An input that becomes what you mean. One text box that morphs into the right UI as you type — an event card, a checklist, a timer, a color picker, a bill splitter, a poll, a converter and more.

_{Try it live · Watch the full 60-second demo (1080p60)}

dinner with priya friday 8pm on zoom   →  Event card · Friday · 8 PM · Priya · Video call
buy milk, eggs, bread and coffee       →  Shopping checklist
split 2400 between 3                   →  ₹800 each
minecraft diamond                      →  #4AEDD9

Intent is classified by TypeSafe AI's Jev model: one call answers 14 typed questions in parallel (which card, plus signals like "is it a video call?", "is it urgent?"). Everything else — dates, amounts, units, math — is deterministic code. Jev decides, code computes.

It works fully offline by default with a built-in keyword classifier, so you can run it without an account.

Requires Bun 1.2+.

bun install
bun dev

Open http://localhost:3000 and start typing. Press / to see every card type.

cp .env.example .env.local
# then set TYPESAFE_API_KEY=... (get one at https://console.typesafe.ai/keys)

Restart bun dev. The latency readout in the bottom-right corner switches from jev-offline to jev-1.13.0. The key is only ever read on the server (/api/intent); it never reaches the browser. If the API is unreachable or rate-limited, Shapeshift quietly falls back to offline mode.

Saved cards live in your browser (localStorage) until you delete them. Click one to edit it.

URL flags: ?debug=1 shows every probability; ?demo=1&loop=1 plays a scripted demo.

Raw model output flickers as you type, so a small state machine turns confidence into calm UI states. A card only changes when a challenger wins twice in a row (or is very sure), and signal badges use an on/off hysteresis band.

The diagrams are Excalidraw files — open any docs/diagrams/*.excalidraw at excalidraw.com to edit them.

1. Add the key to INTENT_KEYS in src/lib/jev/types.ts.
2. Add a non-overlapping criterion to intent in src/lib/jev/questions.ts.
3. Write a parser in src/lib/parse/ and register it in src/lib/parse/index.ts.
4. Write a card component in src/components/intents/ and add a registry entry.
5. Teach the offline classifier in src/lib/jev/mock.ts, and add tests.

TypeScript will point at anything you missed.

bun run check    # typecheck + lint + tests
bun test         # parser, decision, signal and classifier tests
bun run build

Stack: Next.js (App Router) · React 19 · TypeScript (strict) · Tailwind CSS v4 · shadcn/ui · Motion · chrono-node · zod.

MIT