New Repeat tasks feature — set up commands that automatically run at intervals — view on GitHub ↗

[+] AI PC Automation

Your Personal
AI PC Assistant.

Control your Windows PC with plain English — from anywhere in the world. OmniCtrl sees your screen, clicks buttons, types text, and fixes its own mistakes. Command it over Telegram or straight from your terminal.

↓ Install Now View Source → 
 ___  __  __ _  _ ___  ___ _____ ____  _     
/ _ \|  \/  | \| |_ _|/ __|_   _|  _ \| |    
| (_) | |\/| | .` || || (__  | | | |_) | |__  
\___/|_|  |_|_|\_|___|\___| |_| |____/|____|
│ Open Chrome and search for "weather today" Gemini Flash
[✓] Screenshot captured — analyzing screen state
[✓] Located Chrome icon at coords (42, 847)
[✓] Browser launched — navigating to google.com
[✓] Typed "weather today" in search bar
[✓] Task complete — sending final screenshot to Telegram
─────────────────────────────────────────
5 actions · 4.2s · no errors
ctrl-p commands ctrl-s screenshot ctrl-c stop task tab switch mode
20+
Telegram commands
built in
2
AI models working
together
∞
Miles away you can
be from your PC

[+] Capabilities

Everything you need to control your PC.

OmniCtrl combines computer vision, AI planning, and Telegram — giving you a complete remote-control suite with no extra software on your phone.

[+]Plain English CommandsJust say what you want — "Open Chrome and search for cute cat videos" — and it figures out the rest.
[+]Screen VisionTakes screenshots to understand the current state of your screen before every action.
[+]Mouse & Click ControlAutomatically locates and clicks buttons, links, menus, and UI elements using AI coordinate extraction.
[+]Keyboard AutomationTypes text in any application — forms, documents, search bars — exactly as instructed.
[+]Self-Healing RecoveryWhen something goes wrong, it analyzes the situation, adapts, and retries with a different approach.
[+]Remote Control via TelegramSend commands from your phone wherever you are. Your PC obeys.
[+]Voice CommandsSend a voice note — it transcribes and executes your spoken instruction automatically.
[+]Notification ForwardingPC desktop notifications are forwarded straight to your phone in real time.
[+]Task SchedulingSchedule tasks for specific times — /schedule 09:00 Send daily report
[+]Repeat TasksSet up commands that automatically run at specified intervals — /repeat add 300 sysinfo (every 5 min)
[+]System MonitorCheck CPU, RAM, and disk usage remotely with a single Telegram command.
[+]Screen RecordingRecord a 10-second video of your screen and receive it directly in Telegram.
[+]Microphone & WebcamRecord audio from your PC's mic or snap a webcam photo remotely.
[+]Popup MessagesDisplay important messages directly on your PC screen from your phone.

[+] Architecture

How OmniCtrl works.

A two-model AI pipeline handles planning and execution — Gemini Flash for strategy, Gemma for coordinate precision.

Command arrives via Telegram (text or voice)
You send a message or voice note from your phone. Voice is automatically transcribed using Gemini before being processed.
Gemini Flash plans a step-by-step action sequence
The AI understands your intent and decomposes it into atomic steps: open app, find element, click, type, verify.
Screenshot captured — screen state analyzed
A screenshot is taken before every action. The AI sees exactly what you see and identifies interactive elements.
Gemma extracts pixel-accurate coordinates
A specialized model identifies exactly where to click — button centers, input fields, menu items — with sub-pixel precision.
Actions executed — mouse moves, clicks, keystrokes
pyautogui performs the physical interactions. Each action is verified before proceeding to the next step.
Error detected → self-healing retry
If a step fails (element not found, unexpected dialog), the AI re-analyzes and tries a different approach automatically.
Summary + final screenshot sent to Telegram
You receive a confirmation of what was done along with a screenshot proving the outcome.

[+] Setup

Get started in minutes.

Works on Windows 10 and Windows 11. Requires Python 3.8+, Tesseract OCR, and API keys (all free tiers available).

git clone https://github.com/Aryan-crypt/Computer-Agent.git && cd Computer-Agent
pip install -r requirements.txt
python Core/core_agent.py
python main.py
[+] FULL INSTALLATION GUIDE
Step-by-step setup on GitHub
The complete guide covers Python setup, Tesseract OCR, API key configuration, Telegram bot creation, and troubleshooting — with screenshots and beginner-friendly instructions.
→

[+] Reference

Telegram command reference.

Every built-in command available once your bot is running. Send these from your phone to control your PC.

Command Description
/startStart the bot and see available features
/helpDisplay full help guide
/task <text>Explicitly execute a task on your PC
/statusCheck if the agent is busy or idle
/screenshotTake a screenshot of your PC screen
/stopEmergency stop for all running tasks
/historySee your recent task history
/clearClear task history
/sysinfoCheck CPU, RAM, and disk usage remotely
/clipRead your PC's clipboard content
/clip set <text>Write text to your PC's clipboard
/webcamTake a photo with your webcam
/getfile <name>Get a file from your Downloads folder
/streamRecord a 10-second screen video
/micRecord 10 seconds of microphone audio
/popup <msg>Show a message popup on your PC screen
/alias add <w> <cmd>Create a command shortcut
/alias listList all your shortcuts
/alias del <w>Delete a shortcut
/schedule HH:MM <task>Schedule a task for a specific time
/schedule listList all scheduled tasks
/schedule clearClear all scheduled tasks
/repeat every <sec> <cmd>Create a repeating task (runs every N seconds)
/repeat listView all active repeating tasks
/repeat stop <id>Stop a specific repeating task
/repeat stop allStop all active repeating tasks
[+] You can also send a voice message to transcribe and execute, or drop a file to save it to your Downloads folder.

[+] Examples

Things you can ask OmniCtrl.

Plain English commands — no syntax to memorize, no scripts to write.

[→]"Open Chrome and search for allotropes of carbon, then write the results in Notepad"
[→]"Open Calculator and compute 25 * 37"
[→]"Take a screenshot and save it to the desktop"
[→]"Open Notepad and write a short poem about computers"
[→]"Open File Explorer and navigate to the Documents folder"
[→]"Open Spotify and play my liked songs"
[→]"Create a new folder on my desktop called Projects"
[→]"Open Word and write a short letter to my friend"
[→]"Set up a task to check my system info every 5 minutes"
[→]"/Repeat every 30 seconds screenshot"

[x] Safety

Built-in safety features.

Automation can go wrong. OmniCtrl includes multiple layers of protection.

[x] Emergency Stop
Send /stop in Telegram or move your mouse to the top-left corner to halt all actions instantly.
[x] Confirmation Required
The agent always asks for your confirmation before executing any task. No surprises.
[x] Authorization
Only Telegram user IDs you whitelist in AUTHORIZED_USERS can control your PC.
[x] Self-Healing Errors
Errors are caught and retried intelligently. If recovery fails, the task stops safely.
[x] Rate Limiting
Prevents abuse by capping how many commands can be issued per minute.
[x] Task Timeout
Tasks automatically abort after 5 minutes to prevent infinite loops or hangs.

[+] FAQ

Frequently asked questions.

Does it work on Mac or Linux?
Currently OmniCtrl is designed for Windows 10 and Windows 11 only. Mac and Linux support may be added in future versions as the underlying automation libraries are extended.
Is my PC secure? Can anyone control it?
Only Telegram user IDs listed in your AUTHORIZED_USERS config can issue commands. The bot ignores all messages from unauthorized users. Your API keys are stored locally and never sent to third parties.
Do I need a paid API plan?
Both Gemini and OpenRouter offer free tiers that are sufficient for most personal use. Heavy or continuous automation workloads may eventually hit rate limits on free plans.
What happens if the agent clicks the wrong thing?
The self-healing system detects unexpected screen states after each action. If something goes wrong, it re-analyzes and tries a different approach. You can also send /stop at any time for an instant kill switch.
Can it work while I'm using my computer?
Yes — but tasks will interact with whatever is on screen. For best results, run tasks when you don't need the machine, or use the confirmation dialog to schedule them in advance.
How does the repeat feature work?
The repeat feature allows you to set up tasks that automatically execute at specified intervals. For example, you can use /repeat every 300 seconds sysinfo to check your system info every 5 minutes, or /repeat every 30 seconds screenshot to take a screenshot every 30 seconds. You can manage repeating tasks with /repeat list and /repeat stop commands.
How do I update to the latest version?
Navigate to your Computer-Agent folder in Command Prompt and run: git pull && pip install -r requirements.txt — then restart the program.

[+] OPEN SOURCE · FREE TO USE

Your PC, controlled from
anywhere in the world.

OmniCtrl is free and open source. Star the repo, fork it, and make it yours.

↓ Get Started ★ Star on GitHub