NOVA is a free journey and workflow design tool for mapping agentic AI journeys across all 22 dimensions of the HAI framework. Dropdown-enforced taxonomy, multi-journey projects, ZIP export with AI skill files — running entirely in your browser.
Agentic AI lives at the intersection of disciplines. NOVA gives every team a shared language and a shared canvas so handoffs stop dropping context.
Specify agent behaviour at every step before engineering writes a line of code.
Map user emotion, action, and pain points alongside agent reasoning — in one canvas.
Document MCP servers, A2A handoffs, memory tiers, and protocol layers in their proper context.
Design lifecycle journeys where trust evolves and autonomy grows across stages.
Tag every step with EU AI Act risk classification and required human oversight.
A canonical 22-dimension taxonomy. Two journey modes plus governed workflows. ZIP export with AI skill files — ready to build with Claude, ChatGPT, Cursor, or any AI tool.
Every step covers Human, Agent, and System layers — 11 fields are dropdowns enforcing the canonical HAI taxonomy.
One click downloads a ZIP bundle: your journey spec as a structured `.md` file plus two AI skill files — drop them into Claude, ChatGPT, or Cursor and build immediately.
One project can hold many journeys with sub-tabs — perfect for products spanning multiple lifecycle stages.
Switch between visual journey graph and tabular matrix view depending on the work — same data, two lenses.
Tag each step with EU AI Act risk tier and specify sovereignty boundaries — data residency, compute location, model provenance, and cross-border constraints — built into the matrix.
Specify MCP, A2A, AG-UI per step — designed for the modern agentic protocol stack.
Your projects live in your browser. No accounts, no servers, no analytics — privacy by architecture.
Add a Governed Loop to any workflow: a bounded iteration with an enforced ceiling, an exit condition, and a defined exhaustion path — Degrade, Escalate, or Proceed-Partial. No unbounded loops; the design gates how many times an agent can retry before control hands over.
Every agentic journey involves three perspectives running in parallel — what the human sees, how the agent reasons, and what the system enables. The HAI framework structures all three at every step.
What the person is trying to accomplish, how they feel, and where they get stuck.
How the AI reasons, what mode it operates in, and how it handles failure.
What infrastructure enables, constrains, and governs the agent's behaviour.
In 2026 the discipline moved from prompt engineering to context engineering to harness engineering — the runtime that wraps a model: its tools, its memory, its stop conditions, its guardrails. A harness is where reliability comes from. HAI's answer is that the harness shouldn't be improvised in config files after the fact — it should be specified, at design time, in the System Layer.
Same four elements a runtime harness needs — a tool interface, context management, control mechanisms, and an agent loop — but captured upstream, as an editable specification that governs what the harness is allowed to do. The builder assembles the harness; NOVA writes its contract.
The same three layers, at two scales of design time — then a governed execution mode that runs the design. Journeys specify; workflows execute, and the specification governs what runs.
For specific tasks inside a product. Onboarding, checkout, support resolution. Where agent behaviour at each click matters and failure has immediate, localised consequences.
For full customer lifecycles. Awareness through retention. Where trust evolves, channels shift, memory accumulates, and failure affects long-term relationships.
For turning a designed journey into a governed, step-based process that runs. A workflow inherits the same three layers, but adds enforced autonomy ceilings, retry limits, and failure-handover triggers — so the specification governs what the agents can actually do at runtime.
Each template ships with all 22 dimensions filled in across every step — user journeys, customer lifecycles, and governed workflows. Edit, extend, or strip back — they're a starting point, not a constraint.
New-user activation flow with sensible-default recommendations.
E-commerce checkout with fraud-detection guardrails baked in.
Multi-agent pipeline: triage → specialist → human escalation.
Awareness through retention with trust progression across stages.
Trial → Renewal lifecycle with autonomy increasing over time.
Regulated banking journey with EU AI Act high-risk classifications mapped throughout.
A governed, runnable workflow with enforced autonomy ceilings and failure-handover triggers per step.
Accounts-payable workflow with retry limits and a human-approval handover before any payment is released.
An on-call workflow that classifies incoming alerts, escalates by severity, and hands off to a human when confidence is low.
NOVA is where you specify the human–agent–system journey with rigour. Every export is a ZIP bundle — your complete journey spec, the exact agent behaviours, failure protocols, and system constraints your AI tool needs to generate production-ready output, not a guess.
Open a blank canvas in User or Customer Journey mode, or modify one of the pre-filled templates as a starting point.
Fill in the 22 dimensions across Human, Agent, and System layers. Dropdowns enforce HAI taxonomy; free text everywhere else.
One click downloads a ZIP — your journey spec (.md) plus two AI instruction files. Drop them into Claude, ChatGPT, or Cursor and the AI knows exactly how to build from it.
Upload the spec and one skill file to any AI tool. Say what you want to build. The AI has everything it needs — no briefing, no re-explaining, no guessing.
Most vibe coding starts with a chat message and a guess. The AI invents the requirements, fills in the blanks, and produces something that resembles a prototype but reflects nothing you actually designed.
A NOVA export changes that. The bundle carries every decision you made in the canvas — who the user is, what the agent does at each step, what autonomy level it operates at, what happens when it fails, what data it needs, and what compliance constraints apply. The AI doesn't need to invent any of it.
Export once, get the format you need: a Journey Spec PDF, a Runtime Config YAML for LangGraph, CrewAI, or the Agents SDK, RTCROS Prompts ready to paste into any model, a self-contained Portable Workflow JSON, and AI Skills that teach the framework on the spot. Upload it, say one sentence — "Build a working prototype of this journey." — and that's the whole brief.
The full NOVA canvas — clean, fast, and honest about what's specified versus what's still open.
Most deployments discover sovereignty constraints at audit. HAI makes them explicit before a single line of code is written — per step, per stage, per layer.
The HAI Sovereign AI Extension adds a Sovereignty Boundary dimension to the System Layer. At every step or stage it records five things: data residency jurisdiction, compute residency requirement, model provenance constraint, cross-border data flow permission, and accountable jurisdiction.
Sovereignty is rarely binary. A single deployment may permit a foreign foundation model while mandating in-jurisdiction inference and in-country storage. HAI separates these decisions cleanly across layers — so each can be decided independently, by the right team, at design time.
NOVA operationalises the Human-Centred Agentic Intelligence framework — the first design methodology purpose-built for agentic AI journeys. The framework is published, peer-referenced, and openly licensed. The May 2026 edition adds the Sovereign AI Extension.
No installation. No account needed. Start designing your first HAI journey in seconds.
Start for free
