How it works Features Source available Download FAQ GitHub Download

AI agents that actually use
your computer.

They see your screen, click, type, schedule, code, and get the job done. Autonomously.
No technical skills needed. Any model. Runs locally.

Download for macOS Download for Windows
Free* and source available
Any model, cloud or local
1-click install, macOS & Windows

* Free for personal use, and for commercial use by organizations under $500,000 in annual revenue.

Screen vision Mouse control Keyboard input Shell commands Python code Scheduled tasks Image reading Agent Skills Telegram Multiple agents Any LLM Runs locally Screen vision Mouse control Keyboard input Shell commands Python code Scheduled tasks Image reading Agent Skills Telegram Multiple agents Any LLM Runs locally
How it works

From download to a working agent in a minute.

A guided first run takes care of everything. No setup, no configuration, no technical skills.

01

Install

Download, open, grant the permissions it asks for. Add a key from Anthropic, OpenAI, DeepSeek, GLM or any compatible provider, or connect a local model through LM Studio.

02

Give it a job

Describe the outcome in plain language, attach files or folders if it helps: "Fill in this week's expense report from the receipts in Downloads."

03

Watch it work

The agent takes a screenshot, clicks, types, runs code, checks the result, and keeps going until it is done. You see every step live and can pause at any time.

Use cases

Tasks that need hands.

Back-office forms

"Enter the 14 invoices in this folder into the accounting software, one by one, and flag any that don't match the purchase order."

Apps that don't talk to each other

"Copy today's new orders from the web dashboard into the inventory spreadsheet and update the stock column."

Routines

"Every Monday at 9, export last week's report from the admin panel, save it as PDF, and send it to me on Telegram."

Research

"Find three suppliers for this part, compare their prices and lead times, and put it all in a document on my desktop."

Development

"Clone this repo, run the test suite, and tell me which tests fail and why."

From your phone

On Telegram, from the train: "Is the build green? Send me a screenshot."

Any app

If it is on your screen, it can use it.

No plugins, no API keys per app. The agent works through the same screen, mouse, and keyboard you do.

Chrome
Safari
Firefox
Excel
Word
PowerPoint
Gmail
Outlook
Slack
Teams
Notion
Figma
VS Code
Terminal
Finder
Explorer
Salesforce
HubSpot
SAP
QuickBooks
Shopify
Jira
Zoom
WhatsApp
Photoshop
Canva
Your ERP
Your CRM
That internal tool
+ Anything with a window
Features

Hands, not words.

Most AI assistants can talk about your computer. Very few can use it. Alent.Red agents work the way you would: look at the screen, act, check the result, keep going.

Sees your screen

The agent takes screenshots of any connected display and understands what is on it: text, buttons, menus, images. It can also open image files you point it to.

Mouse and keyboard

Native control on macOS and Windows, no automation library in between. Moves, clicks, shortcuts, typing that respects your keyboard layout. Works in any app you have open.

Runs code and commands

When clicking is not the fastest way, the agent runs shell commands or Python in an isolated environment, installs the packages it needs, and reads the output.

Schedules itself

Ask for something in an hour, tonight, or every Monday morning. The agent sets a reminder for itself, wakes up, and does it.

As many agents as you need

Each agent has its own name, prompt, model, and set of capabilities. One sorts your invoices, another runs your reports, another answers you on Telegram.

Any model

Anthropic, OpenAI, LM Studio for models running on your own machine, or any compatible endpoint such as DeepSeek or GLM. Bring your own key. Switch models per agent, no lock-in.

Runs locally

The app and its engine run on your machine. Conversations live in a local database, files are encrypted, API keys stay in your OS keychain. The only network calls are the ones to the model you chose.

Reach it from your phone

Connect an agent to Telegram and send it tasks from anywhere. It works on your computer at home and replies with the result. Other chat platforms are on the way.

Skills, open standard

Reuses the Agent Skills already on your machine from Claude Code or Codex, and any skill you write in the open SKILL.md format. Teach an agent once, every agent can use it.

Why Alent.Red

Source available, one-click install.

The freedom of the tools developers build for themselves, with the install of an app anyone can use.

Source available

The desktop app and the engine behind it are both on GitHub. Read the code, build it yourself, open an issue, send a pull request. Nothing hidden, no closed component.

One-click install

No terminal, no Docker, no config files, no Python to set up. Download the app, open it, and a guided first run creates your first agent. Everything it needs ships inside.

Free and yours

No account, no subscription, no usage cap. Free for personal use and for organizations under $500,000 in annual revenue; above that, a commercial license applies. The only other cost is your model provider, or nothing with a local model.

FAQ

Common questions.

Your conversations, history, configuration, and API keys stay on your device. The one thing that leaves is what the agent sends to the model you chose: your messages and the screenshots it takes, the same way any AI app talks to its model. With a local model in LM Studio, nothing leaves at all. There is no telemetry on your data.

Anthropic (Claude) and OpenAI (GPT) with your own API key, any model running in LM Studio on your machine, and any endpoint compatible with the Anthropic or OpenAI APIs, which covers DeepSeek, GLM, and most hosted models. Each agent picks its own model, so you can mix them.

It takes a screenshot, decides what to do, and sends real mouse and keyboard events through the operating system, with no third-party automation library. On macOS it asks for Screen Recording and Accessibility permissions; on Windows it runs as administrator so it can interact with every window. When clicking is not the best way, it can also run shell commands and Python.

You see every action live, you can pause a conversation at any moment, and the app can hide its own window from screenshots. Still, an autonomous agent can make mistakes. Start with tasks you can check, keep it away from actions you cannot undo, and give each agent only the capabilities it needs.

It's free for personal use, and free for commercial use by organizations under $500,000 in annual revenue, with no paid tier, no usage limit and no account. Above $500,000 in annual revenue, a commercial license applies: write to hello@alent.red.

macOS 11 or later, or Windows 10 or later. The app bundles its own engine, so there is nothing else to install. You need an API key from a model provider, or LM Studio running a local model.

Yes, through Telegram. Create a bot, paste its token into a gateway, choose which usernames are allowed, and the agent answers from your computer wherever you are. Other platforms will follow.

Download

Free. No account. No credit card.

Free for personal use, and for commercial use by organizations under $500,000 in annual revenue. Pick your OS; the app ships with everything it needs.

Download for:

macOS Windows

macOS 11 or later. Windows 10 or later.

Works with:

Claude (Anthropic)
GPT (OpenAI)
LM Studio (local models)
DeepSeek
GLM
+ Any compatible endpoint