft provides session APIs and MCP tools for agent integrations. Your implementation must handle authentication, authorization, session checkout, and the tool calls used during the task. Support for MCP alone does not establish that a client is configured to acquire and operate a Cloud PC in your tenant.

What Licenses Do I Need for Microsoft's Agent Cloud PCs?

For the Intune provisioning path used with Agent 365, Microsoft lists an Intune license, an Agent 365 license in the tenant, and an active Windows 365 for Agents billing plan. Supported partner solutions have their own setup paths, so check the requirements for the integration you will use.

Will a More Powerful GPU Make Claude Code Respond Faster?

A workstation GPU accelerates supported work performed on that workstation. In our example, that work is Blender rendering. It does not directly accelerate a Claude model hosted by Anthropic. Choose resources around the application's workload; our guide to when an AI agent workflow needs a GPU explains the distinction.

Can I Install Software Beyond Vagon's Preinstall Catalog?

Yes. Our preinstall catalog gives you a prepared starting point. You can install additional compatible applications, plugins, and tools inside your cloud computer. For paid software, bring a license that permits your intended cloud use. Your assistant account and its connection to the application are configured separately.

Can I Transfer Files When My Vagon Computer Is Stopped?

Yes. Open Vagon Files from your dashboard to upload or download files without starting the computer. Upload assets in advance, then sync the working copy onto the computer's internal disk when you start a session. Completed outputs must be transferred to Vagon Files before they are available there.

Does Saving My Project Mean the Agent Will Keep Running?

Saving preserves the project state; continued execution requires the computer and relevant processes to remain active. Check shutdown settings before leaving a long task. When you finish, save the scene and transfer your deliverables so you can resume production from a known result in your next session.

Vagon Computer

Run it on a cloud computer

Open it from any browser, Mac, iPad or Chromebook. No local hardware needed.

Vagon Computer

Run it on a cloud computer

Open it from any browser, Mac, iPad or Chromebook. No local hardware needed.

Written by

Mert Boz

Mert is a philosophy student with a strong interest in economics, motorsports, and basketball. He enjoys thinking about how the world works, both in theory and in real life. In his free time, you can often find him watching a race, following a basketball game, or reading about economic ideas.

More posts by this writer →

Related

More on AI Agents

All posts →

AI Agents

Comparison

Choose Claude Code on the Web or a Cloud Computer

·

13 min read

AI Agents

Fix

Fix MCP Server Not Connecting in Claude Code and Codex

·

13 min read

AI Agents

Comparison

AI Sandbox vs Cloud Desktop: Which Your Agent Needs

·

13 min read

Vagon Computer

Run this guide on a cloud computer.

Open it from any browser, Mac, iPad or Chromebook. No local hardware needed.

Vagon Computer

Run this guide on a cloud computer.

Open it from any browser, Mac, iPad or Chromebook. No local hardware needed.