The reasoning layer · MCP-native

Watch an idea
become an app.

Prompt-to-app tools start building on your first sentence — and the plan never exists. Appestry is the deliberate middle: you talk, your agent weaves the reasoning, drafts the real screens, and a dev-ready spec comes out the other side — every feature traceable to the decision that demanded it.

Bring an idea See it come to life ↓
rambledecisionsfeaturesUI screensjourneysroadmapdev handoff
Act I — Ramble

It starts the way every idea does. Out loud.

No forms, no “user stories,” no blank canvas. You just say what you want. Your agent listens and quietly sorts it — commitments become decisions, worries become questions, evidence becomes claims, every “maybe” becomes a proposal — then shows you exactly what it heard.

Capture is free. The human only rules.

You
I want a tiny app, just for me — to keep track of my houseplants and remind me when to water each one. Some need water weekly, some monthly. I always forget and they die 😔
Appestry
Extracted — your callwindowsill · 5 nodes
A personal houseplant watering trackerdecision · fiat — challenge it any time
I forget which plant needs water when — and they dieclaim · insight — the whole reason
Just me. No accounts, no sharingclaim · constraint
How should a reminder actually reach me?question · zero positions — the frontier
A photo of each plant so I recognize themproposal · uncommitted

You didn’t fill out a single form.

You had a feeling about some plants. It’s already a structured product with the reasoning attached.

Act II — The frontier

The open questions, and the answers still looking for them.

The frontier is your agenda — open questions with the uncharted ones first, because a question nobody has answered yet is the most valuable thing in the whole system. Beside them, the mirror rail: proposals, ideas that haven’t earned a commitment. Sweep them in seconds.

The frontier is sacred. Zero positions isn’t a gap — it’s the point.

You
Hmm, I’m honestly not sure how the reminder should reach me. And should it show an exact date or just “due soon”?
Appestry
Frontier2 open · ◇ uncharted first · ▷ 2 proposals
How should a reminder reach me?uncharted — the most valuable state in the system
uncharted
Exact next-water date, or just “due soon”?
2 positions
proposals — answers seeking questions
A photo of each plant
AdoptParkDismiss
Group plants by room
find its question ✦
Act III — Rule & compile

You decide. The features write themselves.

Say “yes, do that” and a decision is recorded — one ruling compiles into many concrete features, each landing in a bounded review batch. Triage at keyboard speed: approve, reject with a reason, defer. Your attention goes to exactly one thing — verdicts.

Nothing is deleted. Reversal is supersession, never editing.

You
Yes — weekly, biweekly, and monthly intervals per plant. And yeah, give me a “what needs water today” screen.
Appestry
Batch 1core watering · 3/5 decided
Add a plant — name, photo, room
REQUIRED
Per-plant watering interval — weekly / biweekly / monthly
REQUIRED
“Water today” home screento review
✓ Approve✕ RejectLater
j/k move · enter approve · x reject · l later · 1–5 priority
Your app, taking shape
Idea
1 decisionthe tracker exists
Ruled
5 features3 required
Drawn
3 screensrendered live
Proven
1 journeythe morning check
Act IV — It comes to life

Now watch it actually appear.

This is the part other spec tools can’t do. Your agent drafts real, versioned screens and components — not wireframes, working React — and renders them right here in the chat. The plant app you described a minute ago is now a thing you can look at, tweak, and version.

A living catalog. Every screen traces back to the feature it serves.

You
Okay, show me the “today” screen — the plants I need to water, with a tap-to-water button.
Appestry
Todayio.windowsill.today · v2 · live preview
Today· 2 plants need water
🌿
MonsteraLiving room · every 7 days
🪴
Fiddle-leaf figBedroom · every 10 days
🌵
Snake plantHallway · every 30 days
in 4 days
catalog · 3 screens · 2 components
Plant cardcomponent · v2
used by 3
Add-plant formscreen · v1
Act V — Walk it

Take the whole thing for a stroll before it’s built.

Accepted features are a list; an app is a path. Your agent stitches the real screens into a journey and walks it as you would — open Today, tap a plant, mark it watered. Where a walk breaks, a question is minted. Coverage gaps surface in minutes, not in week three.

You
Walk me through my morning: I open the app, see what’s thirsty, and water the Monstera.
Appestry
Journey — the morning checkme · 3 steps · live screens
1
Open Today — see who’s thirstyscreen: today
✓ accepted
2
Tap Monstera → its care detailscreen: plant-detail
✓ accepted
3
Mark watered → next date recomputesscreen: today · asserts “mark watered” feature
○ to review

A spec is only worth keeping if it remembers why.

— the one thing prompt-to-app tools throw away
Act VI — Sequence & ship

The roadmap writes itself. You just say what needs what.

Order lives in dependencies — real product truths like “the Today screen needs watering intervals first.” Lanes, colors, and milestones derive from the graph on every read. The release lens answers the only question that matters: is every required feature covered by a journey? For a weekend project or a funded startup, same machine.

Roadmaps are lenses, never artifacts. Sequence, never sprints.

You
What should I build this weekend, and in what order?
Appestry
Roadmap6 features · v1 this weekend
SHIPPED TODAY THIS WEEKEND v1 P1 Add a plant done P2 Plant list done W1 Watering intervals 2 of 3 T1 Today screen building R1 Reminders next ● PH Plant photos next
The full loom

Everything in the box.

The reasoning graph

Decisions, questions, positions, claims — captured as a side effect of talking. Fiat is normal; deliberation attaches only when challenged.

Proposals inbox

“Maybe” and “what if” get a first-class home. Adopt, park with a wake condition, or dismiss with a reason that teaches your agent.

One-gesture triage

Features arrive in bounded batches. Approve, reject, defer — keyboard-speed, with visible progress. You only ever spend attention on verdicts.

Live UI catalog

Your agent drafts real, versioned React screens and components — rendered live, right in the chat. Watch the actual app take shape, not a wireframe.

Journeys

A persona walks the real screens end to end. Every step asserts the features it needs — a broken walk mints a question before a line of code exists.

Priority bands + graveyard

None → required, ordered within bands. Rejections keep their reasons forever and stay revivable — memory, not a tomb.

A roadmap that derives itself

Dependencies are the truth; lanes, colors, and milestones are computed on read. No authored roadmap document to rot.

The release lens

Launch-ready = every required feature covered by a journey. One call tells you exactly what is and isn’t.

Blast radius

Supersede a decision or invalidate a claim and the downstream set is named — features to revisit, questions reopened. Never from memory.

Authorship, always

Every node records who: a human gesture or an agent acting for one. “Whose idea was this?” is a query, not an argument.

Projects, roles, invites

Each product is its own tenant-walled graph. Owner, collaborator, viewer — invite your co-founder, your client, or keep it just for you.

Native to claude.ai

An MCP connector: the windows render inside your chat. Talk in the same place the spec lives — plus CLI and API access.

Where it sits

The WHY layer of your stack.

Appestry doesn’t replace your dev tools — it feeds them. Accepted features and journeys are the handoff artifact; your issue tracker runs the build; Appestry keeps the reasoning current as reality reports back.

Appestry — why & what

Decisions, questions, evidence, proposals → features, screens, journeys, roadmap.

Your dev board — who & when

Linear, Jira, a notebook. Takes the ordered spec; owns tasks and sprints.

Your agent — the build

Claude Code and friends build from a spec that finally deserves the name.

appestry.ai

Bring an idea. Leave with an app you can build.

Connect it to claude.ai in two minutes — the windows render right inside your chat.

Start free Sign in