Skip to main content
This guide gets a chat’s machine connected again and the run it stopped carrying on. Work through it when a chat says its machine is offline, a cloud machine stays on Waking up…, or a run has paused. Every machine runs a daemon that dials out to Reliant and holds one stream open. reliant daemon status reports that stream, not just whether a process exists, so it tells you which link is broken (how a tool call reaches your machine).

Before you begin

  • The name of the chat’s machine. The chat shows it, and Settings → Machines lists all of yours.
  • For your own computer or server: a terminal on it, and the reliant CLI (Installation).
1

Read what the chat says

The message tells you where to look:A run pauses only after three steps in a row failed to reach the machine over at least two minutes, so a short drop never stops it.
2

On a cloud machine, check Settings → Machines

Open Settings → Machines and look at the machine’s status. Click Resume on a suspended machine. A machine that failed to start, or stays on Waking up… for many minutes, can be restarted from its page, or deleted and created again (waking and suspending). A cloud machine’s logs aren’t available to you, so the next steps are for your own computer.
3

On your own computer, check the daemon's stream

From the repository the daemon serves (or with --workspace <path>):
terminal
status exits 0 only when the stream is established and has held for two minutes. With no daemon running:
terminal
Start it with reliant daemon start (connect your computer). A daemon that runs but has no credential says so:
terminal
If you run several daemons, reliant daemon ls lists every instance on the computer with its stream state:
terminal
4

Read the daemon's log

reliant daemon logs shows the last 50 lines and follows, but reads only the last 1 MB. For anything older, search the file itself, logs/tools-daemon.log beside the record status printed:
terminal
Session ended; reconnecting and Gateway stream is dead are drops the daemon recovers from by itself. Rejected: credentials not accepted — retrying, repeated, usually means the credential was revoked or the machine removed: go to the next step. Fatal error — not reconnecting means the daemon stopped, and the line says why.
5

Re-register a daemon whose credential is refused

From the repository, stop the daemon, move its credential aside, and pair the computer again:
terminal
daemon start prints a pairing code to approve in the app. On a server with no browser, give it a new token with --token instead (run the daemon on your own server).
6

Check that the machine is back and the run resumed

Run reliant daemon status again. It now starts with Daemon is running and its gateway stream is established and exits 0, and the machine shows as connected in Settings → Machines.A paused chat resumes on its own when its machine connects. To retry sooner, send a message in the chat, or resume a run from the CLI with reliant workflow resume <execution-id>.

What this does not do

  • Tool calls are not queued for an absent machine. A call made after the daemon has been gone for 90 seconds fails at once.
  • A crash takes up to 90 seconds to show. A clean daemon stop shows as disconnected immediately; a crashed daemon looks connected until 90 seconds pass without a heartbeat.
  • Tool calls don’t wake a cloud machine. Sending a message, or the check a run makes when it starts, wakes it.

Troubleshooting

A tool call reached no daemon. Check the machine as above.
A workflow’s daemon: names a machine that isn’t connected, or that matches none of yours. Check the name and the kind in the workflow (choose which machine runs each step).
The run waited for its machine when it started, and the machine never connected. In a chat the message ends Your message is queued and will be sent as soon as your machine connects, so start the machine and the message is delivered.
A daemon for this workspace is already running. Use it, or run reliant daemon stop first. To run a second daemon on purpose, give it its own --data-dir.
Two processes started with the same identity, and only one may hold it, so this one stopped. Run reliant daemon ls, stop the copy you don’t want, and keep one.
The stream connects and drops again. Look in the log for the Session ended; reconnecting lines and their errors. A network that blocks long-lived HTTPS connections to gateway.reliantapi.com causes this.

Next steps

How a tool call reaches your machine

Heartbeats, reconnects, and exactly when a run pauses.

Choose which machine runs each step

Pin a workflow, or one node, to a machine with daemon:.

Run the daemon on your own server

Keep a machine connected across reboots with systemd.

Daemon security model

What the daemon’s credential is and where it lives.