Jobs and billing
Configure the released CLI
Install envloop-cli as described in Installation. The commands here were checked against release 0.1.1, whose query commands differ from the envloop source CLI's train submit/status.
export ENVLOOP_BASE_URL=https://api.envloop.ai
export ENVLOOP_WORKSPACE_ID=YOUR_WORKSPACE_ID
export ENVLOOP_API_KEY="$(cat /absolute/private/platform.key)"
Use your workspace and existing private KEY file. Explicit global --api-url and --workspace-id options override these defaults and must appear before train.
Inspect jobs
el train job list --limit 25
el train job list --status running --limit 25
el train job status TRAIN_JOB_ID
Output is JSON. Use the returned train_job_id, not an Agent job_id or Trial ID. If the list returns next_cursor, continue with:
el train job list --cursor NEXT_CURSOR --limit 25
Train Job status describes training progress; a running cloud machine does not prove the model is ready. A successful terminal status does not independently verify provider resource release or checkpoint downloads. Record the job/session IDs and inspect the script's finish result.
List checkpoints
el train checkpoint list --limit 25
el train checkpoint list --session-id SESSION_ID --limit 25
el train checkpoint list --run-id RUN_ID --limit 25
Use session/run IDs from training details or script records. Do not substitute train_job_id. Only registered artifacts appear. The current endpoint returns at most 100 checkpoints and has no cursor pagination. See Checkpoints and recovery for save, restore and download verification.
Read estimated usage
el train billing usage 2026-10-10 2026-10-11
Dates use YYYY-MM-DD, ordered as starting_on and ending_before. The current platform validates dates but does not filter usage by that range. estimate: true does not represent a settled invoice. Compare usage against the exact session and provider records; the recorded GPU smoke has an unresolved rate/card attribution discrepancy.
Troubleshoot
| Symptom | Check |
|---|---|
| Command is missing | command -v el, package version and el train --help; source and released CLI have different commands |
| 401 | Correct platform machine KEY, expiration and revocation; SDK scripts additionally require credential-command configuration |
| 403 or resource not found | Workspace membership, permissions and submitting user; another user's KEY does not automatically own the run |
| 503 | Ask the administrator to check the configured training service; installing the CLI does not deploy a backend |
| Checkpoint list is empty | Correct session/run and whether save, publication and registration completed |
| Timeout | Network/proxy, provisioning or backend failure; timeout does not prove the GPU stopped |
SDK failures should be followed by an explicit errored/interrupted finish where possible. The SFT example attempts this in its cleanup path; a failed finish requires administrator follow-up for the exact session. Do not assume read-only access implies managed submission or cancellation is enabled.