How zenonum works — step by step

zenonum is the AI-backed startup engine for founders who mean it — structured strategy on a visual business blueprint, then compile ideas into software and run targeted validation experiments.

You iterate, optimize, and pivot when the graph says so. No technical background required — passion and judgment are enough.

The full chain lives in one live graph: what you believe, what is missing, what to research, what to test, and what to build next. One graph truth → one next best move at every step.

Discovery starts on the marketing site or in the dashboard: a structured conversation with the session coach that fills project context, founder profile, and your North Star — often in the first few turns.

The agent asks what a generic chatbot would miss: who you serve, what you believe, what you refuse to assume. Talkers get gently redirected; serious founders get depth.

After OAuth, your thread continues — marketing pre-login chat merges into the canonical session. Unlock materializes founder, project, and North Star nodes plus the Lean canvas scaffold under your star.

The graph follows Lean Startup and Lean Canvas thinking: strategy bands (North Star, problem, segments, value proposition, channels, revenue, costs, key metrics) connect downward to goals, roadmap epics, experiments, tools, and value-added services.

Each node carries a statement, variables (existence, confidence, gravity, success, failure, income, burn), and typed edges that show support, impact, and contradiction — not just pretty boxes.

Lean cycles are explicit: hypothesize on the graph → coach or research → extract facts → run experiments → update variables → recompute next best. The canvas is the operating system, not a poster.

Open the coach on any node or run session-level coaching. The coach challenges weak statements, proposes sharper copy, and respects what you already committed to the graph.

Extraction turns dialogue into structured updates — titles, statements, variables — with preview before save. Children and edges can be suggested; you accept or reject.

Gauges on active cards summarize existence and confidence at a glance; the inspector shows the full variable set, impacts, income, burn, friction, and edge neighborhood.

Agent Control lists roles — graph builder, variable estimator, prerequisite scaffolder, priority recompute, deep research, and more. Each runs Manual, Auto, or Launch per your policy.

Background agents work while you coach: research completes with a notification; graph enrichment lands on nodes you can inspect. Nothing happens in a black box — status surfaces in the chrome.

You steer where it matters: auto-tune strategy loops on Builder+, launch one-off research from a node, or keep everything manual until you trust the graph.

Lenses re-color the graph: Next best highlights priority-ranked moves; Risk, Contradictions, Gravity, and Success show where the methodology wants your eyes.

Semantic search filters loaded session nodes by title and statement — fast recall when the mesh grows.

Required-move methodology can front overlays when policy is active — the graph tells you what to do next, not a static backlog.

One wallet feeds Discovery, coaching, node chat, agents, and research. Daily and monthly token pools reset on UTC boundaries; the wallet modal shows consumed vs cap and next reset times.

Plans tier daily caps (Starter 10k, Builder 50k, Pro 200k) with pay-per-usage top-ups on the roadmap — buy more capacity without changing tier when a research sprint runs hot.

zenonum routes each job to the best agent and model tier automatically — frontier models for hard reasoning, local Llama 4 class models for routine tasks — so spend tracks value, not vanity.

When the graph says build, zenonum compiles product slices from your spec — incremental features, design tokens, API contracts, and rollback-ready experiments. Not one prompt to a million-dollar company; tens of controlled versions you can promote or revert.

PoC4 mesh coding connects strategy nodes to materialized repos: change batches, environment promote, secrets via Doppler, project-edge routing — everything traceable back to the graph node that requested it.

You interfere anywhere: strategy, features, UX, code, marketing copy. Agents run autonomous by default only where you allow it; every action remains visible on the mesh.