AskUI Desktop is now available

Deploy computer-use agents on any device

Install AgentOS on real desktops, embedded HMIs, and physical devices. Configure prompts, connect tools, route models, and ship agents to production.

Get started Talk to sales

macOS - Windows - LinuxFree tier availableOn-prem available

DeviceWindows POS - display 5

TargetCheckout FrontEnd

ToolsScreenshots + input

Execute the following task and write a report.task.md

## Task
# Cash transaction with 7% tax

Verify checkout, capture evidence, and report the result.

Read display context and select the active screen before acting.

+3.8s

Observe screen

active display 5

+3.6s · 333ms

Recognize Checkview, skip the Tableview branch, and proceed to the product action.

+5.4s

Save evidence

87943_FrontEnd/step_1

+0.0s · 305ms

Move mouse

(238, 182)Coca-Cola

+4.1s · 834ms

Observescreenshots

Reasonstate-aware steps

Actmouse + keyboard

Reportevidence trail

Trusted by engineering teams at

Code-first setup

Python SDK. Same runtime underneath.

Use the demo project structure to organize prompts, tools, procedures, plans, and tests that execute through AgentOS.

git clone github.com/askui/AskUI-Demo-ProjectGitHub

python -m venv .venv
source .venv/bin/activate

pip install -r requirements.txt

cp .env.template .env
# Add AskUI Hub credentials to .env

Python 3.10+MITAgentOS required for device control

What you configure

From install to production in one app.

Three things every team configures. Three views in the control plane.

Pair every device once.

Install AgentOS on macOS, Windows, or Linux targets. For embedded screens and devices you cannot instrument, use Companion Mode with HDMI capture.

Host Mode for full OS access
Companion Mode through a Raspberry Pi
CI/CD ready for VM logon and privilege flows

bench-04linux host mode

paired

head-unit-Aandroid companion

paired

kiosk-zone-2windows host

offline

pi-companion-2arm64 hdmi capture

paired

Configure prompts, tools, and models.

Each agent gets a system prompt, a model, and an allowlist of tools. Swap models without touching the prompt. Version everything.

Versioned system prompts
Per-agent model routing
Tool allowlists for controlled access

System prompt v4

Operate the head unit as a QA validator. After each step, capture a screenshot. If a label is missing, halt and report.

claude-sonnetcache on

Allowed toolsscreenshotcan-busfile-storehttp-fetch

Run it. Watch every step.

Trigger from the app, CI, a schedule, or an integration. Every action is traced, every screenshot is stored, every cache hit is visible.

Live run streaming
Cache hit and miss telemetry
HTML reports for audit and review

// run: head-unit-validator @ bench-04

[00.0s] screenshot   -> 1080x1920
[00.3s] locate       "settings gear"  [cache hit]
[00.4s] click        -> (912, 64)
[01.1s] screenshot   -> panel opened
[01.4s] locate       "Language"       [cache hit]
[01.6s] click        -> (240, 480)
[02.2s] scroll       -> "Deutsch" into view
[02.5s] click        -> tap dispatched
[03.1s] verify       label reads "Einstellungen"

pass - 3.1s - 2 LLM calls - 4 cache hits

Why teams pick AskUI

Computer-use is not the hard part. Production is.

Engineering leaders choose AskUI when they need agents to run reliably on real infrastructure.

Any device, any OS.

Windows, macOS, Linux, Android, iOS, plus embedded screens through Companion Mode when software access is not possible.

5+ OS targets

Any model, no lock-in.

Route each agent to Claude, Gemini, OpenAI, AskUI models, or your own provider through your credentials.

BYOM ready

Caching that pays for itself.

Skip redundant model calls on stable workflows and keep cost visible at every step of the run.

70-90% lower LLM spend

Built for production.

On-prem and air-gapped deployment, ISO 27001 certification, GDPR posture, audit logs, and no training on customer data.

ISO 27001

Where it runs

One runtime. Every surface your team ships.

AskUI runs anywhere there is a screen, from a browser tab to a bench instrument.

Desktop apps

Native and legacy software with no APIs, DOM, or selectors.

windows - macos - linux

Mobile apps

Native iOS and Android apps on real devices or simulators.

android - ios

Web apps

Browser flows using selectors and DOM tools, with screen-based execution when flows leave the browser.

chromium - firefox - webkit

Embedded HMIs

Touchscreens and device panels driven through capture and input.

qt - qml - custom stacks

Infotainment

Head units and in-vehicle displays across automotive stacks.

android auto - qnx - linux

Kiosks and ATMs

Locked-down terminals, POS systems, and ticket machines.

windows iot - linux

Lab instruments

Bench equipment and instruments beside the device under test.

vxi - gpib - usb

VMs and cloud desktops

Headless CI runs with logon and privilege flows handled.

vmware - hyper-v - cloud vm

Why classical automation stalls

Rule-based scripts break. Agents adapt.

Classical automation works for happy paths and breaks on drift, popups, and edge cases. AskUI agents reason about the screen and the goal.

Manual ops

Linear in headcount. Reliable for what people actually do, but every new workflow needs more people and time.

Classical automation

Fast for the happy path. Then it breaks on layout changes, drifted environments, and unnamed edge cases.

Computer-use agents

Reason about the screen and the goal, then handle conditions classical scripts cannot absorb.

ISO 27001 CertifiedGDPR CompliantOn-prem availableBYOM supportEU hosting

For builders

Start on the Free tier.

Install in minutes. Free includes a 14-day trial with 5,000 inference credits and a non-commercial AgentOS license.

Get started

For teams

Compare paid plans.

Starter, Pro, and Enterprise add commercial AgentOS, support, and Hub token-based inference usage.

View pricing

Deploy computer-use agents on any device

Python SDK. Same runtime underneath.

From install to production in one app.

Pair every device once.

Configure prompts, tools, and models.

Run it. Watch every step.

Computer-use is not the hard part. Production is.

Any device, any OS.

Any model, no lock-in.

Caching that pays for itself.

Built for production.

One runtime. Every surface your team ships.

Desktop apps

Mobile apps

Web apps

Embedded HMIs

Infotainment

Kiosks and ATMs

Lab instruments

VMs and cloud desktops

Rule-based scripts break. Agents adapt.

Manual ops

Classical automation

Computer-use agents

Start on the Free tier.

Compare paid plans.

We value your privacy