Skip to main content
The cyberun CLI runs your team’s workflows from a terminal: submit a task, poll it to completion, download the result, and manage agents — without leaving the shell. It’s the scripting companion to the dashboard, useful for local experiments, cron jobs, and CI. The CLI ships on its own release track, separate from the dashboard, so its version moves independently.

​
Install

The installer drops the binary in ~/.cyberun/bin on Linux and macOS, and ~/.cyberun\bin on Windows — no sudo needed. Override with --install-dir; a root-owned target like /usr/local/bin triggers sudo only for the final move:
Prefer to grab the binary directly? Builds are published per platform under releases.cyberun.cloud/cli/<os>-<arch>/latest/, with a single shared cli/checksums.txt, for Linux, macOS, and Windows on amd64 and arm64. Verify the install:

​
Sign in

A fresh install has no API host configured yet, so point the CLI at the Cloud API first — either with an environment variable or the --api-host flag:
Then sign in:
auth login saves the host for you, so you only set it once. (Pass --api-host https://core.cyberun.cloud on the command instead if you prefer not to set the variable.) This pairs your machine with a team. The CLI prints a short code and opens your browser to confirm it while you’re signed in to app.cyberun.cloud; approve it there and the CLI stores the credentials locally. Pairing issues a device credential (dk-) for API calls and a companion agent credential (ak-) — the same companion agent described in Credentials, which lets the machine serve workflows. Check or end the session:
cyberun auth logout revokes the device credential server-side — which also revokes its companion agent — and then removes the stored credentials from this machine.

​
Run a workflow

The quickest path submits a workflow and waits for the result in one command:
run accepts either --slug or --id, takes parameters as a JSON object via --params, prints the finished task as JSON, and exits non-zero if the task fails or is cancelled. Add --download to fetch the result file when the task completes (defaults to <task-id>.bin, or pass --output <path>), and --poll-interval to change how often it checks (default 3s):
To submit without waiting, use workflow run instead — it returns the task id immediately:

​
Inspect workflows and tasks

workflow is also available as wf or workflows.
task is also available as tasks. download writes to <task-id>.bin if you omit --output, and only works once the task has completed.

​
Run a local agent

If the machine you’re signed in on has a GPU to share, turn it into an agent for the team:
This connects using the companion agent credential stored at login and streams tasks until you stop it (Ctrl-C). The gateway endpoint is derived from your API host; override it with --gateway-url if needed. List the team’s connected agents with cyberun agent list. agent serve is a foreground worker tied to your logged-in session — handy for ad-hoc sharing from a machine you’re already using. For a dedicated or headless GPU host you want a managed, always-on service instead: use the standalone agent installer described in Connect an agent.

​
Configuration

The CLI keeps its state in ~/.cyberun/cyberun (or $XDG_CONFIG_HOME/cyberun when that’s set) — cyberun auth login writes it for you, so there’s normally nothing to edit by hand. Use profiles to keep more than one account or team on the same machine. Every command takes --profile <name>; auth login creates the profile if it doesn’t exist. Point the CLI at a different API host or gateway without editing anything — pass a flag for a single command, or set an environment variable for the session: