gpu●aiDocs
CH·02CLI reference

gpu instances get

Show details for one instance

Show details for one instance

Synopsis

Fetches a single GPU instance by ID and renders its full detail view, including the Connection block (hostname, port, ssh_command) when the instance is running. A Disk row shows the filesystem size GPU.ai configured for the launch (disk_gb in JSON); it is omitted on machines whose disk is provider-determined (fixed-disk capacity). Errored instances additionally show a Status Reason row (status_reason in JSON) explaining why the launch failed, e.g. a provisioning timeout or stall.

A Last reachable row (last_reachable_at in JSON) shows when GPU.ai last verified the SSH path to the instance end-to-end. The row is omitted until the first successful check — absent means "not probed yet", not "unreachable". If it stops advancing while Status stays running, the machine is up but SSH inside it is not: the usual causes are a crashed SSH daemon or an out-of-memory/disk-full instance.

When GPU.ai has *confirmed* a tunnel outage (several consecutive failed probes, not a single blip), Status shows unreachable instead of running and an Unreachable since: row (unreachable_since in JSON) carries the confirmed outage start. This is a display state, not a lifecycle state — the instance is still provisioned and metered, but you are not billed for the confirmed-dead window: it is credited automatically on recovery, or billing is closed back to the outage start if the outage persists and the instance is failed. See docs/gpu-instances.md for the full semantics.

An Auto-terminate row (auto_terminate_hours / auto_terminate_at in JSON) appears on instances launched with --auto-terminate-hours: it shows the deadline the platform will terminate the instance at, with the requested limit in parentheses (2026-09-02T18:00:00Z (8h limit)). Before the instance is running there is no deadline yet, so the row reads 8h after it starts running — the clock starts at running, not at create time. The row is omitted entirely for instances with no limit. After the limit fires, Status Reason records that it did.

For instances with a web app or web console, the detail view includes the App Login (basic-auth user/password). The CLI fetches it with the explicit ?include=credentials opt-in on GET /v1/instances/{id} — bare API reads omit the password so instance responses can be logged safely.

gpu instances get <instance-id> [flags]

Examples

gpu instances get gpu-abc123

# JSON for scripting
gpu instances get gpu-abc123 --output json

Options

  -h, --help   help for get

Options inherited from parent commands

      --api-base string   API base URL (override with GPUAI_API_BASE env) (default "https://api.gpu.ai/v1")
      --debug             Enable debug logging to stderr
  -o, --output string     Output format: table|json (default table on TTY, json otherwise)

SEE ALSO

← The gpu CLI