Atomesus Code docs

Atomesus Code is a full desktop IDE with an AI agent built in. It is a standalone app, not a plugin: you install it like any editor, open a folder, and work with the Atomesus Agent in a panel next to your code.

It is built on Code-OSS, the open-source base of VS Code, so the editor, terminal, Git, debugging and most extensions work the way you expect. On top of that, the agent can read your codebase, edit files, run commands with your permission, and plan larger changes before it makes them.

Atomesus Code runs on Windows, macOS and Linux. Pick the build for your machine.

Windows

1. Download the Windows installer (x64, or ARM64 for ARM devices).
2. Run it and follow the steps. It installs for your user account, no admin rights needed.
3. Open Atomesus IDE from the Start menu.

macOS

1. Download the build for Apple Silicon (M1 and newer) or Intel.
2. Open the download and drag Atomesus into Applications.
3. Launch it from Applications or Spotlight.

Linux

Use the package that matches your distribution:

sudo apt install ./atomesus_*.deb

sudo dnf install ./atomesus-*.rpm

chmod +x Atomesus-*.AppImage
./Atomesus-*.AppImage

1. Open a folder with File → Open Folder. The welcome screen lists the shortcuts you will use most.
2. Coming from VS Code or another editor? Use Import from another editor to bring over your settings, keyboard shortcuts and extensions. Settings you already have are kept.
3. Choose a model (see Models), then press Ctrl+L and ask the agent something. Summarize my codebase is a good first question.

Atomesus keeps its own data in ~/.atomesus-ide, separate from any other editor you have installed.

The agent works with the model you choose. Open the model picker in the chat box, or go to Settings, and add an API key for a provider:

Local models

To keep everything on your machine, run a model with Ollama or LM Studio and select it in the picker. Nothing leaves your computer in that setup.

Press Ctrl+L to open the Atomesus Agent panel. Describe a task or ask a question in plain language. The agent explores the files it needs, explains what it is doing, and shows every edit it makes.

The chats list beside the panel keeps your conversations per project. Start a fresh one with Ctrl+Shift+L, or open the Agents Window to see them in their own window.

The agent asks for your permission before it runs a terminal command.

Pick a mode from the selector in the chat box:

• Agent: reads, edits and runs commands to finish the task. Use it when you want the change made.
• Ask: read-only. It explores and explains your code but never edits files or runs commands.
• Plan: writes a step-by-step plan first and saves it in .atomesus/plans/. Edit the plan if you like, then click Build to have the agent carry it out.

• Type @ to attach context, such as a file or folder from your project.
• Type / to run a skill or one of your custom commands.
• Use the image button in the chat box to attach a screenshot or design.

Select some code, press Ctrl+K, and say what to change: add input validation, convert to async/await, write a docstring. The edit appears in place as a diff you accept or reject.

Every change the agent makes shows up as a diff in the editor, so you can read it before you keep it. Changes are applied with Fast Apply, which edits only the changed parts of a file; Slow Apply rewrites the whole file and is there as a fallback.

Checkpoints

The agent keeps checkpoints as it works. Click Restore checkpoint on an earlier message to roll your files back to how they were at that point.

Put a .atomesusrules or AGENTS.md file in the root of your workspace to tell the agent how your project works: the stack, conventions, commands to run, things to avoid. It is read at the start of every chat.

This is a Next.js app using TypeScript and Tailwind.
- Run tests with: npm test
- Use server components unless a file needs state.
- Never edit files in /generated.

Not sure what to write? Ask the agent: Create a .atomesusrules file for me.

A skill is reusable know-how the agent can load when it needs it, such as how to write a migration or how your team reviews pull requests. Each skill is a folder with a SKILL.md file.

• For every project: ~/.atomesus/skills/<name>/SKILL.md
• For one project only: .atomesus/skills/<name>/SKILL.md (overrides a user skill with the same name)

Run a skill with / in the chat, or let the agent pick it up when the task matches. You can also install skills from the Customize page; you always review an install before anything is written.

Save prompts you use often as commands. Each command is a Markdown file; its name becomes the slash command.

• Your commands: ~/.atomesus/commands
• Project commands: .atomesus/commands

Review the staged changes. List bugs first, then risky
patterns, then style nits. Keep each point to one line.

Type /review in any chat to add that prompt to your message.

Connect Model Context Protocol servers to give the agent new tools, such as your database, issue tracker or browser. Add them in .atomesus/mcp.json; a project's servers are approved for that folder only.

{
  "mcpServers": {
    "github": {
      "command": "npx",
      "args": ["-y", "@modelcontextprotocol/server-github"],
      "env": { "GITHUB_TOKEN": "your-token" }
    }
  }
}

Already set up servers somewhere else? Use Import servers from VS Code or another editor to bring them over.

All other shortcuts are under All keyboard shortcuts on the welcome screen, and you can rebind any of them.

Extensions install from Open VSX, the open registry used by Code-OSS editors, so most VS Code extensions are available. Language support for the common languages ships built in.

Remote SSH and WSL are included, so you can open a folder on a server or inside WSL and use the agent there just like locally.

Is this a VS Code extension?

No. Atomesus Code is its own desktop app. It shares the Code-OSS base with VS Code, but you install and run it separately.

Will my VS Code settings and extensions carry over?

Yes. Use Import from another editor on first run. Extensions come from Open VSX.

Where does my code go?

Only to the model provider you pick, when the agent needs it to answer. With a local model through Ollama or LM Studio, it stays on your machine.

What licence is it under?

Apache-2.0.