Finds buttons by name, not pixels
aos reads the accessibility tree, so an agent presses "Equals" instead of guessing coordinates from a screenshot. No vision model, no brittle pixel loop.
aos mouse click --app=Calculator --find=EqualsmacOS · Windows · Linux · one binary
Let your AI use your computer, without taking it over.
aos is a CLI and an agent skill. Install it once, and your agent can arrange your windows, open apps, click, type, show you where to click, and record what it did, on macOS, Windows and Linux.
curl -fsSL https://raw.githubusercontent.com/muthuishere/aos/main/install.sh | shcurl -fsSL -o install.cmd https://raw.githubusercontent.com/muthuishere/aos/main/install.cmd && install.cmdnpm i -g @muthuishere/aosThe agent skill installs itself on first run (opt out: AOS_NO_SKILL_AUTOINSTALL=1). Then run aos doctor. More install options.
aos capture video start smart. Captions show each exact command.Why aos
Works beside you, not instead of you
Input goes to the app you name, even while you keep typing somewhere else. When you should act, it draws an arrow, a label or a subtitle that your clicks pass through.
aos draw --app=Calculator --find=7 --text="Press 7"One binary, every OS, every agent
The same commands on macOS, Windows and Linux. The skill installs itself, so Claude Code or any CLI agent knows every command on first run, and MCP is there too.
aos window move Safari --zone=1A --monitor=2What it does
121 commands across 36 groups. Same verbs, same exit codes, on every OS.
Windows & layout
List, focus, move and resize any window. Snap to halves, quarters or N-way splits on any monitor.
aos window move Safari --zone=1A --monitor=2
Find on screen
Locate a button, menu item or text inside a window via accessibility or OCR, and get its exact center.
aos find --app=Calculator --text=7
Point for humans
Draw a big click-through arrow, ring or box with a sign, so a person knows exactly where to click.
aos draw --app=Calculator --find=7 --text="Press 7"
Focus-free input
Type and press key combos into a named app while you keep working in another window. --verify reads the text back.
aos key type --app=TextEdit --verify "hello"
Labels & subtitles
Put a styled text label or a film-style caption on screen so the agent can narrate what it is doing.
aos subtitle show "Saving the report…"
Capture video
Record the screen and get a .final.mp4 where a smart camera has already zoomed in on every action.
aos capture video start smart
Screenshots
Capture a monitor, a region or one app window so the agent can see what it did.
aos capture screenshot window --app=Safari
Shell & packages
Run commands with JSON-captured output, and one package verb set over brew, winget, apt and more.
aos exec capture -- git status
Browser & terminal tabs
List every tab in Chrome, Safari, Edge, Ghostty, iTerm and Windows Terminal, then focus, open, type into or close one by title, URL or folder.
aos tab focus --url=github.com/pulls
From WSL and SSH
Screen commands follow the screen. Inside WSL they act on the Windows desktop; over SSH they reach back to the laptop you connected from.
aos doctor | grep host
Watch for changes
Stream one JSON line whenever the clipboard, the focused window or a file changes, so an agent can react instead of polling.
aos watch window
Clipboard
Copy words or stdin to the clipboard and paste it back out, the same way on every OS.
aos clipboard paste
Launch, open & web apps
Start an app, hand a file or URL to its default app, or open a site as its own chromeless window.
aos webapp open https://mail.google.com
Processes & services
List, find and kill processes, and install any command as a per-user service that survives reboots.
aos service create sync --autostart -- rclone sync a b
Files
Read line ranges, append, list, copy, move and delete files with exit codes an agent can trust.
aos file read notes.md --lines=1:20
System & hardware
Volume and mute, battery, power, Wi-Fi and IP, monitors, fonts, plus lock and sleep.
aos audio volume 30
Audit log
Every command this machine was asked to run is recorded, with failures and latency, and exports as OTLP.
aos obs audit --since=1h --failed
Your own commands
Add adapters as small config files and they show up as aos commands, in the skill and over MCP.
aos adapters example --write
Remote screen share
Hand this screen and its input to a browser on the LAN, and watch the agent work from your phone.
aos remote share
Headless & MCP
Give a Linux server a virtual display, or serve every command as typed MCP tools.
aos serve mcpBeyond windows: your whole machine
Clipboard, files, processes, packages, audio, the audit log. Every one is a command with the same exit codes, and most of them need no screen at all.
aos
Watch your machine at the OS level. One JSON line per change.
$ aos watch clipboard{"event":"clipboard","seq":1,"length":7,"digest":"4b559aab5a29",…}
$ aos watch window --max=1{"event":"window","app":"Google Chrome","title":"aos docs","monitor":3,…}
$ aos watch file ./src --recursive{"event":"file","kind":"created","path":"./src/x.go","size":0,…}
One CLI for you and your agent · macOS · Windows · Linux · muthuishere.github.io/aos
aos
The clipboard, in and out of a pipeline.
$ echo "hello from aos" | aos clipboard copy
$ aos clipboard pastehello from aos
$ aos watch clipboard --max=1 --content{"event":"clipboard","seq":1,"length":14,"text":"hello from aos"}
One CLI for you and your agent · macOS · Windows · Linux · muthuishere.github.io/aos
aos
Files with exit codes an agent can trust.
$ aos file stat package.json --json{"name":"package.json","size":499,"mode":"-rw-r--r--","dir":false,"modified":"2026-10-13T06:02:13+05:30"}
$ aos file read notes.md --lines=1:3# Monday
- call the bank
- ship the docs
$ aos file delete /aos: refusing to delete /
One CLI for you and your agent · macOS · Windows · Linux · muthuishere.github.io/aos
aos
Processes and services, without sudo.
$ aos process find Finder --json[{"pid":2424,"name":"Finder","cpu":0,"memory_mb":97.4,"user":"you"}]
$ aos process kill Calculator
$ aos service create mcp --autostart --now -- aos serve mcpname mcp
label aos.mcp
state running
One CLI for you and your agent · macOS · Windows · Linux · muthuishere.github.io/aos
aos
Audio, battery, network, lock. Same verbs on every OS.
$ aos audio volume 40 && aos audio mute off40
$ aos battery status && aos network status100% charged
interface en0
address 192.168.1.24
wifi HomeNet
$ aos system lock
One CLI for you and your agent · macOS · Windows · Linux · muthuishere.github.io/aos
aos
Every command this machine was asked to run, recorded.
$ aos obs tail --limit=306:55:20 ok cli 56ms process find
06:55:21 err cli 47ms process kill
06:55:21 ok cli 351ms doctor
$ aos obs audit --since=24h --source=mcp2026-10-13 06:02:59 mcp window move 121ms ok
2026-10-13 06:03:07 mcp key type 306ms ok
$ aos obs export --since=1h > spans.json
One CLI for you and your agent · macOS · Windows · Linux · muthuishere.github.io/aos
aos
One package verb set over brew, winget, apt and more.
$ aos pkg managerhomebrew
$ aos pkg search ripgrepripgrep
ripgrep-all
$ aos pkg install ripgrep
One CLI for you and your agent · macOS · Windows · Linux · muthuishere.github.io/aos
aos
Watch it from your phone. Hand it to an agent over MCP.
$ aos remote shareLAN http://192.168.1.24:8765/?k=…
Press Ctrl-C to revoke the link and stop.
$ aos serve mcpaos mcp http://127.0.0.1:14320/mcp
token YOUR_TOKEN
tools 96
$ aos serve tools --groups=windowwindow_list List open windows
window_move Move a window to a zone, a split, or bounds
One CLI for you and your agent · macOS · Windows · Linux · muthuishere.github.io/aos
Works with your agent
The skill ships inside the binary, so Claude Code already knows every command. Just ask in plain English.
> put TextEdit on the left half of monitor 2,
then point at its Format menu with a green arrow$ claude -p "screenshot Safari and tell me
what page is open"