Executing commands

Run arbitrary commands without opening a full session:

thinr device exec <deviceId> "uname -a"

Output streams back to your terminal in real time, and the local process exits with the remote command's exit code, so exec composes with set -e, conditionals and CI steps:

thinr device exec store-14 "systemctl is-active app" && echo "app is up"

Commands run through the device's shell as the agent's user (root on a system install); see the command execution feature for the agent-side semantics.

JSON mode for automation

thinr device exec <deviceId> "df -P /" --json

Buffers output and emits a single envelope with { stdout, stderr, exitCode }. See JSON and automation for the scripting patterns.

Long-running commands and cancellation

Output streams as it's produced, so tail -f-style commands work; stop them with Ctrl+C, which terminates the remote command and exits with code 130. For following logs specifically, prefer thinr device logs -f, which adds units, files and product log sources.

Fleet-wide execution

Fan a command out to every active device of a product, in parallel:

thinr product exec edge-gateways "apt-get update -qq" -c 5 --timeout 60

You get a status table per device plus the outputs, with -c controlling concurrency (default 10), --timeout per device (default 30 s), --fail-fast to stop dequeueing on the first failure, and -g to scope by group. In JSON mode the envelope carries a summary block and per-device results[]; remember that partial failures don't change the exit code, so check summary.failed when scripting. Full flags in the thinr product reference.

For procedures with more than one step (check a condition, copy files, run, verify), reach for playbooks instead of chaining exec calls.

Older agents

--legacy switches to the non-streaming one-shot API for agents that predate streaming exec. You only need it if a very old agent rejects the default mode; updating the agent is the better fix.