User repository on GitHub
Chess moves, evaluations, persona opponents, and game classification with TypeSafe AI System One
- Forks
- 0
- Last pushed
- Updated Oct 2026
Repo stats from the GitHub API, cached Oct 2026.Homepage
CommunityJev Practices & Patterns
Jev Chess — hemanth
TypeScript chess library where Jev resolves natural-language move intent against verified legal moves, scores sharpness, strategic theme and king risk in one call, and drives persona opponents.
- Stars
- 2
- Language
- TypeScript
- Published by
- Community
- Author
- hemanth
- Last verified
- Oct 2026
About this project
Chess moves, evaluations, persona opponents, and game classification using TypeSafe AI System One models. resolveIntent() maps natural language to verified legal moves via Choice. evaluateMove() assesses sharpness, strategic themes, and king risk in parallel. selectMove() weighs candidates against persona archetypes. That's the whole loop.
Quoted from the project's README. Read the full README.
- Languages
- TypeScript
Highlights
- resolveIntent() matches "Develop knight to attack center" against legal moves with Choice, returning confidence plus alternatives, not fresh coordinates.
- One systemOne() call returns tacticalSharpness as a Score 0.0-3.0, a strategicTheme such as pawn_break or prophylaxis, and kingAttackRisk as a Noul 0.0-1.0.
- Commentary is synthesized by deterministic code instead of model prose, and personas are client-side weight vectors over the same atomic dimensions.
- ClassicMatchStudio.classifyMatch() returns an archetype, an aesthetic-brilliance score with a level, and a turning point as a move number plus SAN.
Watch out
Strength claims are the gap. Tal, Petrosian, Capablanca and Coffeehouse are described as client-side weight vectors over the System One dimensions — Tal weighted toward tactical sharpness, king attack and psychological pressure, Petrosian toward prophylaxis and king safety — but no games, ratings or results against a real engine appear anywhere, and the "Tal plays ..." rationale is not traced back to the weight that produced it. The same applies to ClassicMatchStudio: "ROMANTIC SWASHBUCKLER", a 2.9 aesthetic score and a turn-20 turning point are shown as sample output, with no ground truth for what counts as a verified sacrifice or a real turning point. The README says a single systemOne() call covers four orthogonal dimensions while the documented example exposes three, with commentaryBadge built deterministically. Cost per call is never mentioned, and every intent resolution, evaluation, persona choice and classification looks like a live API request. Copy reviewed against the repository README on 2026-10-04, not from running it.
Questions about this entry
Short answers drawn from the linked project or page.
Intent resolution runs through Choice against the legal moves the engine already holds rather than generating coordinates, so a match can only be a legal move. When confidence falls under your threshold, resolveIntent returns the alternativeCandidates instead of picking one for you — the quick start gates on confidence > 0.6 before calling makeMove, and the whole loop is engine.makeMove(matchedMove.san), so the engine remains the source of legality.
The README gives you no way to check. Personas are client-side weight vectors over atomic System One dimensions, with Tal weighted toward tactical sharpness, king attack and psychological pressure. There are no games, no ratings and no results against a real engine, so treat them as themed behaviour settings rather than as strength estimates, and expect the rationale string to read as an explanation of the weights rather than of good chess.
One systemOne() call per evaluation, returning tacticalSharpness as a Score from 0.0 to 3.0, strategicTheme as a Choice among pawn_break, tactical_strike, prophylaxis and similar values, kingAttackRisk as a Noul probability from 0.0 to 1.0, and a commentaryBadge such as "Sharp Tactical Clash". The badge is synthesized by deterministic code from those numbers, so no model is asked to write the prose. The README notes no price.
classifyMatch(CLASSIC_MATCHES[0]) returns an archetype string, an aestheticBrilliance object with a score and level, and a turningPoint with a move number and SAN. The README also says it detects turning points, verifies sacrifices and produces structural tension breakdowns — but it defines none of those terms and reports no accuracy, so treat the sample output as the shape of the answer, not as a quality bar you can rely on.
It is a TypeScript library rather than an app: engine.makeMove takes the matched SAN move, resolveIntent returns a match with confidence and alternatives, and the evaluation comes back from one systemOne() call. You supply the target role or intent in plain language and gate on confidence yourself, with the quick start using a 0.6 threshold before calling makeMove.
The README says a single systemOne() call covers four orthogonal dimensions, but the documented example exposes three: tacticalSharpness as a Score from 0.0 to 3.0, strategicTheme as a Choice such as pawn_break, tactical_strike, or prophylaxis, and kingAttackRisk as a Noul from 0.0 to 1.0. The commentaryBadge that goes with them, such as Sharp Tactical Clash, is built deterministically in code from those numbers.
Yes, and no price is documented. Intent resolution, each evaluation, the persona choice, and the match classification all look like live System One API requests, so a full game multiplies them: every move is a resolution, an evaluation, and a persona selection. The README never converts any of that into a cost figure.
Other listings of this project
Forks and copies of this project in the directory. This entry is the canonical one.
A live 3D chess match where Jev plays both sides with different personas and exposes the probability split behind every move.
Jev CookbooksPlayground#community#playground#games
Community
A single shared chess game where visitors play against Jev, which weighs every legal move at once as one Choice question and shades the board with its probabilities.
Jev CookbooksPlayground#community#playground#demos
Community
More Jev resources like this
7GitHub stars
Word-level language model experiment that uses Jev as its output layer, with explicit tokenizer, sampler, repetition mask, stop rules, and chunk verification.
Jev Practices & Patterns#community#jevbest#catalog-import
Community
y0usaf/jev-lmTypeScript
y0usaf/jev-lmTypeScript5GitHub stars
Open arena pitting Jev against a chosen opponent on your question, hiding identities until you vote, then revealing latency, cost, and confidence data.
Jev Practices & Patterns#community#awesomejev#catalog-import
Community
chenmingtang830/jevarenaTypeScript
chenmingtang830/jevarenaTypeScript2GitHub stars
A TypeScript starter for the layer around the model: decision contracts, confidence thresholds that route to auto, fallback or human review, deterministic mock providers, and an offline evaluation harness.
Jev Practices & Patterns#community#awesomejev#catalog-import
Community
hamakyo/jev-starterTypeScript
hamakyo/jev-starterTypeScript2GitHub stars
A Turkish chat toy that answers every message with one of five fixed phrases, using Jev as an evaluation model so the five-word limit is enforced by the type system rather than by prompting.
Jev Practices & Patterns#community#awesomejev#catalog-import
2GitHub stars
An MCP server with two tools for agent memory: record writes facts into a per-project ledger as they happen, and recall scores each one against the task you are starting and hands back only what matters.
Jev Practices & Patterns#community#awesomejev#catalog-import
Community
Dharundp6/jev-carryforwardTypeScript
Dharundp6/jev-carryforwardTypeScript1GitHub stars
An independent benchmark asking whether Jev can replace an LLM reranker, run over natural-language search across all 6,245 YC companies, plus the live search app it measures: Jev routes the query, BM25 and embeddings retrieve, Jev reranks.
Jev Practices & Patterns#community#awesomejev#catalog-import
Community
PPRAMANIK62/yc-jev-benchTypeScript
PPRAMANIK62/yc-jev-benchTypeScriptFrom the Jev guides
Original write-ups on what this entry is for.
Most Jev accuracy problems are state problems. What belongs in a state, what is a pre-made judgment to leave out, and why naming the slice you ask about matters more than the wording.
Jev Practices & Patterns#state#design#primitives
Read article
A Noul returns one number and no separate confidence. How to read the value, pin its boundary with criteria, split compound questions, and set a per-action threshold that matches the cost of being wrong.
Jev Practices & Patterns#noul#primitives#confidence
Read article
A Score is a weighted position, not a ruler. Why the number can land between levels, why levels are judged separately, and how to combine several atomic scores in code instead of one crowded rubric.
Jev Practices & Patterns#score#primitives#scoring
Read article
From the Jev community
Posts from builders shipping with Jev right now.