User repository on GitHub
Semantic command palette (Ctrl+K) ranked by intent with TypeSafe Jev — one Choice call, the probability distribution IS the ranking
- Forks
- 0
- Last pushed
- Updated Sep 2026
Repo stats from the GitHub API, cached Oct 2026.
Jev directory
CommunityJev Demos & Experiments
Jev Palette
A semantic command palette demo where Jev ranks 77 commands by intent in Portuguese or English: one Choice question per keystroke, with the full probability distribution over the command catalog as the ranking. Runs side by side with a classic fuzzy matcher that fails honestly on the same queries.
- Stars
- 0
- Language
- JavaScript
- Reading time
- 3 min
- Published by
- Community
- Author
- amycardoso
- Last verified
- Oct 2026
Used inJev for Software & DevTools
- Languages
- JavaScriptTypeScript
- Platforms
- Browser
- Capabilities
- Open sourceRequires API key
Highlights
- Choice question per keystroke, all 77 commands as options; Jev returns the full distribution in one ~100ms call — no embeddings, no output tokens billed.
- About 900 input tokens per call at $0.042 per million input tokens puts each keystroke at roughly $0.00004.
- Eval with 50 labeled queries against jev-1.13.0 (2026-09-19): Jev top-1 98% and top-3 100% overall versus fuzzy 20% and 20%; mean Jev latency 413ms.
- On the name-like slice where fuzzy matching is strong, both hit 100% — the README suggests production palettes should run fuzzy first and fall through to Jev.
- npm run dev:mock serves a local keyword-based stand-in offline, with a modo mock badge until a real key is configured.
Quickstart
bash
npm install
cp .env.example .env # put your TYPESAFE_API_KEY in it
npm run dev # proxy on :8787 + Vite on :5173Watch out
Grounded in the README and GitHub metadata reviewed on 2026-10-15; the demo ranks a fictional mail app's 77 commands, the 50-query eval costs about $0.004 to run live, the single top-1 miss is documented, and the GitHub metadata lists no license.
Questions about this entry
Short answers drawn from the linked project or page.
A semantic command palette (Cmd+K) demo: you type your intent in Portuguese or English and Jev ranks 77 commands of a fictional mail app by meaning, side by side with a deliberately classic fuzzy matcher that fails honestly on the same queries.
It asks one Choice question per keystroke with the whole command catalog as options; Jev returns the full probability distribution over all options in a single ~100ms call, and that distribution is the ranking. No embeddings or vector index are involved.
Roughly $0.00004 per keystroke: about 900 input tokens per call at $0.042 per million input tokens, with no output tokens billed. The README's 50-query eval costs about $0.004 to run against the live API.
Against jev-1.13.0 on 2026-09-19, Jev scored 98% top-1 and 100% top-3 over 50 labeled queries versus 20% for the fuzzy baseline; mean Jev latency was 413ms. The single miss ranked 'throw this away' as Discard draft over Delete message.
No — the README's own caveat: on name-like queries where users already know a command's name, fuzzy is just as accurate and effectively instant, so a production palette should run fuzzy first and fall through to Jev when it comes back empty or weak.
Yes: npm run dev:mock, or npm run dev with no key, serves a local keyword-based stand-in so the UI works offline, with a modo mock badge in the header until a real key is configured. The proxy keeps real keys server-side.
Verified against jev-1.13.0: a choice question takes instructions and a criteria map of option id to description, and the answer returns choice, probabilities summing to 1 across every option, and confidence. rankFromAnswer also tolerates a distribution field name.
More Jev resources like this
An AI companion that replies only with facial expressions: Jev judges each message and picks one of 16 moods while blobatar morphs its face, with no text replies and no visible chat history.
Jev Demos & Experiments#community#catalog-import#logicrw
Community
appchainai/jevatarTypeScript
appchainai/jevatarTypeScriptA ChatGPT-style chat UI for Jev where answers are decisions, not prose. Pick Yes/No or styles like Yes/No/Maybe, Pirate or Tabloid, and expand any reply to see Jev's calibrated probability for every option. Runs on Deno.
Jev Demos & Experiments#community#catalog-import#logicrw
Community
kt3k/jevchatJavaScript
kt3k/jevchatJavaScriptA small React web app where Jev, TypeSafe's System One API, estimates a name's age band and gender from naming trends across Korea, the US, Japan and China. A Vite server plugin keeps the API key server-side and the UI draws the returned band probabilities as bars.
Jev Demos & Experiments#community#catalog-import#madewithjev
Community
ximhear/jev-kr-name-ageTypeScript
ximhear/jev-kr-name-ageTypeScriptA playable Tetris where Jev never sees the board: code enumerates every legal placement as a sentence and Jev returns a probability for each in about 380 ms. The field glows with the whole distribution before the piece drops.
Jev Demos & Experiments#community#catalog-import#cobanov
Community
thelau/jev-tetrisJavaScript
thelau/jev-tetrisJavaScriptJevSpeak explores whether a model that cannot generate text can still talk: about 13 parallel Jev decisions per turn become a typed Semantic IR that a deterministic compiler renders in English or Chinese. Live at jevspeak.org.
Jev Demos & Experiments#community#catalog-import#logicrw
Community
mm-sheng/jevspeakTypeScript
mm-sheng/jevspeakTypeScriptA one-page playground and test bench for TypeSafe Jev: pick a backend, load an example and run typed questions against Jev through OpenRouter or TypeSafe, an LLM baseline, or a keyless mock — all speaking the same /v1/systemone contract, with an inspector for the answer distributions.
Jev Demos & Experiments#community#catalog-import#logicrw
Community
creativoma/here-we-go-jevTypeScript
creativoma/here-we-go-jevTypeScriptFrom the Jev guides
Original write-ups on what this entry is for.
From the Jev AI model community
Posts from builders shipping with Jev right now.