Read & learn
Written guides.
Understand Jev, one idea at a time. Walkthroughs, recipes, and writeups — our notes first, the original next.
A public registry · independent & community-run
Explore the Jev ecosystem.
Learn, build, and play with Jev — from deep technical guides to creative experiments.
Watch & learn
Video guides.
See an idea take shape. Tutorials, demos, and deep dives, organized by topic and credited to their creators.
Make & explore
Experiments.
See what builders made with Jev. Games, tools, repositories, articles, and more — traced to their sources.
The directory03 / 03 · Experiments
Find something worth exploring.
Every category, from games and tools to repositories and writeups, traced to a primary source.
Same ideas.
Different paths.
This is the full registry, not just the featured picks. Figures like speed, cost, and stars are maker-reported or captured snapshots, not JevMade measurements.
1613 experiments · showing 1081–1140
Lets Cline browse inside an isolated Chromium session by choosing indexed DOM actions instead of guessing screen coordinates.
SOURCE SCREENSHOTEXPAND ↗
A paper-trading loop for BTC and ETH that tests Jev's market judgments under explicit fees, latency and risk limits.
SOURCE SCREENSHOTEXPAND ↗
Replays historical stock data to compare Jev-directed paper trades with moving-average and buy-and-hold baselines.
SOURCE SCREENSHOTEXPAND ↗
A Chinese-language experiment suite for Jev's context limits, bilingual judgments, batching and game decisions.
SOURCE SCREENSHOTEXPAND ↗
A read-only crypto alert bot that studies settled signals and adjusts its filters without ever placing a trade.
SOURCE SCREENSHOTEXPAND ↗
A browser agent that asks Jev to choose from the buttons and controls on the current page.
SOURCE SCREENSHOTEXPAND ↗
Compacts Pi history without generated summaries, keeping selected tool exchanges and optional assistant prose exactly as they appeared.
SOURCE SCREENSHOTEXPAND ↗
Batches small agent-control judgments into one validated plan before any generative model is allowed to run.
SOURCE SCREENSHOTEXPAND ↗
Keep track of Codex workers with Jev progress checks, then accept their work only when your chosen command passes.
SOURCE SCREENSHOTEXPAND ↗
A browser shooter with Jev directing the tanks and riflemen you face.
SOURCE SCREENSHOTEXPAND ↗
from the source
const providerSchema = z.object({ model: z.string().min(1).max(100), answers: z.object({ tactic: z.object({ type: z.literal("choice"),
A reference decision layer for Minecraft that turns typed judgments into tool calls behind hard-coded survival reflexes.
SOURCE SCREENSHOTEXPAND ↗
from the source
export const API = { baseUrl: 'https://api.typesafe.ai', evaluatePath: '/v1/systemone', modelsPath: '/v1/models', apiKeyEnv: 'TYPESAFE_API_KEY',
A GraphRAG talk demo where Jev routes questions, checks evidence and decides whether deeper retrieval is needed.
SOURCE SCREENSHOTEXPAND ↗
from the source
JEV_MODEL = os.getenv("DEMO_JEV_MODEL", "jev-1.13.0")JEV_URL = "https://api.typesafe.ai/v1/systemone"LAYA_REPO = os.getenv("DEMO_LAYA_REPO", "convaiinnovations/laya")LAYA_MODEL_ID = "laya-421m-en"llm.PRICES_PER_MTOK[JEV_MODEL] = (0.042, 0.0)
Tell Jev what you want to watch and let it pick from a catalog of 4,800 films.
SOURCE SCREENSHOTEXPAND ↗
from the source
class ChoiceQuestion(BaseModel): type: Literal["choice"] = "choice" instructions: JSONContent = None criteria: dict[str, JSONContent]
A wildfire console that verifies citizen reports, rates evacuation routes and chooses a tactical response as the incident changes.
SOURCE SCREENSHOTEXPAND ↗
from the source
import { TypeSafeClient, choice, noul, score } from "@typesafe-ai/sdk";/** * Server-side TypeSafe Jev client. Never import this from a client component: * the API key must stay on the server (Next.js Route Handlers only).
Adds an optional second opinion to fraud investigations, comparing Jev's action and grounding check with the existing evidence pipeline.
SOURCE SCREENSHOTEXPAND ↗
from the source
questions={ "recommended_action": { "type": "choice", "instructions": "Given this fraud-risk evidence for one transaction, whi... "criteria": ACTION_CRITERIA,
A retro platformer where Jev chooses maneuvers and detours while code keeps control of physics, jumps and pit-saving reflexes.
SOURCE SCREENSHOTEXPAND ↗
from the source
import { choice, noul, score } from "@typesafe-ai/sdk";import type { Personality } from "@/lib/game/types";const MANEUVERS = { advance: {
Type a line and watch a 3D avatar choose one of fourteen performance styles and play the matching animation.
SOURCE SCREENSHOTEXPAND ↗
from the source
const TS_URL = "https://api.typesafe.ai/v1/systemone";const KEY_ITEM = "jev-stage.typesafe.key";function buildQuestions(){ const crit = {};
A Clash Royale-style sandbox where one request chooses whether to play, which affordable card, which lane and how deep.
SOURCE SCREENSHOTEXPAND ↗
from the source
const response = await client.systemOne({ state: { gameTimeRemaining: Math.round(state.timeLeftSeconds), myState: { playerId: 1,
Review your Valorant rounds with ratings from local rules or an optional Jev judge.
SOURCE SCREENSHOTEXPAND ↗
from the source
const payload = { state: input.state, model: this.model, questions: toPayloadQuestions(questions), }
A local second brain where Jev reranks search evidence and suggests useful links between related notes.
SOURCE SCREENSHOTEXPAND ↗
from the source
export function passageScoreQuestion(query: string, pointer: string): JudgmentQuestion { return { type: 'score', instructions: `Score how useful the passage at \`${pointer}\` is as evidence for the... criteria: [
A browser colony simulation where support tickets carry real text and Jev assigns the categories agents use to route them.
SOURCE SCREENSHOTEXPAND ↗
from the source
var INSTRUCTIONS_PREFIX = 'Classify the support ticket in '; var CRITERIA = { incident: 'Something that was working is broken or degraded right now and a ' + 'user is blocked or impaired. The ticket reports a failure, an outage, an ' +
Play trivia by assigning probabilities, then compare your calibration with Jev and a coin flip.
SOURCE SCREENSHOTEXPAND ↗
from the source
export const JEV_MODEL = "jev-1.13.0";const ENDPOINT = "https://api.typesafe.ai/v1/systemone"; state, questions: { verdict: { type: "noul", instructions } },
A top-down duel where Jev chooses among reachable positions and local code handles paths, leading aim and firing.
SOURCE SCREENSHOTEXPAND ↗
from the source
const questions = { move: { type: "choice", instructions: { doctrine: DOCTRINE,
An original platformer and reinforcement-learning lab testing whether frozen Jev risk features help PPO learn faster.
SOURCE SCREENSHOTEXPAND ↗
from the source
API_URL = "https://api.typesafe.ai/v1/systemone"MODEL = "jev-1.13.0"ACTION_NAMES = tuple(action_names("simple"))ACTION_CRITERIA = { "noop": "Release all buttons; coast or wait, and release jump.",
A Spanish-language cancer podcast library where Jev sorts new videos by format, topic and relevance before code decides publication.
SOURCE SCREENSHOTEXPAND ↗
from the source
export const editorialQuestions={ category:{type:'choice',instructions:'Identify the PRIMARY topic of this video from its...
A three-level interrogation game where your statement receives a release probability, persuasiveness score and persuasion tactic.
SOURCE SCREENSHOTEXPAND ↗
from the source
should_release: { type: 'noul', instructions: instruction, criteria: { true: 'Th... persuasiveness: { type: 'score', instructions: criteria.persuasivenessInstructio... tactic: { type: 'choice', instructions: criteria.tacticInstructions, criteria: c...
A collaborative story game where players write every line and Jev chooses the winner, screens content and senses an ending.
SOURCE SCREENSHOTEXPAND ↗
from the source
questions[`dim_${d.key}`] = { type: 'choice', instructions: d.instructions, criteria... questions[`end_${c.id}`] = { type: 'noul', instructions: `If this sentence were added next, would the story feel finished, li...
Political Compass and Pew typology probes for Jev, alongside a deliberately awkward character-by-character text decoder.
SOURCE SCREENSHOTEXPAND ↗
from the source
def choice(instructions: str, criteria: dict[str, str | None]) -> dict: """A Choice question: pick one of up to 255 options.""" return {"type": "choice", "instructions": instructions, "criteria": criteria}
Follows one Jev-controlled gambler through spins and shopping, while code owns the reels, bankroll and intentionally losing arithmetic.
SOURCE SCREENSHOTEXPAND ↗
from the source
export function gamblingQuestions(s, ladder, { canSpend }) { const questions = { bet_size: choice('How much does Jev bet?', betCriteria(ladder, s.balance)), pace: score('How fast is Jev playing?', [
Turns a text and a pack of criteria into a shareable wall of verdict cards, appearing as each batch returns.
SOURCE SCREENSHOTEXPAND ↗
from the source
const TYPES = new Set(['noul', 'score', 'choice']); if (item.type === 'score' && (!Array.isArray(item.criteria) || item.criteria.length...function asQuestion(item) { return { type: item.type, instructions: item.instructions, ....
Play oversized tic-tac-toe against Jev, without a search algorithm rescuing its mistakes.
SOURCE SCREENSHOTEXPAND ↗
from the source
export function buildJevRequest(snapshot) { const { size, board, currentPlayer, legalMoves } = jevState(snapshot); const opponent = otherPlayer(currentPlayer); const showReplies = board.filter((mark) => !mark).length <= MAX_REPLY_CONTEXT_EMPTY_CE... const criteria = [
A crossword agent where Python manages the grid and crossings while Jev ranks short wordlist candidates for each clue.
SOURCE SCREENSHOTEXPAND ↗
from the source
answers = self._jev( state, { "fits": { "type": "noul",
A Portuguese platformer where typing a category changes which objects float, bridge, bounce or turn ghostly.
SOURCE SCREENSHOTEXPAND ↗
from the source
test("API request asks one constrained Noul per object", () => { const request = makeRequest("madeira"); assert.equal(Object.keys(request.questions).length, 24); assert.equal(request.state.category, "madeira");});
A maze lab that reveals only local facts to Jev and measures the resulting path against breadth-first search.
SOURCE SCREENSHOTEXPAND ↗
from the source
const DIRECTION_LABEL: Record<Direction, string> = { up: "Move one cell up (north)", down: "Move one cell down (south)", left: "Move one cell left (west)", right: "Move one cell right (east)",
Watch Jev play heads-up Hold'em against a heuristic bot, with the probabilities behind each move visible beside the table.
SOURCE SCREENSHOTEXPAND ↗
from the source
ENDPOINT = "https://api.typesafe.ai/v1/systemone"CLIENT = httpx.Client(http2=True, timeout=30)POKER_RULES = ( "You are a top heads-up No-Limit Texas Hold'em player. Choose exactly one offered ac...
Battleship where Jev returns a probability of a ship for every open cell, creating both the heat map and the shot.
SOURCE SCREENSHOTEXPAND ↗
from the source
def argmax_cell(probabilities: Mapping[str, float], cells: Sequence[Cell]) -> Cell: if not cells: raise ValueError("no cells to choose from") best_cell = cells[0] best_probability = float(probabilities.get(cell_id(best_cell), 0.0))
A Telegram fantasy league and token-research copilot that asks Jev whether deeper Nansen data is worth the credits.
SOURCE SCREENSHOTEXPAND ↗
from the source
"worth_deep_dive": { "type": "noul", "instructions": "Is this token move worth spending 5 credits of deep Nan... },
A Processing treasure hunt where each robot sees five sectors and Jev chooses its movement and turn size.
SOURCE SCREENSHOTEXPAND ↗
from the source
"action": { "type": "choice", "choice": "forward", "probabilities": { "forward": 0.55, "backward": 0, "turn_left": 0.45, "turn_right"... "confidence": 0.4
Run a Mineflayer survival bot where Jev picks the next high-level action and code handles the execution knowledge.
SOURCE SCREENSHOTEXPAND ↗
from the source
id: "next_action", type: "choice", instructions: "現在の状況(HP/満腹度と食料の所持有無/周辺の敵性エンティティとその種別/周辺の動... "夜間かどうか/装備/木材・道具のクラフト進行状況(crafting)/直近の行動履歴/周辺地...
A twenty-questions game that tries to guess a Japanese persona's home prefecture and age.
SOURCE SCREENSHOTEXPAND ↗
from the source
export type NoulQuestion = { type: "noul"; instructions: string | object; criteria?: { true?: string; false?: string };};
A roguelike dungeon director where Jev sketches the next room's role, size, danger, enemies, loot and secrets.
SOURCE SCREENSHOTEXPAND ↗
from the source
"room_type": { "type": "choice", "choice": "chamber", "confidence": 0.8, "probabilities": {"chamber": 0.8, "room": 0.2},
An agent that predicts what actions will do, notices confident mistakes and rewrites its explicit model of the world.
SOURCE SCREENSHOTEXPAND ↗
from the source
model: this.opts.model ?? 'jev-latest', state: { world: stateText(state), action: actionText(action), ...(withRules ? { co... questions: Object.fromEntries(questions.map(q => [q.id, { type: 'noul', instructio... }
A small playground that turns one line into several visible Jev judgments about romance, social posting or city routing.
SOURCE SCREENSHOTEXPAND ↗
from the source
import { choice, noul, score, type Questions } from '@typesafe-ai/sdk';romantic_frame: choice(love_signal_strength: score(is_passionate: noul('Does `text` read as intense, burning romantic feeling?', {
Play music and get one of twenty house cocktails chosen to match its sound.
SOURCE SCREENSHOTEXPAND ↗
from the source
const upstream=await fetch('https://api.typesafe.ai/v1/systemone',{method:'POST',headers...const data=await upstream.json(), answer=data.answers?.cocktail;const recipe=candidates(input).find(r=>r.id===answer?.choice);
Runs Jev and a frontier model down parallel lanes, answering the same timed traffic-hazard choices and comparing the scorecards.
SOURCE SCREENSHOTEXPAND ↗
from the source
export const CALL_CRITERIA = { left: "Swerve left. The hazard is real and the left side of the road is open.", right: "Swerve right. The hazard is real and the right side of the road is open.", hold: "Hold the line. Do not swerve. The markings disagree with the open side, neither...
A local research desk where Jev chooses methods, searches and leads, then checks evidence and claims under fixed budgets.
ACTUAL RECORDING12 SEC ↗
from the source
from typesafe_sdk import AsyncTypeSafeClient,Choice,Score,Noul,RetryPolicy if a.get('type')!=q.type: raise DecisionError('Answer type mismatch') if q.type=='choice': if a.get('choice') not in q.criteria or set(a.get('probabilities',{}))!=set(...
A pinball table that keeps moving while Jev chooses flipper states, applying only answers fresh enough to matter.
SOURCE SCREENSHOTEXPAND ↗
from the source
questions: { flippers: { type: 'choice' as const, instructions: 'Choose the flipper buttons to hold NOW to keep this moving ball in play... criteria: { left: 'Press left flipper; release right.',
A Factorio agent loop where Jev picks the macro goal and next action while deterministic code owns rules and actuation.
SOURCE SCREENSHOTEXPAND ↗
from the source
def evaluate(self, state: dict, questions: dict) -> dict: """One system-one call. Returns the `answers` map keyed by question id.""" resp = requests.post( headers={"Authorization": f"Bearer {self.api_key}"}, json={"state": state, "model": self.model, "questions": questions},
A controlled city simulation comparing four kinds of institutional memory, with Jev scoring each term beside a deterministic scoreboard.
SOURCE SCREENSHOTEXPAND ↗
from the source
ENDPOINT = "https://api.typesafe.ai/v1/systemone"MODEL = os.getenv("TYPESAFE_MODEL", "jev-latest") "prosperity": { "type": "score", "instructions": "Prosperity over the whole term: jobs relative to population at...
Add classification, probability and scoring columns to CSV rows from questions written in YAML.
SOURCE SCREENSHOTEXPAND ↗
from the source
questions: - name: department type: choice question: Which team should handle this message? options:
A Magic 8-Ball that first checks whether your question is askable, then judges which way the outlook leans.
SOURCE SCREENSHOTEXPAND ↗
from the source
askable: noul( "Is `question` a yes-or-no question about the future or the unknown, the kind a Magi... { true: { meaning: "A question that can be answered yes or no",
Type a word or phrase and see Jev spread it across sixteen fixed colors in a bar chart.
ACTUAL RECORDING12 SEC ↗
from the source
const QUESTIONS = { colour: { type: 'choice', instructions: 'A person typed a word or phrase. If they had to paint it using only these sixteen...
Jev dispatches taxis through a simulation of Baku's streets.
SOURCE SCREENSHOTEXPAND ↗
from the source
const INSTRUCTIONS = { surucu: "Bu sifarişi hansı sürücü götürməlidir? Məsafə, gözlənilən çatma vaxtı və reyt... tecililik: "Sərnişinin mesajına əsasən sifariş nə qədər təcilidir?", saxta_sifaris: "Bu sifariş saxta, zarafat və ya sui-istifadə cəhdidir?", xidmet_tipi: "Sərnişinə hansı xidmət tipi uyğundur?",
A customer-support demo that uses GPT-4o to write replies and Jev to check them.
SOURCE SCREENSHOTEXPAND ↗
from the source
JEV_URL = "https://api.typesafe.ai/v1/systemone"JEV_MODEL = "jev-1.13.0" # pinned on purpose; "jev-latest" can change under youHAZARD_BLOCK_AT = 0.5INTENT_MIN_CONFIDENCE = 0.5EVAL_PASS_AT = 0.75
A small web crawler that asks Jev which page to visit next.
SOURCE SCREENSHOTEXPAND ↗
from the source
API = os.environ.get("JEV_API_URL", "https://api.typesafe.ai/v1/systemone")DEFAULT_MODEL = os.environ.get("JEV_MODEL", "jev-latest")def yes(answer: dict, key: str) -> bool: """Read a ``noul`` answer as a boolean.""" return bool((answer.get("answers", {}).get(key) or {}).get("noul"))
A tokenized-stock trading demo that uses Jev to interpret what you want to do.
SOURCE SCREENSHOTEXPAND ↗
from the source
execution_intent: { type: "choice", instructions: "What does the user want to do about `request.text` given `mar... criteria: { market_now: "User explicitly accepts current on-chain price and wants imme...
Keep a coding agent out of the click loop while Jev chooses bounded browser operations and compatible on-page targets.
SOURCE SCREENSHOTEXPAND ↗
from the source
questions.insert( "operation".into(), json!({ "type": "choice", "criteria": Value::Object(operations),
Turns Jev probability distributions into one plain-language sureness verdict, from certain through torn to clueless.
SOURCE SCREENSHOTEXPAND ↗
from the source
QUESTIONS = { "department": {"type": "choice", "instructions": "Which team should handle this mess... "severity": { "type": "score", "instructions": "How severe is the reported issue? If there is no issue, pick th...
Label rows in a CSV or JSONL file with Jev, and set uncertain answers aside for a person to review.
SOURCE SCREENSHOTEXPAND ↗
from the source
questions: category: type: choice instructions: "Is `message` unsolicited commercial spam, or a message the recipient... options:
A parody type guard that asks Jev whether a value is a record—and makes another call for the negative form.
SOURCE SCREENSHOTEXPAND ↗
noul
Is `value` a record? · non-null object that is not an array / null, array, primitive, or otherwise not a record
Jev, briefly
A decision model, not a chatbot.
A chat model generates text token by token and hopes you parse it. Jev never generates a word. Your code sends the state of the world plus typed questions; Jev returns every answer in one parallel pass — typed values with calibrated probabilities and a confidence score your code can trust. About 70–500 ms end to end, $0.042 per million input tokens, output free. Trained with what TypeSafe calls Reinforcement Learning for Calibrated Decisions.
choice Pick one of these.
You define the options. Jev returns the chosen option, a probability for every option, and a confidence. The workhorse of routing, agents, and games — the answer space is always legal, so there is nothing to parse and nothing to hallucinate.
score Rate this on a rubric.
You describe ordered levels — say trivial / normal / critical. Jev returns the level, the probability of each, and a confidence. Scores turn fuzzy judgment (“how severe is this log line?”) into a number-free decision your code can branch on.
noul Is this statement true?
You assert a statement; Jev returns the probability that it is true — a noul. Moderation, verification, guardrails, “does this diff actually fix the bug?”: one question, one calibrated probability.
Official material lives at typesafe.ai and docs.typesafe.ai. JevMade is an independent community registry — not affiliated with or endorsed by TypeSafe AI.
Recurring lessons
Patterns that keep showing up.
One decision per tick
Games, drones, trading bots, and browser agents all converge on the same loop: serialize the state, ask one decisive question, act, repeat. Jev’s latency makes the loop feel instant — the model lives inside the control loop, not outside it.
Confidence as a gate
The answer says what; the confidence says whether to act. The most reliable entries threshold on confidence to route edge cases to a slower model or a human — automation with an honest escape hatch.
Shrink the choice space in code
Makers rarely ask Jev to pick from everything. Local tactics prune 225 gomoku moves to ~40; DOM filters turn a page into an element table; code narrows options, Jev judges within them.
Jev judges, LLMs talk
A recurring split: Jev makes every decision cheaply and instantly, and a small LLM is only invoked when a human-facing string must actually be written. Decision and generation are separate budgets.
Made something with Jev?
Send the link — repo, live demo, post, or video — plus a line on what it does and which primitives it uses. Every entry is verified against its primary source before it ships.
Submit your experiment → hello@JevMade.com