How does a classifier play chess without illegal moves?

Each turn becomes one Choice question whose options are the legal moves, which always fits the option limit, so a System One model cannot return an illegal move by construction.

How do I read whether a model actually played well?

Stockfish ranks every played move against all legal alternatives by centipawns and percentile versus random play, with distance and rank breakdowns plus worst-loss columns.

Do I need an API key just to look at the bundled games?

Run the published container and open the saved tournament for replay with no key; live play and new tournaments need a gateway key or a per-session browser key.

What changes if I play the offline model instead of Jev?

Pick the local model as the opponent; it needs a larger disk footprint and receives bare notation labels because its option token budget is smaller than the full move descriptions.

How do I start it with Docker?

Run docker run --rm -p 127.0.0.1:8000:8000 ghcr.io/dperezcabrera/ai-chess-lab:latest, then open http://localhost:8000/tournament and pick Tournament, Saved, Review; the 72-game September 2026 tournament is bundled and replays with no key.

How do I keep my tournaments?

Tournaments are saved in /app/tournaments inside the container, so they go when it is removed; mount a folder there to keep them, and copy results/20260922-141451-7ccf.json in to preserve the bundled tournament too.

What are Kev and Laya here?

Both are open System One models that answer the same typed questions: Laya (Apache-2.0) runs locally at no per-move cost, and Kev (0.8B, 4B, 9B) speaks the /v1/systemone protocol and plays with full descriptions and up to 255 options.

CommunityJev Integrations & Adapters

AI Chess Lab

Browser chess lab where LLMs, System One models, and humans play single games, Swiss tournaments, and leagues with Stockfish analysis and cost tracking.

See highlights

Open github.com

Stars: 4
Language: Python
License: GPL-3.0
Published by: Community
Author: dperezcabrera
Last verified: Oct 2026

dperezcabrera/ai-chess-lab

User repository on GitHub

View on GitHub

Chess against Jev, TypeSafe AI's System One model, through OpenRouter. Built with the pico framework.

Forks: 1
Last pushed: Updated Oct 2026

Repo stats from the GitHub API, cached Oct 2026.

About this project

ai-chess-lab: LLMs, System One models and you play chess in the browser, one game at a time or in Swiss tournaments and full leagues, with every move, token, second and dollar on the record and Stockfish analysing the games in your browser. The first battle, 16 players and 72 games, is in the repository.

Quoted from the project's README. Read the full README.

Tagscommunity github

Works forJev for Routing

Used inJev for Software & DevTools

Languages: Python

Highlights

Maps every chess turn to one Choice over legal moves, letting chat models and decision models share the same position.
Runs Swiss rounds and full leagues with per-move saves, standings, tie-breaks, and round boards.
Evaluates games locally with Stockfish charts, centipawn loss, and inaccuracy, mistake, and blunder counts.
Tracks calls, tokens, latency, and dollars live per game, with PGN export and Docker replay support.

Watch out

Jev is a fast classifier rather than a chess engine, so expect plausible moves rather than strong play in this demo of the decision pattern. 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.

More Jev resources like this

systemone

Compatibility server exposing the /v1/systemone typed-decision API on top of vLLM, SGLang, or OpenRouter models, with vision models enabling image inputs that Jev and Laya lack.

Jev Integrations & Adapters#community#github

Community

infinitylogesh/systemonePython

grok-jev-router

Dependency-free Python toolkit that puts Jev in front of Grok Bot, turning calibrated answers into proceed, dry_run, ask_human, or skip with hard-coded safety overrides.

Jev Integrations & Adapters#community#github

Community

colinmcdermott/grok-jev-routerPython

jev-site-auditor

Site-wide content auditor combining free deterministic regex and structural checks with one batched Jev judgment call per page.

Jev Integrations & Adapters#community#github

Community

tjkimcloud/jev-site-auditorPython

AutoGPT

187.5k

Autonomous agent platform with 187,504 stars for building and running agents from plain English or a visual builder. Its platform features first-class TypeSafe Jev decision blocks for typed routing, filtering, scoring and confidence-gated next-action dispatching.

Jev Integrations & Adapters#community#autonomous-agents#platform

Community

Significant-Gravitas/AutoGPTPython

LiteLLM

60.9k

LiteLLM's proxy adds TypeSafe guardrail hooks and a Jev complexity classifier for policy judgments and request routing across 100+ model providers.

Jev Integrations & Adapters#community#python#gateway

Community

BerriAI/litellmPython

AgentScope

33k

Alibaba's multi-agent platform, which implements native TypeSafe Jev classification through JevClassifierModel for binary, choice and score routing across agent pipelines. Python 3.11+, Apache-2.0, with a FastAPI agent service and web UI.

Jev Integrations & Adapters#community#multi-agent#framework

Community

agentscope-ai/agentscopePython

Back to all resources

From the Jev community

Posts from builders shipping with Jev right now.

Follow @typesafeai