Skip to content
JevDirectory.orgTraffic count unavailable. View site traffic

Jev directory

CommunityJev Demos & Experiments

JevSpeak

JevSpeak 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.
See highlightsOpen github.com
Stars
2
Language
TypeScript
License
MIT
Reading time
3 min
Published by
Community
Author
mm-sheng
Last verified
Oct 2026
Tagscommunitycatalog-importlogicrwnlgnextjsexplainable-ai
Works forJev for ClassificationJev for Scoring
Used inJev for Software & DevTools
Languages
TypeScript
Platforms
Next.jsBrowser
Capabilities
Open sourceRequires API key

Highlights

  • Per turn Jev answers ~13 parallel questions — intent, topic, emotion, speech act, stance, claim, tone, length — each a distribution over finite options.
  • Decisions normalize into a typed Semantic IR; a deterministic compiler turns it into English, so same IR in means same sentence out.
  • Confidence is visible in wording: below 0.50 it declines or asks to clarify, 0.50–0.60 says Maybe., and 0.60–0.75 begins I think so.
  • No GPT, Claude or Gemini runs at runtime; the README says there is no hidden rewrite, grammar fix or fallback generative call anywhere.
  • The app has a Jev Brain view for decision distributions and a debug view tracing how they become a sentence; EN and 中文 switch the rendering.

Watch out

Copy reviewed against the truncated repo README on 2026-10-15; the excerpt stops in the confidence table and in the modes and supported-domains sections. The finite claim vocabulary bounds what it can answer, and mock mode uses a deterministic development scorer rather than real Jev.

Questions about this entry

Short answers drawn from the linked project or page.

From 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.

Follow @typesafeai