Blog · October 30, 2026

What Is an AI Agent on My Computer?

An AI agent on your computer is software that uses an AI model to take actions for you, not just answer questions. Instead of replying with text, an agent can open apps, click through websites, read and edit files, run commands or send messages to finish a task you describe. Some agents run on your own Windows 11 PC, some run in the cloud on their own virtual computer, and some connect the two. This guide explains what agents are, how they differ from chatbots and regular apps, the main examples you will run into, what access they can get, and why it helps to see what they are doing.

AI agents in plain English

Think of a chatbot as someone who gives you directions, and an agent as someone who takes the wheel. You give an agent a goal, such as “find three flights for next Friday and put them in a spreadsheet,” and it decides the steps, uses tools to carry them out, checks the results and keeps going until the task is done or it needs your input.

Most agents share three traits:

They act, not just talk. They use tools: a browser, a terminal, files, or connections to apps like email and calendars.
They work in steps. An agent plans, takes an action, looks at what happened, and picks the next action.
They work on your behalf. When an agent signs in, opens a file or clicks a button, it does so with access you or an app gave it.

AI agent vs. chatbot vs. regular app

	Regular app	Chatbot	AI agent
What you give it	Clicks and typing	A question	A goal or task
What it does	Exactly what you tell it, one step at a time	Writes an answer	Plans and carries out several steps
Uses other tools	Only its own features	Usually not	Yes: browser, files, terminal, connected apps
Who is “driving”	You	You (it only replies)	The agent, within the access it has

A regular app, like a spreadsheet program, does one thing when you click. A chatbot answers in a chat window and leaves the doing to you. An agent closes that gap: it can carry out the work itself. The same product can be both: many AI assistants that started as chatbots now include agent features that can take actions.

Common examples of AI agents

These are some of the best-known agents and agent features. Details change often, so check each vendor’s current documentation.

Windows 11 agentic features and Copilot Actions

Microsoft is adding agent support to Windows 11 itself. According to Microsoft, the Experimental agentic features setting enables an agent workspace: “a separate, contained space in Windows where you can grant agents access to your apps and files so they can complete tasks for you in the background while you continue to use your device.” Copilot Actions, in preview for Windows Insiders, uses it. Microsoft says agents typically get access to known folders or specific shared folders, and each agent has its own workspace and its own permissions.

The setting is off by default, and only an administrator can turn it on. Our guide to Windows 11 agentic AI and how to turn it off walks through where it is and what it changes.

ChatGPT agent (OpenAI)

OpenAI introduced ChatGPT agent in July 2025. It combined Operator, which could scroll, click and type on websites, with deep research, and OpenAI said it worked “using its own virtual computer” with a visual browser, a text-based browser, a terminal and direct API access. It could also use ChatGPT connectors to apps like Gmail and GitHub, and OpenAI said it asked for permission before “actions of consequence.”

OpenAI’s help center now says ChatGPT agent is no longer available and points users to ChatGPT Work for longer, multi-step tasks and to a cloud browser in ChatGPT for supported browser workflows. The idea is the same: these tools work on OpenAI’s computers rather than as a program on your PC, but they can still act on your data through connected apps and websites you sign them into.

Claude computer use (Anthropic)

Anthropic offers a computer use tool for developers building with Claude. Anthropic says it “provides screenshot capabilities and mouse/keyboard control for autonomous desktop interaction.” In other words, the model looks at the screen and operates a computer much like a person would. Anthropic’s documentation says computer use “requires a sandboxed computing environment” and advises developers to “isolate Claude from sensitive data and actions.” Developers build this tool into their own products, so you may meet it inside another app rather than under Anthropic’s name.

OpenClaw

OpenClaw is an open-source project that its documentation describes as “a self-hosted gateway that connects your favorite chat apps,” such as Discord, Slack, Telegram and WhatsApp, “to AI coding agents.” You run a gateway process on your own machine or a server, and it becomes the bridge between your messaging apps and an AI assistant. This is a good example of an agent that runs on your own hardware and that you can send tasks to from a phone chat app.

Other agents you may already have

Many everyday tools now include agent features:

AI coding assistants that can edit files and run commands in a project folder.
Browser extensions that can read and act on the pages you visit.
Desktop AI assistants that can see your screen or control apps.
Background helpers that start with Windows and keep running after you close the main window.

Where AI agents run: on your PC, in the cloud, or both

It helps to sort agents by where the “hands” are:

On your PC. The agent’s program runs on your computer. Examples include Windows’ agent workspace, desktop assistants, coding tools, browser extensions and self-hosted tools like OpenClaw. These can reach whatever their access allows on that machine.
In the cloud. The agent works on the vendor’s own virtual computer or cloud browser, as OpenAI’s ChatGPT agent did. It doesn’t run on your PC, but it can still act on accounts and data you connect.
Both. A cloud model can drive a program on your PC, for example a desktop app that sends screenshots to an AI model and carries out the clicks it sends back.

For your own Windows 11 PC, the first and third groups matter most, because that is where actions happen on your machine.

What access can an AI agent get?

An agent can do only what its access allows, but that access can be broad. Depending on the tool and what you approve, an agent may be able to:

Read and change files, either specific shared folders (as with Windows’ agent workspace) or anything your user account can open.
See your screen and control the mouse and keyboard.
Use your browser, including sites where you are signed in.
Run commands in a terminal.
Use connected accounts, such as email, calendars and code repositories, through connectors or app permissions.
Keep running in the background or start automatically with Windows.

Many agent apps you install run with your own user permissions, outside Windows’ agent workspace, so they can reach whatever your account can. Vendors also warn about a specific risk called prompt injection: instructions hidden in a webpage, document or image that try to trick an agent into actions you didn’t ask for. Anthropic warns that instructions on webpages or in images “might override your instructions,” and Microsoft calls the same risk cross-prompt injection. That is why vendors recommend keeping agents away from sensitive data they don’t need.

Why visibility into AI agents matters

Agents are useful. The point isn’t to avoid them, it’s to know which ones are acting on your computer and to stay in charge of them. Visibility helps you:

Spot agents you didn’t set up, such as a helper bundled with another app or an extension someone else installed.
Match activity to your intent. If you didn’t ask an agent to do anything, it shouldn’t be busy.
Make clear decisions for a team. On shared or work PCs, owners need to know which AI tools are in use before they can set a policy.
Clean up old tools you tried once and no longer use.

You can check a lot by hand. Our step-by-step guide on how to tell if an AI agent is running on your Windows PC covers Task Manager, startup apps, installed apps, browser extensions and privacy settings.

How PERCARIS helps

Manual checks give you a snapshot. PERCARIS surfaces recognized AI-agent activity on your Windows computer and gives you a Stop control for agents you did not authorize.

PERCARIS recognizes known AI tools. It won’t flag every AI agent.

You stay in control: PERCARIS shows you what it sees, and you decide whether to press Stop. It works alongside your existing security tools, and each PC is managed on its own.

Plans cover 1 PC (Personal), up to 5 PCs (Power) or up to 25 PCs (Small Business). You pay through Lemon Squeezy, your installer and license key arrive by email, and every plan includes a 14-day refund window. See plans.

FAQ

What is an AI agent on my computer?

It is software that uses an AI model to carry out tasks for you, such as opening apps, using websites, editing files or running commands, instead of only answering questions.

Is an AI agent the same as a chatbot?

No. A chatbot replies with text and leaves the work to you. An agent plans and takes actions with tools. Some AI assistants can work as either.

Is Copilot an AI agent?

Copilot can act as one through Copilot Actions, in preview for Windows Insiders, which uses the Windows 11 agent workspace when the Experimental agentic features setting is on. That setting is off by default, and only an administrator can turn it on.

Can an AI agent access my files?

It depends on the agent. Windows’ agent workspace typically limits agents to known folders or folders you share. Many installed agent apps run with your own user permissions, so they can reach whatever your account can.

How do I see which AI agents are on my PC?

Check Task Manager, Startup apps, installed apps, browser extensions and Settings > System > AI components. Our guide on how to tell if an AI agent is running has the steps.

Sources

Product features change often. Checked against each vendor’s documentation on October 30, 2026.

— The PERCARIS Team