What you need

  • A Linux machine, x64 or arm64.
  • Node.js 22.19 or newer. Check with node --version.
  • An OmniRush account. Sign-up needs a GitHub account at least 7 days old, and you accept the Terms and Privacy Policy first.
  • A browser on any device for the first sign-in.

1. Get Node.js 22.19 or newer

Many distributions ship an older Node.js. If node --version prints something below v22.19, a version manager keeps a current one in your home folder, with no sudo for global npm packages:

curl -fsSL https://raw.githubusercontent.com/nvm-sh/nvm/master/install.sh | bash
# open a new terminal, then:
nvm install 22
node --version

Your distribution's own packages, or NodeSource's, work too, as long as the version is 22.19 or newer.

2. Install and sign in

install
$ npm i -g omnirush@latest
$ omnirush login
$ omnirush whoami

The omnirush package has no dependencies and no install scripts. npm adds one prebuilt package for your platform with the runtime and the agent engine.

omnirush login signs you in through your browser. When no browser can open, as over SSH, it prints a link to open on any device. If anything looks wrong, omnirush doctor checks sign-in, runtime and configuration, and says what to fix.

3. Your first task

~/code/api
$ cd ~/code/api
$ omnirush

That opens the full-screen agent in the current folder. Describe the task the way you would to a colleague, including how to check it:

Hard to verify
make /health better
Easy to verify
The /health endpoint returns 500 when Redis is
down. Make it return 503 with a JSON body, add a
test, and run the test suite.

The agent reads the code, runs commands and edits files. Git writes, destructive commands and sudo ask first. --yolo, or /yolo on in a session, lets every command run without asking: keep that for projects you trust, ideally inside the sandbox.

Start in the project folder, not your home folder. The agent finds the right files sooner.

Start in the project folder, not your home folder. The agent finds the right files sooner, and unrelated files stay out of the session.

4. Commands you will use every day

Everyday OmniRush CLI commands
CommandWhat it does
omnirushStart an interactive session in the current folder
omnirush -p "…"Answer once and exit: reply to stdout, errors to stderr, exit code 1 if the turn fails
git diff | omnirush -p "review this"Pipe input into a one-shot run
omnirush --continueResume your last session
omnirush -rPick a session to resume from a list
omnirush -m gpt-6-sol:highChoose a model and an effort level for this run
omnirush doctorCheck sign-in, runtime and configuration
omnirush updateUpdate now (updates also install in the background)

Inside a session: /subagents and /agents hand work to sub-agents, /jobs lists long commands running in the background, /compact shortens a long chat, and /voice lets you dictate.

5. Use it in scripts

Without a terminal (a script, a cron job, CI), omnirush answers once and exits, like -p. The exit code reflects the turn, so it slots into pipelines:

scripts
$ omnirush -p "Summarise what changed in the last 10 commits" > CHANGES.md
$ git diff main... | omnirush -p "List risky changes in this diff"

Treat scripted runs like any other session: never pipe secrets, private keys or logs with customer data into them.

6. The sandbox

Without the sandbox, the agent runs commands with your permissions. With it, a project's commands run in a Linux sandbox kept for that project, while your files and chats stay where they are.

sandbox
$ omnirush sandbox on      # this folder runs in the sandbox from now on
$ omnirush --sandbox       # or just this run
$ omnirush sandbox status  # which tool is used, and why

On Linux it uses bubblewrap, Podman or Docker. bubblewrap ships in OmniRush's Linux packages, so most machines need nothing extra. Packages the agent installs are still there next time, and omnirush sandbox off turns it off for the folder.

It can be on by default for new chats, depending on your system and OmniRush's current settings; omnirush sandbox status tells you. Our sandbox guide explains why it is worth turning on.

7. Long tasks, laptops and servers

While the agent works, OmniRush keeps the computer from sleeping with a logind inhibitor (systemd-inhibit) that ends with the task. The screen still turns off and locks as usual, and it is skipped over SSH, in CI and in containers.

omnirush keep-awake off turns it off, and omnirush keep-awake status shows whether it works on your machine. For long jobs on a remote box, run omnirush inside tmux or screen.

8. Prefer a window? The Linux desktop app

The desktop app runs the same agent with project workspaces and swarm agents. On Debian or Ubuntu, one line installs the newest .deb:

curl -fsSLo omnirush.deb https://omnirush.ai/download/linux-deb && sudo apt install ./omnirush.deb

For arm64, use linux-arm64-deb. AppImage and Arch packages for both architectures are in the download section, and every build is on GitHub with SHA-256 checksums. Not sure which to use? See desktop app or terminal.

Funding, settings and uninstalling

OmniRush is funded by brand ads, and data use described in its privacy policy.

Settings, sign-in tokens and local session history live in ~/.omnirush. npm uninstall -g omnirush removes the tool.

Questions

Which Linux distributions does the OmniRush CLI support?

Any distribution where Node.js 22.19 or newer runs, on x64 or arm64. The desktop app comes as .deb, AppImage and Arch packages.

Does the CLI work over SSH?

Yes. It runs in any terminal, including over SSH and in tmux. Run omnirush login and open the link it prints in a browser on any device.

Does the CLI support MCP servers?

No. MCP server connections are disabled in the OmniRush CLI.

More from the blog