Rube

Rigs that build themselves.

The automation platform that AI agents can actually use. No drag-and-drop. No YAML. Describe what you want, and Claude builds a working pipeline.

Why Rube exists

You want automation, not a chore.

No SaaS. No YAML. No drag-and-drop. No infrastructure to deploy.

You describe what you want. An agent builds it. You get a folder of shell scripts you can read, run, git commit, and forget about.

Automation you own. Code you didn't have to write.

What you get

stale-tickets/
├── steps/
│   ├── fetch.sh        # Get data from Jira
│   ├── transform.sh    # Process with DuckDB
│   └── notify.sh       # Post to Slack
├── out/                # Outputs live here
└── .beads/             # DAG state (dependencies)

Shell scripts. Files. Git-committable. Debug with bash steps/fetch.sh.

Watch it build itself

You: I need to find Jira tickets that haven't been updated
     in 7 days and post them to Slack every morning.

Claude: I'll build a rig for that.

        $ mkdir stale-tickets && cd stale-tickets
        $ rube init
        $ rube step add fetch -d "Get stale tickets" -s '...'
        $ rube step add format -d "Format message" -s '...'
        $ rube step add notify -d "Post to Slack" -s '...'
        $ rube dep add format --needs fetch
        $ rube dep add notify --needs format

        Done. Run it with: rube run

You: Run it.

Claude: $ rube run
        ✔ fetch (2.1s)
        ✔ format (0.1s)
        ✔ notify (0.3s)

        ✓ Completed 3 steps (2.5s)

No UI. No tinkering. The agent did it.

150 integrations built-in

Rube uses Steampipe under the hood. Query any API with SQL:

SELECT key, summary, assignee
FROM jira_issue
WHERE updated < now() - interval '7 days'

Available: jira · slack · github · gitlab · postgres · aws · gcp · azure · pagerduty · datadog · okta · zendesk · 140+ more

Transforms are DuckDB. Secrets are psst. All SQL. All stuff the agent already knows.

Parallel by default

Steps without dependencies run in parallel:

$ rube run
✔ fetch-jira (2.1s)      # ─┐
✔ fetch-github (1.8s)    # ─┼─ parallel
✔ fetch-pagerduty (0.9s) # ─┘
✔ merge (0.3s)           # waits for all three
✔ notify (0.1s)

No configuration. If it can run, it runs.

Get started

# Install
npm install -g rube

# Set up dependencies
rube install

# Then either:

# Option A: Ask Claude to build it
# "Build me a rig that checks for open PRs older than 7 days and posts to Slack"

# Option B: Build it yourself
mkdir my-rig && cd my-rig
rube init
rube step add fetch -d "Get data" -s '#!/bin/bash
curl -s https://api.example.com/data > out/data.json'
rube run

Commands

| Command | What it does | |---------|--------------| | rube install | Install all dependencies | | rube doctor | Check dependencies are installed | | rube init | Initialize a rig | | rube list | List all registered rigs | | rube step add <name> -d "desc" -s 'script' | Add a step | | rube dep add <step> --needs <dep> | Wire dependencies | | rube status | See what's ready, waiting, done | | rube run | Execute the DAG | | rube run --dry | Preview execution plan | | rube secrets set <name> | Store a secret | | rube secrets list | List stored secrets | | rube export | Generate Dockerfile + compose |

Runs anywhere

Local: rube run

GitHub Actions: Free cron, no infra. rube export gives you a Docker image with everything baked in.

# .github/workflows/stale-tickets.yml
name: Stale tickets
on:
  schedule:
    - cron: '0 9 * * *'  # Every day at 9am
jobs:
  run:
    runs-on: ubuntu-latest
    steps:
      - uses: actions/checkout@v4
      - run: docker build -t stale-tickets .
      - run: docker run --env-file .env stale-tickets
        env:
          JIRA_TOKEN: ${{ secrets.JIRA_TOKEN }}
          SLACK_WEBHOOK: ${{ secrets.SLACK_WEBHOOK }}

Commit. Push. Your automation runs forever for free.

Docker: rube export generates Dockerfile + compose.

Philosophy

1. Shell scripts are fine. Stop overthinking.
2. Files over memory. Steps communicate via out/.
3. Git is truth. Commit your rigs.
4. Agents can drive it. That's the point.

Requirements

• Node.js 20+

All other dependencies are installed automatically via rube install.

Credits

• Steampipe - APIs as SQL
• DuckDB - transforms
• psst - secrets management
• beads - DAG execution
• April Dunford - positioning inspiration

License

MIT