questions: { stuck: boolean, destructive: boolean, }
jev-playground
TypeSafe AI の System One モデル Jev を MoonBit から触るためのプレイグラウンド。

Every project here really calls Jev in its source · refreshed daily
questions: { stuck: boolean, destructive: boolean, }
TypeSafe AI の System One モデル Jev を MoonBit から触るためのプレイグラウンド。
questions: { does_pass: boolean, }
Using Jev as an evaluator.
questions: { route: choice, toolFree: boolean, }
Jev-native AI security harness for autonomous research, multi-agent swarms, persistent hunt boards, and long-running agent workflows. CLI-first, open source, and built for authorized security research.
questions: { a: boolean, b: boolean, }
Inspired by TypeSafe Ai, Ask a local LLM typed questions, get calibrated probabilities instead of text. Structured output without generation or parsing. MLX / Apple Silicon.
questions: { urgent: boolean, }
Connect JEV to MCP clients and compare its judgments against general-purpose LLMs using shared datasets and measurable accuracy.
questions: { instructions: boolean, }
TypeSafe Jev routing for Oh My Pi, with an opt-in checkpoint orchestrator and editable XDG configuration. Requires Bun = 1.3.14 and OMP = 18.2.3.
questions: { done: boolean, }
A chatbot from typed Jev decisions: hierarchical speculative decoding over System One probabilities.
questions: { done: boolean, blocked: boolean, }
jev-browse is an unofficial project and isn't affiliated with TypeSafe or Vercel.
questions: { what: choice, }
Operator lab for TypeSafe Jev. Chess Arena, closed-schema booths, Stockfish HUD for review only.
questions: { department: choice, }
Measure when to use Jev and other models on your data, then route accordingly.
questions: { task_completion: score, instruction_following: boolean, grounded_in_evidence: boolean, // +5 }
Affordable for parallel online agent evals and observability. Powered by JEV.
questions: { action: choice, }
Jev and Microsandbox explore alternate game futures with a reusable TypeScript learning harness
questions: { label: choice, }
Owned, deterministic classifier for checking whether agent claims are supported by evidence, with optional Jev comparison through Cloudflare AI Gateway.
questions: { is_urgent: boolean, severity: score, }
Shadow-mode validation harness for a pre-execution firewall on AI agent tool calls (TypeSafe/Jev). Real run, findings in report.md.
questions: { is_regulation: boolean, mentions_deadline: boolean, mentions_fines: boolean, // +5 }
Never confidently wrong: a TLA+-verified consensus kernel around TypeSafe's Jev, run through 1,680 chaos-tested pharmacy decisions with zero wrong verdicts. Film, code, and every captured call.
questions: { topic: choice, requests_credentials: boolean, sender_identity_mismatch: boolean, // +4 }
Experiments on TypeSafe Jev (System One decision model) via OpenRouter: repeatability, perturbation, and LLM baseline comparison
questions: { failure_domain: choice, customer_impact: score, systemic_outage: boolean, }
A garage full of tiny experiments for building critical systems with System One & Jev 🔧🧠⚡
questions: { impact: score, }
TypeSafe AI (jev) 시범 사용 프로젝트
questions: { is_urgent: boolean, needs_human: boolean, is_financial: boolean, }
Compare GPT generated language with JEV structured Noul decisions on the same input.
parser = argparse.ArgumentParser(description="Local visual Choice/Noul/Score judgments")
An educational Jev-like visual inference experiment on Apple Silicon: shared context, direct candidate scoring, and local visual demos.
base_url: str = "https://api.typesafe.ai",
Read-only trading journal and review harness: Jev typed judgments, agent integration, and a reproducible finance benchmark. No orders, no advice.
"""回傳一個已設定好 key 的 TypeSafeClient;拿不到 key 就直接退出、說清楚為什麼。
Independent, evidence-based map of when TypeSafe's Jev actually holds up vs. breaks down — real API-call receipts, not a leaderboard. 中文為主的雙語 repo。
jev_raw=label['jev'], jev_reported=label['jev_reported'])
Typed JSON inference with DiffusionGemma, with Every and Jev benchmark results
const json = await postJson(fetchImpl, "https://api.typesafe.ai/v1/systemone", key, jevRequest(batch, model));
Jev vs Gemini 3.8 Flash: labelling 1,000 app reviews, 4.1× faster and 7× cheaper
from typesafe_sdk import Choice, TypeSafeClient
Probability-aware evaluation for typed decision models: calibration, selective risk, latency, and reproducible benchmarks.
id: i ? 'deepseek' : 'jev',
Jev 模型介绍与实测:通过 Choice / Score / Noul 将自然语言转为带类型的判断与概率,用于分类、评分和路由;支持与 DeepSeek 等模型对比评论打标、速度与结果,含 CSV/Excel 导入、原速回放与离线报告。
DEFAULT_BASE_URL = "https://api.typesafe.ai"
Stop guessing confidence thresholds: calibrate, threshold, and drift-check typed decision models (TypeSafe Jev) against an LLM teacher.
TRACKED_PACKAGES = ["httpx", "numpy", "pydantic", "PyYAML", "typer", "typesafe-sdk"]
Reproducible benchmark for measuring Jev reranking quality, latency, and cost in RAG
* TypeSafe-compatible /v1/systemone (native typed interface).
JevBench v1 - a benchmark for Jev-class typed decision models: smart, cheap, fast, reliable, open.
from typesafe_sdk import Choice, RetryPolicy, TypeSafeClient
Reproducible early-access evaluation of Jev on Korean understanding and medical text, with runtime and cost evidence
import typesafe_sdk
High-speed recursive AI Elo tournament engine powered by Jev and Swiss matchmaking
export const ENDPOINT = "https://api.typesafe.ai/v1/systemone";
A word-level language model whose output layer is Jev: n-gram drafter, Noul chunk verification, bits-per-token eval
self.model = os.environ.get("JEV_MODEL", "typesafe-ai/jev")A coding agent that filters every tool result through Jev before the model sees it, with an A/B harness measuring pass@1 and cost against the unfiltered control
import { experimental_evaluate as evaluate } from 'ai';Calibrated alignment verifier for LLM responses and agent plans — powered by Jev
for name,url,key in [('openrouter','https://openrouter.ai/api/v1/models',auth['OPENROUTER_KEY']),('typesafe','https://api.typesafe.ai/v1/models',auth['TYPESAFE_Comparing Jev, Gemini Flash, and Claude Fable on Python code review rules: cost, speed, accuracy, and consistency. Includes results, charts, and reproducible experiments.
// api.typesafe.ai allows no localhost CORS origins (and keys belong server-side).
A show-and-tell capability study for Jev, TypeSafe's System One decision model.
"jev": jev[k], "llm": llm[k], "label": None}
Does a TypeSafe Jev rerank beat embedding search? Graded relevance eval (9,831 pairs, 164 zh/en queries) over the Agent Skills Hub catalog, with the judge-circularity bias measured.
import type { Questions } from "@typesafe-ai/sdk";Decision harness for TypeSafe Jev — confidence gates, shadow mode, recipes, and evals. Claude CLI 48.9s → Jev 1.3s on the same row-filter job.
req=urllib.request.Request('https://api.typesafe.ai/v1/systemone',data=body,headers={'Authorization':'Bearer '+key,'Content-Type':'application/json','Cache-ContIndependent Jev 1.13.0 behavior study: report, controlled prompt experiments, raw results, and offline verification.
from typesafe_sdk import AsyncTypeSafeClient
TypeSafe Jev demonstration for new analyzation — experimenting with Jev for fast analysis of news and tickers
from typesafe_sdk import TypeSafeError
Make room for useful evidence. Inspectable context selection for RAG, with Jev reranking and open benchmark studies.
// Probability is the Noul answer: how likely Jev thinks the positive is.
Blind security benchmarks for Jev, TypeSafe's System One model: prompt injection and vulnerable code detection, built on jev-go
from typesafe_sdk import AsyncTypeSafeClient, Choice
Backtest Jev (TypeSafe) as a BUY/SELL/HOLD trader on NQ L10 order-book data
Calls the real ``https://api.typesafe.ai/v1/systemone`` endpoint. The
Evidence-driven frontend QA built on Jev Ultrafast and Browser Harness, with a synthetic todo demo.
d, wall, srv = post("api.typesafe.ai", "/v1/systemone", os.environ["TYPESAFE_API_KEY"], {"model": "jev-1.13.0", "state": state, "questions": questions})An independent benchmark of TypeSafe's Jev, a model that does not write text. You send it some content and a list of typed questions (yes/no, pick one option, rate on a scale) and
api.typesafe.ai; the CLI, the verifier and the lab harness all go through it.
Jev (TypeSafe) exploratory thread: claim audit, live demos, and runnable code
import { experimental_evaluate as evaluate } from 'ai';Independent calibration test of TypeSafe's Jev on a task it cannot have seen: 900 rule-generated support tickets (choice / score / boolean) plus 3 public benchmarks via Vercel AI Gateway. Raw responses, ECE with noise floor, temperature refit, per-type sign of miscalibration. Reproducible for ~$0.06.
const curlCmd = `curl -i -X POST https://api.typesafe.ai/v1/systemone \\
Autonomous System-One Triage Engine & Benchmark powered by TypeSafe AI (Jev). 75ms inference, $0 output tokens, and RLCD epistemic safety gates.
from typesafe_sdk import Choice, Noul, Score
Benchmarks and a playground for TypeSafe's Jev (System One) model: chess, and who-is-the-player-talking-to for speech-to-text game NPCs
MODEL = "~typesafe/jev-latest"
Cross-domain check on MIND news: a zero-shot Jev headline prior is worth ~500 labelled articles, adds +0.069 ρ as features, and lifts a Thompson-sampling cold start by 25%.
JEV_URL = "https://api.typesafe.ai/v1/systemone"
Benchmark TypeSafe Jev against any OpenRouter model on your own labelled classification data: accuracy, calibration, latency, cost
MODEL = "typesafe/jev-1.13"
Mini benchmark of TypeSafe's jev-1.13 structured decision model (OpenRouter Decisions API) on labeled support-triage: noul/choice/score, consistency, cost, lessons learned
const BASE_URL = "https://api.typesafe.ai";
Test bench for TypeSafe's Jev
typesafe = env("TYPESAFE_BASE_URL", "https://api.typesafe.ai").rstrip("/")Jev (TypeSafe) vs Claude Haiku 4.5 on 2 000 phishing emails: accuracy, calibration, latency, cost. Reproducible benchmark.
API_URL = "https://api.typesafe.ai/v1/systemone"
NYC 311 complaint heatmaps with TypeSafe JEV: reproducible pipeline, live research results, and interactive geographic visualizations.
import { TypeSafeClient } from "@typesafe-ai/sdk";Local evaluation workbench for TypeSafe Jev
ENDPOINT = "https://api.typesafe.ai/v1/systemone"
Jevaluate: evaluate before you trust. Field notes, runnable scripts and an agent skill for TypeSafe Jev: gated evals, a browser loop, a product walk with DeepSeek vision, a UI text judge and a first-click tree test. Co-authored with Claude Fable 5.1.
export const MODELS = { chat: 'google/gemini-3.7-flash', decision: 'typesafe/jev-1.13' } as const;A repository for jev-expirements
let md = `# Results\n\nGenerated by \`node eval.js\` on ${new Date().toISOString().slice(0, 10)} with \`${process.env.JEV_MODEL ?? 'typesafe/jev-1.13'}\`.\n\n`;Experiment: using TypeSafe Jev as a chatbot by choosing replies one letter or word at a time
ENDPOINT = "https://api.typesafe.ai/v1/systemone"
Jev x NetHack: bounded runner, research code, and completed recording releases
"""Thin wrapper around TypeSafe AI's official Python SDK (`typesafe-sdk`,
Google ADK vs Microsoft Agent Framework for structured-output agents, with TypeSafe Jev as a vendor-neutral QC gate
ap.add_argument("--model", default=os.environ.get("JEV_MODEL", "typesafe/jev-1.13"))Jev-shaped typed-decision model (state + Choice/Score/Noul questions -> calibrated probabilities, one pass) on ModernBERT / DeBERTa / LLaDA-MoE, with measured latency, accuracy, calibration and training cost
API_URL = "https://api.typesafe.ai/v1/systemone"
TypeSafe (Jev) vs DeepSeek-flash: side-by-side speed/token/cost/accuracy comparison across invoice extraction, email classification, and reranking
.unwrap_or_else(|_| "https://api.typesafe.ai/v1/systemone".into()),
A playground for experiments around Jev, TypeSafe's System One model.
import typesafe_sdk as sdk
experiments with system one model jev
JEV_URL = "https://api.typesafe.ai/v1/systemone"
Jev (TypeSafe System One) × ASReview SYNERGY abstract screening demo — Choice/Noul vs gold labels
if (kind === 'typesafe') return 'https://api.typesafe.ai/v1/systemone';
A browser-first AI poker benchmark. Seat Jev and OpenAI-compatible models at the same no-limit Texas Hold'em table, watch every card and decision as a spectator, and let the tournament run autonomously until one model wins.
APIURL = "https://api.typesafe.ai/v1/systemone"
Results page: https://claude.ai/code/artifact/f00ee126-9554-4e2f-b2e7-1fc86c066aa9
URL = "https://api.typesafe.ai/v1/systemone"
Jev (TypeSafe AI) 可复现实测:算术、计数、日期、零幻觉、把握度校准,中英对照
from typesafe_sdk import AsyncTypeSafeClient, RetryPolicy
AI detection in research papers with Jev
import { choice, TypeSafeClient, type EntryType } from '@typesafe-ai/sdk';Reproducible benchmark evaluating JEV as a software decision primitive for dependency-update automation under distribution shift.
req=urllib.request.Request('https://api.typesafe.ai/v1/systemone',data=json.dumps(p).encode(),headers={'Authorization':'Bearer '+self.key,'Content-Type':'applicReproducible evaluation of TypeSafe Jev on all 58,492 BBQ questions: accuracy, stereotype bias, uncertainty, cost and latency.
import { experimental_evaluate as evaluate } from 'ai';iOS 6-inspired Jev calculator demo with a lifetime API budget
.post("https://api.typesafe.ai/v1/systemone")File-scoped maintainability review with TypeSafe Jev
jev_api_url: str = "https://api.typesafe.ai/v1/systemone"
A local playground for comparing Laya and Jev decision models with article recommendations.
const JEV_URL = process.env.JEV_URL || 'https://api.typesafe.ai/v1/systemone';
Every token passes the gate before the screen.
DEFAULT_ENDPOINT = "https://api.typesafe.ai/v1/systemone"
Benchmarking Jev (Typesafe.ai) against a strong LLM on the Who&When Pro agent-failure-attribution benchmark (text subset).
"""Client for TypeSafe's Jev decision model (POST /v1/systemone).
Benchmarking TypeSafe's Jev decision model as a cost-efficient LLM router on RouterArena
MODEL = "~typesafe/jev-latest"
Can a TypeSafe Jev prior read from a README on day one predict which new agent-skill repos gain stars? Zero-shot Jev ≈ a text model trained on ~150 labels; best used as a feature. Prospective test running.
MODEL = "~typesafe/jev-latest"
Can a Jev-labelled GitHub issue stream catch a broken release before the fix? No at daily cadence (null, n=7). Per issue, Jev matches triage labels far better than keywords or sentiment.
MODEL = "~typesafe/jev-latest"
Does a Jev-labelled support-tweet stream spike before a brand admits an outage? At equal false alarms it catches 17 vs 10 incidents (volume), ~4h ahead; a good keyword list is almost as good.
python scripts/run_baseline.py --model jev --base-url https://api.typesafe.ai/v1 --api-key-env TYPESAFE_API_KEY
Typed-decision benchmark from PadFlow (land development SaaS): schemas, anonymized labeled rows, and a runner for confidence-calibrated models like TypeSafe Jev.
const TYPESAFE_ENDPOINT = 'https://api.typesafe.ai/v1/systemone';
Jev-powered integer sorting experiment: serial selection versus parallel rank prediction.