Most AI products embed operational intelligence inside prompts, runtimes, or orchestration code — all ephemeral. Inistate externalizes it into a schema: workflows, states, forms, transitions, and audit trails persisted as the operational source of truth. This page walks the schema, primitive by primitive. By the end, you'll have built one layer of the harness — the substrate humans, AI, and hybrid actors execute against.

The harness has five primitives: Activity, State, Information, Form, and the transition that binds them. Each requires the ones beneath it. Together they form the dependency graph — the architectural moat that no incumbent can copy without a multi-year rebuild. Here's the build order.

1. Start with the Activity block

Every process begins with an activity. It's the action someone performs, whether it's contacting a lead or a manager approving a leave. On our canvas, the journey starts with create, a basic activity.

2. Move to the State block

A state acts as a status indicator, showing the current phase of a process and guiding the next steps. It's akin to real-life scenarios, like flight statuses. On our canvas, we initiate with the Draft state, symbolizing a leave request's inception.

3. Fill in the Information block

This block captures the essence of any process. It's where you list down data points like texts, dates, images, and more. For our leave application, details like:

• Employee for the employee name who makes the requests
• Leave Type for type of leave applied for example Annual Leave, Medical Leave or Unpaid Leave.
• Start Date for leave start date
• End Date for leave end date
• Reject Reason for reason for rejection

4. Integrate Information to Activity as Form

Forms are the bridge between activities and information. They allow users to input or update values when performing an activity. On our canvas, for the create activity, we'd exclude Reject Reason as it's not relevant at this stage.

5. The inversion: Actor Parity.

The transition you just drew — Draft → Pending Approval — is actor-agnostic. A human submits it through a browser form. An AI agent submits it through MCP using the same schema. A hybrid sequence does both. Same form. Same transition. Same audit record shape. No separate API surface for AI. No separate audit pipeline. One primitive, one audit row, one governance regime — for every actor.

That's Actor Parity. It's patent-protected (US Provisional #64/045,012, filed April 21, 2026) and it's why models become interchangeable execution intelligence while the workflow stays the operational source of truth. Each action on the canvas can lead to a new state — whoever performs it.

6. Finalize Your Canvas Design

Voila! You've just crafted a basic leave application process using the Workflow Canvas. But remember, the canvas is flexible. As your needs evolve, you can introduce more complexities. Once satisfied, let Inistate work its magic, turning your canvas into functional tools.

What you can't see on the canvas: the audit row.

Every transition produces one history event with identical fields — by (who), on (when), changes (what), and an ai object containing reasoning, sources, model, model version, and confidence. The shape doesn't vary by actor. Compliance reporting is a query, not a reconstruction project. EU AI Act Article 14 (meaningful human oversight) and FINRA/HIPAA audit requirements map directly to fields the harness ships natively — not features added by professional services.

Dive Deeper with the Canvas

Roles, permissions, confidence gating, intention events, and AI-only / hybrid / any-actor activity types compose on top of what you just designed. Confidence below threshold? The harness flags it as an intention event and pauses for human consent — that's how the Three Laws get enforced inside the primitive. (See Actor Parity for the full mechanism stack.)

You've just built one layer of the dependency graph.

What you sketched in 6 steps, Inistate built from one sentence in Demo 6 — same Leave Application, same four-state machine, same audit trail. Models churn every 4–6 months. The schema you just walked compounds for 5–10 years. → Watch Demo 6 · Read the architecture · Build against the MCP server