Read & learn
Written guides.
Understand Jev, one idea at a time. Walkthroughs, recipes, and writeups — our notes first, the original next.
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.Experiments
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.
311 experiments · showing 121–180
Gives Claude Code's reviews, debugging ideas, plans and searches a separate probabilistic second opinion.
SOURCE SCREENSHOTEXPAND ↗
A language and runtime for budgeted agent programs whose declared decisions can be answered by Jev and replayed later.
SOURCE SCREENSHOTEXPAND ↗
from the source
export const TYPESAFE_SYSTEMONE_URL = "https://api.typesafe.ai/v1/systemone" as const;export const JEV_DEFAULT_MODEL = "jev-latest" as const;export const JEV_CREDENTIAL_ENV = "TYPESAFE_API_KEY" as const;
Jev helps Pi choose skills, rank source files and understand failed commands.
SOURCE SCREENSHOTEXPAND ↗
from the source
export type Question = | { type: "noul"; instructions: string; criteria?: { true: string; false: string } } | { type: "choice"; instructions: string; criteria: Record<string, string | null> };export type Questions = Record<string, Question>;export type Answer =
Uses Jev to keep useful Codex tool output while replacing low-value bulk with bounded excerpts.
SOURCE SCREENSHOTEXPAND ↗
Tracks an imaginary brain budget for Codex using local Laya scores through a Jev-compatible request shape.
SOURCE SCREENSHOTEXPAND ↗
Reviews code against plain-language team rules, using Jev to turn each rule into a probability-backed finding.
SOURCE SCREENSHOTEXPAND ↗
A Pi extension masks stale tool output after Jev review while retaining paged access to the original session evidence.
SOURCE SCREENSHOTEXPAND ↗
An Agent Skill and CLI convert decision prompts or calling code into validated Jev requests and runnable integration scripts.
ACTUAL RECORDING12 SEC ↗
Review changed code for stylistic warning signs before it ships, with Jev mapping findings to a small verdict set.
SOURCE SCREENSHOTEXPAND ↗
Run tool-using agents where Jev makes bounded decisions, deterministic code executes, and an LLM handles language when needed.
SOURCE SCREENSHOTEXPAND ↗
JevGate reviews code, tests and documentation with small typed questions, then turns the answers into located findings.
SOURCE SCREENSHOTEXPAND ↗
from the source
let first: Vec<_> = plan.requests.iter().map(Task::unit).collect();self.dispatch(report, first, |file, asked, body| { crate::units::record(file, &asked, body)})?;compose_files(&plan, report);
A Jev-powered pull-request risk router and code-review coach.
SOURCE SCREENSHOTEXPAND ↗
A Hermes plugin for typed decisions, ranking, verification, and optional tool gating with Jev.
SOURCE SCREENSHOTEXPAND ↗
from the source
hermes config set plugins.entries.hermes-jev.settings.jev_model typesafe/jev-1.13 --forc...hermes config set plugins.entries.hermes-jev.settings.timeout_seconds 15 --force
A Python command-line agent that uses Jev while working through file-based questions.
SOURCE SCREENSHOTEXPAND ↗
from the source
uv sync --group devcp .env.example .env# Fill in TYPESAFE_API_KEY and OPENAI_API_KEY in .env.uv run jevex 'Read prices.md. How much do three notebooks cost?' --trace run.jsonl
Give the Pi coding agent a way to ask Jev for decisions while it works.
SOURCE SCREENSHOTEXPAND ↗
Slidepilot advances Slidev presentations by matching live speech to the next slide with Jev.
SOURCE SCREENSHOTEXPAND ↗
from the source
pnpm installcp apps/worker/.dev.vars.example apps/worker/.dev.vars# Uncomment TYPESAFE_API_KEY in apps/worker/.dev.vars for real Jev inference.# Terminal 1pnpm dev:worker
Ask Jev questions from the command line, with answers as choices, scores or yes-or-no probabilities.
SOURCE SCREENSHOTEXPAND ↗
from the source
typesafe info # endpoints; does not print the keytypesafe modelstypesafe smoke # docs quickstart exampletypesafe ask --state "…" --noul is_urgent="Does this convey urgency?"typesafe ask --state-file s.json --questions-file q.json --json
A Pi extension that checks risky tool calls and notices when a coding agent gets stuck.
SOURCE SCREENSHOTEXPAND ↗
Search a Neovim buffer by describing behavior, then see matching functions ranked in quickfix and beside the code.
SOURCE SCREENSHOTEXPAND ↗
Turns a straightforward request into an MCP tool call by choosing a discovered tool and filling simple arguments.
SOURCE SCREENSHOTEXPAND ↗
Checks completed Pi replies for clarity and sends only the difficult ones to the configured model for a rewrite.
SOURCE SCREENSHOTEXPAND ↗
Search source files, documentation and logs by describing the idea you need rather than its exact wording.
SOURCE SCREENSHOTEXPAND ↗
Predicts which tests are safe to skip for a Git change, then runs the retained set through the existing framework.
SOURCE SCREENSHOTEXPAND ↗
Searches several web backends, then uses Jev to keep useful results, confirm duplicates and filter scraped page chunks.
SOURCE SCREENSHOTEXPAND ↗
from the source
// NoulQuestion builds a yes/no question.func NoulQuestion(instructions string) Question { return Question{Type: TypeNoul, Instructions: instructions}}
Prunes DeepSeek Harness tool results according to Jev judgments while retaining deterministic context receipts.
SOURCE SCREENSHOTEXPAND ↗
Exposes Jev’s three decision types as one MCP tool for coding-agent routing, risk checks, and bounded architectural choices.
SOURCE SCREENSHOTEXPAND ↗
Ailerix uses Jev to route each request to a model from a typed catalog.
SOURCE SCREENSHOTEXPAND ↗
A Codex plugin that uses Jev to restore verbatim context after session compaction.
SOURCE SCREENSHOTEXPAND ↗
from the source
npm installnpm run typecheck # librarynpm run check:hooks # node --check on every hook scriptnpm test # unit tests, fake Jev, no networknpm run build # emit dist/ (checked in: hooks import it)
A local proxy that uses Jev to choose the Claude model and reasoning effort for each message.
SOURCE SCREENSHOTEXPAND ↗
from the source
Claude Code ---> jcm-router (localhost:8787) ---> api.anthropic.com | +---> api.typesafe.ai (Jev: "which model? how much effort?")
An agent skill that sends closed coding decisions to Jev for judgment.
SOURCE SCREENSHOTEXPAND ↗
An MCP server that exposes Jev tools for classification, scoring, checks, and batched questions.
SOURCE SCREENSHOTEXPAND ↗
Put a Jev check between an agent and its MCP tools to screen for unsafe calls and results.
SOURCE SCREENSHOTEXPAND ↗
from the source
npx -y github:caiovicentino/jev-shield install-hooks # opt in: hook + plugin (merges...npx -y github:caiovicentino/jev-shield uninstall-hooks # opt out: removes hook + plugi...JEV_HOOK_OFF=1 # kill switch (env), JEV_FAIL_M...
A Claude Code hook that scores installed skills against the current task and hides irrelevant ones from the manifest.
SOURCE SCREENSHOTEXPAND ↗
An architecture skill for turning fuzzy semantic decisions into small Jev Choice, Score, or Noul questions.
SOURCE SCREENSHOTEXPAND ↗
from the source
# from this repocp -R skills/jev-system-architect ~/.cursor/skills/jev-system-architect# or wherever your agent loads skills from
A local workbench for building and publishing versioned Jev judgment functions.
SOURCE SCREENSHOTEXPAND ↗
from the source
pip install typesafe-sdkexport TYPESAFE_BASE_URL=http://127.0.0.1:17420export TYPESAFE_API_KEY=$JEV_CLIENT_TOKEN
An omp extension that uses Jev scores to reduce context while preserving selected passages verbatim.
SOURCE SCREENSHOTEXPAND ↗
from the source
grep "jev context" ~/.omp/logs/omp.$(date +%F).*.log | tail -3# jev context: 33906->27067 chars, dropped=5, asks=4, cacheHits=20
A Pi extension that checks side-effecting tool calls against the user's stated constraints before they run.
SOURCE SCREENSHOTEXPAND ↗
Skill Router uses Jev to match a Claude Code session's goal with relevant installed skills.
SOURCE SCREENSHOTEXPAND ↗
from the source
$ skill-router intent set "write the cold outreach sequence for our fintech prospects"[skill-router] Session goal: write the cold outreach sequence for our fintech prospectsJev: task kind `marketing_growth`, needs-a-skill 0.84.Relevant installed skills for this session: /cold-email (probability 1.00) — Write B2B cold emails and follow-up sequences that ge...
A model router that uses Jev and a YAML policy to choose the least expensive LLM suited to each request.
SOURCE SCREENSHOTEXPAND ↗
from the source
request ─► Jev: 11 questions, one call, ~180 ms ─► your YAML policy ─► model, fallback,... difficulty · stakes · needs_reasoning thresholds on probabilities; safety · domain · output_length · ... context, capability, and budget...
A Jev-powered task classifier that routes work across three tiers.
SOURCE SCREENSHOTEXPAND ↗
A CI reviewer that uses Jev to assess the safety of database migrations.
SOURCE SCREENSHOTEXPAND ↗
from the source
sequenceDiagram autonumber actor Dev as Developer / Git Push participant CI as GitHub Actions Pipeline participant MG as Migration Guard API (/api/migration-guard)
A Claude Code mod that ranks installed skills for each prompt and answers the agent's binary questions when confidence is high.
SOURCE SCREENSHOTEXPAND ↗
A LiteLLM router that asks Jev to pick a model for each request.
SOURCE SCREENSHOTEXPAND ↗
from the source
pip install "litellm[proxy]" pyyamlexport OPENROUTER_API_KEY=sk-or-...# optional — enables Jev routing instead of the baselineexport TYPESAFE_API_KEY=...litellm --config config.yaml --port 4000
A collection of Jev demos led by a chess match presented like a TV broadcast.
SOURCE SCREENSHOTEXPAND ↗
Describe the change you intended, review the matching Git hunks, and stage only the patches you approve.
SOURCE SCREENSHOTEXPAND ↗
Keeps coding agents away from graders, hidden tests and evaluation machinery while allowing ordinary development work.
SOURCE SCREENSHOTEXPAND ↗
Trims old DeepSeek Harness tool exchanges by keeping them whole, reducing them to excerpts or removing them.
SOURCE SCREENSHOTEXPAND ↗
Reviews pull-request changes against the practical refactoring rules in Five Lines of Code.
SOURCE SCREENSHOTEXPAND ↗
Reads batches of web pages and keeps the passages that best support a research question or planned claim.
SOURCE SCREENSHOTEXPAND ↗
Builds a smaller Codex handoff packet by selecting useful evidence verbatim instead of summarizing it.
SOURCE SCREENSHOTEXPAND ↗
Cuts long command output down before a coding agent reads it, while keeping selected lines exactly as written.
SOURCE SCREENSHOTEXPAND ↗
Temporarily hides older, low-value Pi messages from future requests without deleting them from the session.
SOURCE SCREENSHOTEXPAND ↗
Keeps OpenCode on a lightweight cache-friendly parent model until Jev decides a turn deserves a stronger child agent.
SOURCE SCREENSHOTEXPAND ↗
Five small gates help coding agents triage failures, stop doomed work, route models, verify steps and choose reasoning effort.
SOURCE SCREENSHOTEXPAND ↗
from the source
const questions = { category: { type: "choice" as const, instructions: "What is the root failure type in this error trace?", criteria: {
Uses typed Jev judgments to classify relationships between Hermes skills and plan conservative archive operations.
SOURCE SCREENSHOTEXPAND ↗
Routes an agent request to a small set of skills by asking Jev which descriptions match.
SOURCE SCREENSHOTEXPAND ↗
Uses Jev to select developer tools while leaving code generation and edits to Codex.
SOURCE SCREENSHOTEXPAND ↗
A Mnemosyne-compatible memory layer uses Jev to rank evidence and prefetch relevant memories for Hermes agents.
SOURCE SCREENSHOTEXPAND ↗
siftr gives coding agents Jev-ranked semantic search, focused file reads, and bounded list selection through a CLI and MCP server.
SOURCE SCREENSHOTEXPAND ↗
ActionGate reviews an agent's proposed tool call before allowing it, blocking it or asking for human approval.
SOURCE SCREENSHOTEXPAND ↗
from the source
providerResponse = await this.provider.evaluate( { state: buildMinimalState(effectiveRequest, tool), questions: buildSemanticBattery() }, { timeoutMs: this.options.timeoutMs ?? 2000 });
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 experimentSign in first so we can reply to you. Submissions are reviewed, not published automatically.