Skip to content

Orchestrate SLURM jobs
like code.

A Python toolkit for submitting, monitoring, and chaining compute jobs on HPC clusters — with a web UI, an MCP server, and YAML workflows that feel like CI.

Install Quickstart

01 / Capabilities

  • Simple submission

    One-line SLURM submits with conda, venv, Apptainer, and Pyxis wiring included.

  • Resource control

    Declare nodes, GPUs, memory, and partitions — or let the defaults ride.

  • Workflows as YAML

    Typed jobs with depends_on, retry, and Jinja-templated args.

  • Parameter sweeps

    Matrix cross-product over hyperparameters. Per-cell tracking, bounded SSH pool, Web UI progress.

    Parameter sweeps

  • Live monitoring

    Poll state, fan out Slack deliveries, and snapshot GPU utilization.

  • rsync project sync

    Delta-transfer your repo to any cluster via ProxyJump-aware SSH.

  • Web UI

    Browser dashboard for queue, DAG visualization, run history, and sweep detail pages.

  • MCP server

    Claude Code and other MCP clients drive srunx over stdio — including run_workflow(sweep=..., mount=...).

  • Jinja templates

    Every sbatch script is a rendered template you can override.

02 / In action

srunx sbatch --wrap "python train.py" --gpus-per-node 2 --conda ml_env

srunx sbatch --wrap "python train.py \"
  --container "runtime=apptainer,image=pytorch.sif,nv=true"

name: ml_pipeline
jobs:
  - name: preprocess
    command: ["python", "preprocess.py"]
    resources:
      nodes: 1

  - name: train
    command: ["python", "train.py"]
    depends_on: [preprocess]
    resources:
      gpus_per_node: 1
      memory_per_node: "32GB"
      time_limit: "8:00:00"
    environment:
      conda: ml_env

03 / Documentation

  • Tutorials

    Start here. Install, submit your first job, tour the Web UI, and set up MCP.

    Tutorials

  • How-to guides

    Task-oriented recipes for workflows, monitoring, sync, and MCP usage.

    How-to

  • Reference

    Python API, Web UI endpoints, and the full MCP tool surface.

    Reference

  • Explanation

    Architectural decisions, design trade-offs, and how the pieces fit together.

    Explanation