Skip to main content
By the end of this guide your computer is a Reliant machine: an agent in one of your chats can read files and run commands in a repository on it. The desktop app already does this for you. Follow this guide when you use the browser, or when the computer is one you reach over SSH. The reliant daemon process makes it work. It dials out to Reliant and holds one stream open, so your computer needs no open inbound port (how a tool call reaches your machine).

Before you begin

  • A Reliant account you can sign in to in the browser or the desktop app.
  • The reliant CLI on the computer you want to connect (Installation). Run reliant --version to check.
  • A git repository on that computer for agents to work in.
1

Go to the repository

A daemon serves one workspace: the git worktree root of the directory you start it from. Every subdirectory of the checkout is the same daemon, and status, logs and stop look up the daemon the same way.
terminal
2

Start the daemon

The first time, the daemon has no credential, so it pairs the computer with your account. --name sets the name the app shows. Without it, the name is <instance-id>@<hostname>.
terminal
terminal
The daemon runs in the foreground. Leave this terminal open.
3

Approve the code in the app

Open the link on any device where you are signed in, or type the code into Add machine → Your own computer in the app, and approve the machine. Nothing has to reach back to the terminal, so this works the same over SSH.Once you approve, the daemon saves its own credential to ~/.reliant/daemon.json and connects. This credential is separate from a reliant auth login session, and later starts reuse it without pairing again (how the daemon holds its credential).
4

Check that it is connected

In a second terminal, from the same repository:
terminal
A connected daemon starts with Daemon is running and its gateway stream is established, then shows the stream’s uptime and stability. status exits 0 only when the stream is up and stable. While the code is still waiting for approval, it reports no daemon and exits 1:
terminal
The computer then appears under Settings → Machines in the app. Pick it on the new-chat screen, or move an existing chat to it with Connect a machine (choosing a machine). Ask the agent to run git status: the output comes from your computer.
5

Stop it when you are done

Press Ctrl+C in the daemon’s terminal, or run this from the repository:
terminal
terminal
stop waits until the process has exited. While the daemon is stopped, the app shows the machine as offline, and the next reliant daemon start reconnects with the saved credential.

What this does not do

  • It does not keep running on its own. The daemon stops when its terminal closes. To keep a computer available, run the daemon as a service: Run the Reliant daemon headless on a server.
  • It does not sandbox the agent by default. Commands run as your Unix user and can read what you can read. Whatever an agent reads or a command prints is stored with the chat and sent to your model provider (Privacy & security). credentials: none on a workflow node hides your credentials from that node’s commands (daemon security model).
  • It serves one repository. A second checkout or worktree needs its own daemon. reliant daemon ls lists every daemon on the computer.

Troubleshooting

Pairing codes are short-lived, and the daemon exits when its code expires. Run reliant daemon start again and approve the new code.
Someone declined the code in the app. Run reliant daemon start again. If you declined it by mistake, approve the new code.
When the CLI is signed in (reliant auth login), the daemon pairs only as that same account and does not save an approval from another one. Approve the code while signed in to the app as the CLI’s account, or sign the CLI in as the account you meant.
status, logs and stop look up the daemon by the directory you run them from. Run them from the repository the daemon serves, or pass --workspace <path>.

Next steps

Run the daemon headless on a server

A token instead of pairing, and a systemd service that keeps it up.

Your own computer

Every daemon start flag, several accounts, and TLS options.

How a tool call reaches your machine

The outbound stream, heartbeats, and what happens when it drops.

Troubleshoot an offline machine

When a chat says your machine is offline.