MachineMachine
DesktopGet started

How Machine works

The few ideas behind everything in Machine Desktop, explained in plain words. Read this first if you are new.

You don't need to be technical to use Machine. You do need about ten ideas, and once you have them, the rest of the app makes sense. This page explains each one, with an example.

Bots

A bot is an AI helper with a job. You might have one bot that does research, one that writes code, and one that handles your inbox. You can have as many as you like.

Each bot has:

  • A name and a face, so you can tell them apart in the sidebar.
  • Standing instructions, which Machine calls its Soul. This is a description of who the bot is and how it should work, for example "You are a careful researcher. Always list your sources. Keep summaries under one page." The bot follows these in every conversation, so you don't have to repeat yourself.
  • Skills: extra know-how for specific tasks, like making slide decks or working with spreadsheets.
  • Memory: notes the bot writes for itself so it can remember things about you and your work.
  • Connected apps and secrets: access to outside tools, like Gmail, and the passwords or keys it needs for them.
  • A default model: the AI "brain" it uses. More on that below.

Bots and projects

On Machine, every bot is a project. Your bots are listed under Bots in the sidebar, but the button to make one still says New Project. "Bot" and "project" mean the same thing in the desktop app. If you use the Machine web app, the projects you see there are the bots you see here.

Example: you make a bot called Scout whose Soul says it is a market researcher. Every time you ask Scout something, it researches like a market researcher would, without you explaining that again.

Threads

A bot's thread in Machine Desktop: your request, the steps the bot took, and its reply with a saved file

A thread is one conversation with a bot about one piece of work. Think of it as one email thread, or one ticket.

You can have many threads with the same bot, for example one thread per client or per task. Each thread keeps its own history, its own files, and its own computer (see next). Starting a new thread is like giving the bot a fresh desk for a new job.

Tip: start a new thread when you start a new task. Long threads get slower and more expensive, because the AI re-reads the whole conversation every time you send a message.

The bot's computer

This is the big idea. Every thread runs on its own computer in the cloud. It is a real computer that Machine runs for you, with its own files and programs and an internet connection.

When you ask a bot to do something, it uses that computer to do it. It might write files, run programs, install tools, browse websites, or start a website of its own. Everything it makes stays on that computer, where you can see it and download it.

  • It sleeps when nobody is using it and wakes up when you send a message. Waking takes a few seconds. Files are kept while it sleeps.
  • It costs money while it is awake, so it goes to sleep on its own when the thread is quiet. You can also stop it yourself.
  • Deleting a thread deletes its computer, including any files on it. Download anything you want to keep first.

You can look inside at any time. See The bot's computer.

This PC

The bot's own computer: the files a thread made, in the computer panel

Sometimes you want a bot to work with things on your own computer: a folder of invoices, a program you have installed, or a website you are signed into. For that, you switch a thread from Cloud to This PC.

You choose exactly which folder it may use, you can make it ask before every action, and you can stop it at any moment with one click or one keyboard shortcut. See This PC.

Turns and tools

When you send a message, the bot takes a turn: it thinks, does some work, and replies. One turn can involve many steps, like reading a file, searching the web, running a program, and writing a report.

Each of those steps uses a tool. A tool is an action the bot can take, like read a file, run a command, or search the web. In the chat, tools show up as small chips, so you can see exactly what the bot did. You can turn these off if you only want to see the replies.

While a turn is running, you can Stop it at any time.

Models

A model is the AI that does the thinking, made by a company like Anthropic (Claude), OpenAI (GPT), or Google (Gemini). Different models have different strengths and prices. Bigger models are usually smarter, slower, and more expensive.

Every bot has a default model, and you can change the model for any thread. If you are not sure, keep the default. See Models and approvals.

Credits and cost

Machine is paid for from your balance, shown at the bottom of the sidebar. Your balance goes down when:

  • a model does work. Each turn costs a little, depending on the model and how much the bot reads and writes.
  • a bot's computer is awake.

Your balance refills from your plan and from daily credits, and you can top it up on the web. To see where it went, open Settings → Usage. Each thread also shows its own cost in the chat header. See Settings and usage.

Tips to spend less: start new threads for new tasks, stop computers you are not using, and use a smaller model for simple jobs.

Approvals

Bots can do real things, so you decide how much freedom they get. With Ask for approval, a bot stops and asks before it runs a command or changes a file, and you click Allow once or Deny. With Approve for me, it goes ahead with most things, but still asks before anything is sent or deleted. See Approvals.

Tip: while you are learning, keep Ask for approval on. You will see exactly what the bot wants to do before it does it.

Group chats and teams

A group chat puts several bots in one conversation, so they can work together. For example, a researcher finds facts, a writer turns them into a blog post, and an editor checks it. You @mention a bot to talk to it directly.

A team is just a way to group bots in the sidebar, like a folder, for example Marketing or Client A.

Your group chats and teams follow you: Machine Desktop on your other computers and the Machine phone app show the same ones.

See Group chats.

Automations

An automation (also called a routine) asks a bot to do something on a schedule, like "every weekday at 8am, summarize yesterday's news about our competitors". It runs in the cloud, so your computer can be off. See Automations.

Words you will see

WordMeaning
BotAn AI helper with its own instructions, skills, and memory. Also called a project.
ProjectMachine's name for a bot.
SoulA bot's standing instructions: who it is and how it works.
ThreadOne conversation with a bot, with its own history and computer.
TurnOne round of you sending a message and the bot working and replying.
Tool / tool callOne action a bot takes, like reading a file or running a command.
Computer / cloud computerThe computer in the cloud where a thread's work happens.
This PCYour own computer, when you let a bot work on it.
WorkspaceThe main folder on a bot's computer, where it keeps its work.
ModelThe AI that does the thinking, like Claude or GPT.
EffortHow long a model thinks before it answers. More effort is slower and costs more.
ContextHow much of the conversation the model can keep in mind at once.
Credits / balanceWhat you pay with.
ApprovalYour yes or no before a bot does something.
SkillA saved set of know-how a bot can use for a type of task.
MemoryNotes a bot keeps to remember things between conversations.
Connected appAn outside service, like Gmail, that a bot can use for you.
SecretA password or API key a bot needs, stored so nobody can read it back.
API keyA password-like code that lets a program use a service.
Group chatA conversation with several bots at once.
TeamA group of bots in the sidebar.
RecipeA ready-made team of bots for a job, which you can add in one click.
Automation / routineA task a bot does on a schedule.
DeploymentA website or app a bot built, published on the internet.
KickA button that unsticks a bot that seems frozen.

Ready? Continue to Your first 15 minutes.

On this page