Intelligence that
acts within boundaries

mashin is a governed intelligence substrate. Build AI systems where every effect is recorded, every decision is auditable, and governance is structural, not bolted on.

Get Started Read the Research

Governed by construction

Every effect passes through a governance interpreter. No bypass path exists. 454 machine-checked theorems prove the guarantees hold.

Auditable by default

Every execution produces a tamper-evident, Ed25519-signed behavioral ledger. What happened, why it was allowed, what it cost. Always.

The right level of abstraction

Machines are structured thoughts about how intelligence should act. High enough for humans, tractable for AI generation, governed by construction.

Write intelligence, not infrastructure

machine emailTriage "Email Triage"

  accepts
    email as text, is required

  responds with
    urgency as text
    category as text
    action as text

  implements
    ask classify, using: "claude-sonnet"
      with role
        "You are an email triage specialist."
      with task
        "Classify this email by urgency (high, medium, low) and category (technical, billing, general). Recommend an action (escalate, respond, archive).\n\nEmail: ${input.email}"
      returns
        urgency as text
        category as text
        action as text

    decide route
      when context.classify.urgency == "high"
        then "escalate"
      otherwise
        then "archive"

Every step is governed. Every LLM call is recorded. Every decision is auditable.

Three products, one substrate

mashin app

The native app for macOS, iOS, watchOS, and CarPlay. Where users interact with machines, Koda, and governance. Five tabs: Koda, Machines, Search, Inbox, Settings.

Learn more →

mashin server

The headless BEAM runtime. Runs machines, governance, and the behavioral ledger. Same binary for desktop (Mac Mini), cloud, Docker, and enterprise.

Learn more →

mashin cli

Command-line tool for compiling, running, testing, and deploying machines. mashin run, mashin test, mashin deploy.

Learn more →

Formally verified governance

Kore is the governance kernel, extracted from 454 machine-checked Rocq theorems with zero admitted lemmas. 12,227 lines of proof. Every governance decision in production runs through proved code.

454 theorems | 0 admitted | 36 modules | ~5 us/call
View proofs on GitHub →

Stay informed

Get updates on releases, research, and the governed intelligence thesis.