Skip to content
mashinmashinv0.150.0
Developer Preview — APIs and language features may change before 1.0

Quickstart

1. Install mashin

Section titled “1. Install mashin”
Terminal window
# macOS
brew install mashin-live/tap/mashin


# Or download directly
curl -fsSL https://get.mashin.live | sh

2. Create a machine

Section titled “2. Create a machine”
Terminal window
mashin new classifier --template classifier

This creates classifier.mashin:

machine classifier


  accepts
    text as text, is required


  responds with
    category as text
    confidence as number
    confident as boolean


  behaves
    ask classify, using: "anthropic:claude-sonnet-4-6"
      with task "Classify this text into a category. Return the category name and a confidence score between 0 and 1.\n\nText: ${input.text}"
      returns
        category as text
        confidence as number


    compute assess
      {category: classify.category, confidence: classify.confidence, confident: classify.confidence >= 0.7}


  verifies
    test "a confident classification is flagged confident"
      assuming classify {category: "general", confidence: 0.85}
      given {text: "What are your opening hours?"}
      expect {category: "general", confident: true}


    test "a low-confidence classification is not"
      assuming classify {category: "billing", confidence: 0.4}
      given {text: "asdf qwerty"}
      expect {confident: false}

The machine asks an LLM to classify, then a compute step decides whether the classification is confident enough to act on.

3. Run it

Section titled “3. Run it”
Terminal window
mashin run classifier.mashin --input '{"text": "The quarterly earnings exceeded expectations"}'

Output:

{
  "category": "finance",
  "confidence": 0.92,
  "confident": true
}

4. Test it (without an API key)

Section titled “4. Test it (without an API key)”

The assuming block mocks the LLM step: it supplies the answer the model would give. The expect asserts what the machine derives from that answer. The first test feeds in confidence 0.85 and asserts the machine flags it confident: true; the second feeds in 0.4 and asserts the flag comes back false. The logic under test is the machine’s own, not the mock.

Terminal window
mashin test classifier.mashin

Output:

  Testing Machine: classifier.mashin


  classifier
    PASS  a confident classification is flagged confident (59ms)
    PASS  a low-confidence classification is not (1ms)


  Summary: 2/2 passed (60ms)

No API key needed. Tests are fast, free, and deterministic.

5. See the governance trace

Section titled “5. See the governance trace”
Terminal window
mashin run classifier.mashin --input '{"text": "hello"}' --trace

The trace shows every governance decision: what permissions were checked, what the LLM was asked, how many tokens it used, and the hash chain proving the execution was not tampered with.

Ship it

Section titled “Ship it”

Your machine works locally. Now make it live:

Terminal window
mashin deploy classifier.mashin

Your machine is now running on your Kortex, accessible from any device. The governance, audit trail, and behavioral ledger travel with it.

From your phone: “Hey Siri, run email triage.” From the web: visit your machine’s page. From another machine: ask classify, from: "@you/email-triage".

One command. Zero infrastructure. That is the mashin deployment story.

What’s next?

Section titled “What’s next?”