ON-SCREEN AI FOR WINDOWS

An AI companion that sees your screen
only when you ask.

Chat by voice or text, get click-by-click guidance in real apps, or let the agent do the work with explicit permission. Five modes—from safe Idle to Co-Pilot halo guidance and Pilot automation—with always-visible indicators, instant stop, and no password typing.


Permission-first

It asks before viewing your screen or taking any action—every time.

Instant control

Clear on-screen indicators and immediate stop via Esc, Stop, or mouse.

Permission-first automation with always-visible indicators

Co-Pilot halo guides clicks inside your real apps

Transparent controls

No hidden automation. Clear, visible permissions.

Every capability is labeled with what the agent can see or do, when it will ask first, and how to stop instantly—so you always stay in control.

PERMISSION FIRST

Always asks before acting

Nothing runs silently. The agent requests explicit permission before viewing your screen, guiding clicks, or taking actions—then shows clear indicators while it’s active.

VISIBLE INDICATORS

Know exactly when it can see

Idle mode is chat-only. When screen view is enabled, you’ll see an on-screen border/banner so there’s no ambiguity about what’s happening.

INSTANT STOP

Esc, Stop, or move the mouse

Stop the agent immediately with Esc, the Stop control, or your mouse. Control is designed to be obvious and interruptible at any moment.

NO PASSWORD TYPING

Protect sign-ins by design

The agent won’t sign in for you or type passwords. Risky steps require confirmation, keeping sensitive actions firmly in your hands.

Safety-first automation, by design

An AI companion that helps—only when you ask

AI Desktop Agent lives on your screen to chat, explain, and guide work inside real Windows apps. It can view your screen only with permission, shows clear on-screen indicators, and stops instantly. You stay in control—whether you want gentle click-by-click guidance or fully assisted task runs with confirmations.

  • Five modes: Idle chat, Co‑Pilot guidance halo, Pilot actions, Auditor alerts, Tutor soon.
  • Explicit permission, visible border/banner, and instant stop via Esc/Stop/mouse.
  • Learns from official manuals with cited sources; save repeatable “recipes” for later.

Read our safety model →

Five modes

Pick your level of control—chat to full pilot

Start with chat-only, then turn on screen awareness and action only when you ask. Each mode has clear on-screen indicators, permission prompts, and an instant stop so you stay in charge.

IDLE MODE

Chat only (no screen view)

Ask questions, draft text, and plan workflows while your screen stays private. Great for quick answers, writing help, and troubleshooting before you act.

Safety details →

CO-PILOT MODE

Step-by-step click guidance

The agent points, explains, and guides each click inside real apps—perfect for careful setup, admin tasks, and learning a new tool without surrendering control.

See all modes →

PILOT MODE

Do the task—with instant stop

With explicit permission, the agent can take actions for you with clear control indicators, confirmations for risky steps, and immediate stop via Esc/Stop/mouse.

Check requirements →

How it works

Choose your mode. Stay in control.

Start in Idle, then switch into guided help or hands-on automation—always with permission, visible indicators, and instant stop.

1. Idle (chat-only)

Ask questions by text or voice. The agent won’t view your screen unless you request it—ideal for planning, drafting, and quick help without sharing.

Anytime

Text or voice

2. Co‑Pilot (guide my clicks)

Turn on step-by-step guidance to see exactly where to click inside real apps, with clear on-screen cues and explanations as you go.

Most popular mode

3. Pilot (do it for me)

Let the agent perform tasks with explicit permission and clear control indicators. Stop instantly with Esc/Stop or moving your mouse—no surprises.

Instant-stop control

ON-SCREEN AI COMPANION

A Windows desktop agent that guides clicks—or acts—with your OK

Chat, speak, and get step-by-step help inside real apps. It can view your screen only when you ask, performs tasks only with explicit permission, and stops instantly via Esc/Stop/mouse. Choose safer modes from Idle to Co‑Pilot and Pilot.

Before setup

After 1 week

Trusted by power users

“I used it to set up a tricky workflow in real apps without handing over control. It asked before viewing my screen, showed a clear ‘running’ indicator, and Esc stopped it instantly. The step-by-step Co-Pilot mode feels like having an expert on call.”

— Jordan P., IT support

< 5 min

TIME TO FIRST AUTOMATION

0 logins

PASSWORDS TYPED BY AGENT

REAL USER NOTES

Trusted help—without giving up control.

People use AI Desktop Agent to get unstuck in real Windows apps. Every mode is permission-based, visibly indicated, and stoppable instantly—so the assistant stays helpful, not spooky.

“Co-Pilot mode feels like having an expert over my shoulder—clear halos, clear steps, and it never grabs control unless I explicitly switch to Pilot.”

Maya K.
IT support lead · Austin

“Pilot is the first automation I trust. The border makes it obvious when it can see my screen, and Esc stops everything immediately—no drama, no guessing.”

Jordan P.
Windows power user · Toronto

“I like that it cites official manuals and saves repeatable recipes. It’s not just chat—it actually helps me finish the task, with confirmations before anything risky.”

Samir D.
Ops analyst · London

Agent Journal

How to use an AI agent safely on Windows

Practical reads on click-guidance, permission-based automation, and privacy-first setup. Learn when to stay in chat-only, when to share your screen, and how to keep full control—every step, every time.

Co-Pilot mode · 8 min

Click-by-click guidance that never takes control

See how Co-Pilot adds a clear halo and step prompts inside real apps—so you stay the operator while the agent explains exactly what to click and why.

Continue reading →

Pilot controls · 10 min

Instant stop, clear indicators, and no password typing

A quick tour of Pilot mode safeguards: explicit permission, a visible control state, confirmations for risky actions, and immediate stop via Esc/Stop/mouse—without ever entering passwords.

Continue reading →

Privacy · 7 min

Local models, on-device speech, and what stays on your PC

Understand your options for using Ollama locally, choosing Claude when you want cloud power, and configuring voice and screenshots so sharing happens only when you ask.

Continue reading →

GET STARTED

Download AI Desktop Agent & set up in minutes

Get the installer, system requirements, and a quick-start walkthrough. You stay in control with clear on-screen indicators and instant stop.