Start here

Your first half hour with Coder, step by step: install it, log in, give it a first task in one of your repositories, read what it did, and answer a step that waits for you. Every command is here to copy.

You need a Mac, Linux, or Windows computer, a terminal, and a project in a Git repository. A coding agent you already use, such as Claude Code or Codex, helps but isn't required.

1. Install Coder

On macOS or Linux, in a terminal:

curl -fsSL https://openagents.com/cli/install.sh | bash

On Windows, in PowerShell:

irm https://openagents.com/cli/install.ps1 | iex

Open a new terminal, so it finds the new commands, and check:

coder --version

It prints coder and the version. If it says the command isn't found, see Troubleshooting. More on what gets installed and how updates work: Download and install.

2. Log in

coder login

Your browser opens openagents.com/device with an eight-letter code, such as BCDF-GHJK. Check that it's the same code your terminal shows, then approve. The code works for 10 minutes; run coder login again if it runs out.

Coder then asks once where this computer's chats should live:

Sync all my chats: they show on openagents.com and on your phone.
Keep chats on this computer: nothing is sent.

Type 1 or 2. You can change it later with /sync on or /sync off.

Logging in is optional. Without it, Coder still answers through Claude Code, Codex, or OpenCode if one of them is set up on this computer. Logged in, it can also use our models, with a free daily allowance (see Pricing).

3. Open your project

cd ~/code/my-project
coder

The screen is the chat, with the input at the bottom. At the input's top right is who answers: Auto until the first reply, then the agent and model, such as Auto → Claude (Haiku 5.5). At its bottom right are your folder and Git branch.

4. Give it a first task

Start with a question, so you can see how it reads your code. Type this and press Enter:

What does this project do? Name the three files I should read first.

Then a small change:

Add a short "Getting started" section to the README. Show me the diff.

Coder reads, edits, and runs commands in this folder on its own. You don't need to approve each edit.

5. Read what it did

Each step Coder takes shows as one line in the chat: files read, searches, each edit with its diff, and each command with its output.

Press Ctrl+O to open every step in full, and again to fold them back.
Click one of Coder's messages to see the model that wrote it, such as Claude (Haiku 5.5), for about three seconds.
Use PageUp and PageDown, or the mouse wheel, to scroll; Ctrl+End jumps to the bottom.
Run git diff in another terminal to see every change together.

Don't like the change? Type /rewind, pick your message, and Coder puts the files back as they were before it.

6. Answer a step that waits for you

A short list of steps that are hard to undo always waits for your yes or no: a production deploy, merging a pull request, reading or changing a secret, a payment, and a DNS change. To see one safely, make a file that looks like a secret, with nothing real in it:

echo "DEMO_KEY=not-a-real-key" > .env

Then ask Coder:

Run cat .env and tell me what it prints.

Coder stops and asks you one plain question, with the exact command.

Key	Does
`Y`	Allow it, this one time.
`N` or `Esc`	Say no.
`PageUp`, `PageDown`	Read the whole command.

Press N. The command doesn't run, and Coder carries on with the rest of the work without it. Delete the practice file when you're done: rm .env. Everything on the list, and how a project can tighten it, is in Safety.

7. Keep going

Ctrl+N starts a new chat. /resume opens an earlier one.
Shift+Tab, before a chat's first message, picks who answers.
/ lists every command; Ctrl+C quits, and your chat is saved.

Next, try a task-based guide: Fix a failing test, Use your Claude Code or Codex, or the keyboard shortcuts.

Next: Download and install.