Open Source

Your PC, from
your pocket.

OpenLoop is an AI agent that runs on your Windows machine and takes commands through Telegram. Keyboard, mouse, screenshots, app control — all from your phone.

PowerShell
powershell -Command "iwr -useb https://openloop.sh/install.ps1 | iex"
DownloadSource Code

Capabilities

What it actually does

Snapshot → Plan → Execute
The agent reads your target app's entire UI tree, sends it to an LLM, gets a step-by-step plan, and executes it — autonomously.
Telegram as your remote
Send instructions from Telegram. The agent receives them, picks the right process, runs the task, and replies with status updates.
Seven LLM providers
Gemini, OpenAI, Claude, DeepSeek, Moonshot, OpenRouter, or Ollama for local inference. Pick the model that works for you.
Native UI automation
Uses Windows UI Automation APIs first, with mouse and keyboard simulation as a fallback. Handles dropdowns, trees, lists — the real stuff.
Screenshots on demand
Take a screenshot from your phone and get it sent back instantly. See exactly what the agent is doing before and after each action.
Single binary, ~1MB
Written in C++ with ImGui and DirectX. No runtime dependencies, no Electron, no Python. Just runs.

Setup

Under a minute. Seriously.

1

Run one command

Open PowerShell and paste:
iwr -useb https://openloop.sh/install.ps1 | iex
It downloads the binary, asks for your Telegram bot token, detects your chat ID, and sets up auto-start.

2

Message your bot

Open Telegram on your phone and send anything to your bot. OpenLoop detects the conversation and starts listening for commands.

3

You're in control

Tell it what to do — open apps, type text, click buttons, grab screenshots, run automations. It reads the screen, plans the steps, and executes.

In action

You type. It does.

you

open spotify and play my liked songs

openloop

→ attaching to Spotify.exe

→ found navigation: "Liked Songs"

→ clicking "Liked Songs"

→ pressing Shuffle Play

→ done.

you

skip this track

openloop

→ pressing Next button

→ now playing: "Something New"

❯

MIT Licensed

Read the code.
Run it yourself.

The full source is on GitHub. Star it, fork it, open a PR, or just read through and learn how it works.

View on GitHubDownload Latest