Every card is one step of a clinician's chain of thought, floating in 3D. Green edges are the flow; amber edges are decision branches (★ marks the hardcoded default). Purple text explains why each step exists.
Fly around the reasoning graph. Toggle the CoT lens to see every step’s rationale.
Tap a card, change anything — wording, reasoning, decision defaults ★, wiring. Apply updates the scene live; Save keeps your version.
A guided, gamified walkthrough. The camera flies step to step; you tap decision options and quick-answer chips, dictate values with 🎤, and earn XP. Override a default and it offers to become your default.
A workflow from the FlowForge engine: 11 real examples are built in — diagnostic reasoning chains, full procedures (root canal, crown, extraction, implant…), a patient journey and an office process. Pick one from the ☰ library (top left).
Flow runs along the depth axis; branches fan out sideways. Card borders mean:
The CoT lens button shows the chain-of-thought note on every card at once. Tap any card to pin its full reasoning panel.
Tap a card and the inspector opens. Everything is editable: the label, the instruction, the reasoning note, who performs it, accepted inputs (photo, voice, chips…), materials, and — on decisions — the options themselves, with the radio picking the hardcoded default ★ and dropdowns rewiring where each branch goes.
Click a card in Edit mode and it opens — name, instruction, and reasoning editable right there. 🔗 Connect… starts link mode: tap the step it should lead to (on a decision step this adds a new branch). ↶ / ↷ in the top bar (or Cmd/Ctrl+Z) undo and redo any change.
+ after inserts a step, Delete removes one (wiring heals itself). Apply updates the 3D scene instantly. Save my version keeps your edits on this device — the library copy stays untouched; a ● in the library marks edited workflows and Reset to library restores. Export downloads the JSON to import into FlowForge Studio on the platform.
The camera flies to each step; the card at the bottom tells you what to do and — via ✦ Why this step? — why. Decisions are tapped: ☆ marks the hardcoded default, but you can pick anything. Quick steps build a 🔥 combo; finishing earns XP.
Inputs on a step: decision steps show tappable options (☆ = hardcoded default). Steps that accept chips show quick-answer buttons (Confirmed / Flag / N-A); steps that accept voice show a 🎤 that dictates into the value field (Safari asks for mic permission on first use); typed values are logged with Log.
Sandbox vs Live: the amber SANDBOX mode is your playground — edits stay local. The green LIVE mode shows the published version everyone gets, read-only. The colors follow you: amber top-bar stripe and glow = sandbox, green = live. ⬆ Send to HITL proposes your sandbox version for approval in the HITL Center — approved flows become the live version.
Versions: every Save banks a restorable version. Open Edit with no card selected to see the version list — Restore rolls the workflow back (and banks the current state first, so restores are themselves reversible).
The learning loop: if you override a default, the finish screen offers to make your choice the new default (★). Accepting genuinely edits your saved copy — the same self-improvement loop the platform runs, where 3 repeats trigger the offer and sharing sends it to council review.
Runs entirely on your device — no network, nothing leaves this page. Edits live in this browser’s storage. Safety-gate steps are rule-based and can’t be skipped; AI-assist markers are suggestion points only (human-in-the-loop by design).
Flip anything — it applies instantly and remembers. This is your playground; the live library never changes.
Sandbox security is local to this browser — a convenience lock for shared operatory screens, not account security. Real role-based access lives in the platform's HITL Center and admin gates.
Photograph a tooth to open diagnostics, a supply bottle or instrument to find the procedures that use it, or a printed daily schedule to jump into office operations. Recognition runs through the platform's vision module and feeds the same self-training loop as the rest of the platform.