A clean terminal dashboard for Slurm. Find your projects, see why jobs wait or fail, read their logs, and prepare the next run.

Warning: vibecoded for my own use, I don't have a plan to maintain it.

You need Node.js 22 or newer and Git on the computer where you run seejobs. The cluster needs Python 3 and its usual Slurm commands. If you use SSH, make sure ssh YOUR_ALIAS works first.

git clone https://github.com/jmillanacosta/seejobs.git
cd seejobs
npm install
npm install --global .
seejobs

First use starts setup. Enter:

1. SSH host: your usual SSH alias or hostname. Enter local if you are already on a cluster login node.
2. Slurm username: the user whose jobs you want under “My jobs”.
3. History days: press Enter for 7 days.
4. Refresh interval: press Enter for 1 second. You can choose 1–300 seconds.
5. Projects: press Enter to group by working directory, or enter script.
6. Guide folder: optional. Enter a local folder of Markdown cluster guides, or leave it empty.

No global install permission? Run npm start from this folder instead. Install Node with a user-owned version manager if your system's global npm folder is not writable. You do not need Node on the remote cluster when seejobs connects over SSH.

The Slurm page opens first. It combines live jobs and history, including failures. Script groups start expanded. Names wrap, and the queue shows submission dates. Wide terminals show more columns. Pending is yellow, successful is green, and failed is red.

Use Up/Down to select a row. Press Enter to open it. Esc returns to Slurm when no form is open. The footer shows the actions for the current page.

F opens filters for user, partition, status, GPU request and project. Choose a field, press Enter, choose its value, then press A to apply. “All visible users” requests other users' jobs; their file permissions still apply. K searches job fields. R reloads. Q quits. Uppercase and lowercase keys work the same way.

On D, choose a project and press Enter to see its jobs. G switches between working directories and script paths. Full paths keep projects with the same name separate. X clears the project filter. On S/H, G toggles script groups; Enter, Space or Left/Right folds a selected group.

Click a job to open it. Click a panel to focus it. Tab switches focus between the job list and the document. Page Up/Down and Home/End work in lists and documents. Run seejobs --no-mouse if you prefer your terminal's text selection.

Select a job, then press L. / toggles the split view. Left/Right selects a file. V switches between recent output and the whole file. Whole files load in chunks; [ / ] loads the previous/next chunk. Navigation keys scroll within the chunk.

For older jobs, seejobs tries the retained batch script, then the current script file, then matching files in the job's directory and its log/logs folders. Inferred paths are labelled. Files that moved, were deleted, or are private may not be readable. Error excerpts are clues, not a guaranteed diagnosis.

Press E on a job to recover its script. Or press B to open a script by path or create a new job.

1. Select a field and press Enter to edit it. Check the submission directory. For a new job, enter the command to run.
2. Read the field help and live cluster limits. Press ? for more help, including imported guide advice. An empty option uses the scheduler default.
3. Press V to review the complete script.
4. Press Enter to check it with sbatch --test-only. This does not submit it.
5. Press Enter again to open the final confirmation, then Y to submit.

Editing prepares a new submission and keeps the original file. The check does not guarantee that your program will succeed or that resources will be available. Check array and dependency fields before sending. Cancellation also requires confirmation and uses your normal Slurm permissions. Actions are never automatically retried after connection errors.

Jobs and selected details reload every 1 second by default. Change this in Settings (,). Requests do not overlap; slow servers reduce the effective rate. Metrics poll no faster than the configured refresh interval or Slurm's known collection interval, whichever is longer. The clock redraws twice per second; unchanged job snapshots do not rebuild the full interface.

Metrics show estimated CPU cores from counter changes, reported current task memory, CPU efficiency, time used versus the limit, and final step accounting. GPU counters appear only when Slurm supplies them. A GPU allocation does not prove GPU use. The largest task's peak memory is not whole-job peak memory. Missing measurements stay unknown. Time-limit use is not task progress.

Charts cover the current viewing session. History comes from Slurm, is limited to 5000 accounting records, and depends on cluster retention and privacy settings. seejobs is not a permanent metrics recorder. On Nodes, A requests account fair-share information and reservations. On Overview, W requests start and priority details. These extra requests run on demand.

Press , in the app, or run seejobs --config, to change setup. Settings are saved with owner-only permissions in ~/.config/seejobs/config.json (or $XDG_CONFIG_HOME/seejobs/config.json). SSH credentials stay in your SSH configuration.

Guide advice is optional and stays in that private file. It includes the source file and section. Examples are never applied to jobs automatically. Saving a guide folder imports it again; clearing it removes the advice.

seejobs --import-guides /path/to/markdown-guides
seejobs --host local --user YOUR_USERNAME
seejobs --days 30
seejobs --once  # print one JSON snapshot

SEEJOBS_CONFIG, SEEJOBS_HOST, and SEEJOBS_USER can override the settings path or connection. Imported guide advice stays private and is never executed.

Ink renders the terminal interface with React. React manages its state. wrap-ansi wraps terminal text. Node.js handles SSH and settings; a small Python standard-library helper reads Slurm data on the server. No database, web service, or extra Python package is required.

The in-app ? Help page explains what each Slurm command contributes, how to interpret metrics, and which limitations remain.

For development: run npm run build and npm test. npm run test:smoke performs read-only live checks using your private settings. Use npm link in a source checkout to run local edits immediately.