SHOWCASE

Showcase

Real public projects (repos / demos) that implement one or more Use Case patterns. Heat uses observed stars + awesome-list position — X mentions deferred when unknown. Not a visual redesign of the site.

Browse Use Cases Guides

222 public cases
Sorted by heat (high → low)
Heat ≠ ✦ awesome score on Use Cases
Cross-linked to Use Cases

Default sort: heat_score high → low. Heat is showcase-only; Use Cases keep editorial / category order and may show ✦ awesome score separately.

Computer use heat 5 · Hot · ★ 10900

showcase

jev-ultrafast

Jev picks operation + DOM target; small LLM only for TYPE_TEXT.

Agents heat 5 · Hot · ★ 4800

showcase

fast-jev-compaction

Claude Code plugin: Jev-gated context compaction before the context window blows up.

Routing heat 5 · Hot · ★ 1020

showcase

Shapeshift

One text box that morphs into the right UI as you type; TypeSafe Jev answers 14 typed questions in parallel, and deterministic code fills dates, amounts and math.

Agents heat 5 · Hot · ★ 40345

showcase

OpenHuman

Rust agent harness; README “Jev decides fast” — tool search among 1k+ tools + browser step decisions via System One proxy.

Realtime heat 5 · Hot · ★ 1400

showcase

jev-trader

One AI trade decision every Monad block — treat as dry-run demo, not financial advice.

Agents heat 5 · Hot · ★ 7165

showcase

Jev chat Jarvis

Android OCR chat co-pilot: Jev judges intent/risk/urgency; LLM drafts; human sends.

Computer use heat 5 · Hot · ★ 578

showcase

typesafe-computer-use

Mac computer-use: OCR/a11y → structured actions → Jev chooses → executor clicks.

Realtime heat 4 · Warm · ★ 294

showcase

typesafe-mario

TypeSafe/Jev agent that plays Super Mario — game/sim action tick demo.

Agents heat 4 · Warm · ★ 1935

showcase

jevgrep

CLI/skill: ask what code does; Jev judges folder/file/declaration relevance for coding agents.

Media heat 4 · Warm · ★ 142

showcase

Unclutter

Browser extension: Jev-powered page clutter removal with reusable template rules.

Guardrails heat 4 · Warm · ★ 392

showcase

jev-review

Staged code-review workflow and library with Jev decision checkpoints.

Guardrails heat 4 · Warm · ★ 101

showcase

pi-warden

Guardrails for Pi built on TypeSafe / Jev decision checks.

Guardrails heat 4 · Warm · ★ 421

showcase

foreman

Software factory foreman based on TypeSafe / Jev supervision loops.

Agents heat 4 · Warm · ★ 953

showcase

hermes-jev-skills

Jev skills for Hermes/Claude Code/Codex: routing, memory, compaction, skill select, computer/browser.

Routing heat 4 · Warm · ★ 236

showcase

pg-jev

Ask Postgres tables questions in plain language; Jev shapes the decision layer.

Computer use heat 4 · Warm · ★ 744

showcase

jev-browser-use

Codex skill: Jev clicks/navigates; Codex types and verifies. Distinct from jev-ultrafast / jev-browser.

Guardrails heat 4 · Warm · ★ 108

showcase

pi-jev

TypeSafe Jev as a decision layer for the Pi coding agent.

Routing heat 4 · Warm · ★ 498

showcase

jev-router (Claude Code)

Claude Code plugin that routes each task to the cheapest capable model via Jev.

Retrieval heat 4 · Warm · ★ 493

showcase

jev-search

Search1API + TypeSafe Jev for query understanding, source pick, and relevance ranking.

Agents heat 4 · Warm · ★ 463

showcase

Jev MCP

MCP server exposing TypeSafe Jev as typed agent tools (verify/screen/classify/rerank/gate).

Computer use heat 4 · Warm · ★ 425

showcase

mobile-jev

Mobile/Android agent where Jev chooses actions; live React studio + demo.

Agents heat 4 · Warm · ★ 337

showcase

system-one-connector

Go MCP/CLI exposing System One judgments (Jev and other decision models) to desktop agents.

Computer use heat 4 · Warm · ★ 297

showcase

jev-browser (jkudish)

Browser-use stack with TypeSafe Jev selecting actions (companion to jev-mcp).

Media heat 3 · Mild · ★ 44

showcase

vibecheck

Chrome extension: vibe-check your X drafts with a Jev judgment layer.

Agents heat 3 · Mild · ★ 153

showcase

jev-pruner

Claude Code plugin: Jev trims long Bash/tool output before the coding model sees it.

Guardrails heat 3 · Mild · ★ 108

showcase

jev-code-reviewer

Local PR reviewer: Jev prioritizes P0/P1/P2 attention; posts nothing to GitHub. Distinct from jev-review.

Agents heat 3 · Mild · ★ 29

showcase

winnow

Calibrated context sieve for Claude Code / agent tool results.

Platform heat 3 · Mild · ★ 11

showcase

Jev PowerShell module (dfinke)

A MIT PowerShell module that asks hosted TypeSafe Jev Choice, Noul, and Score questions and returns typed answers beside the input state for script-owned thresholds.

Guardrails heat 3 · Mild · ★ 49

showcase

jev-guard

Multi-agent security hooks: Jev scores risk / user_requested / from_untrusted on every tool call.

Media heat 3 · Mild · ★ 3

showcase

jev-skip

YouTube sponsor skipper that reads captions and lets Jev decide skip windows.

Routing heat 3 · Mild · ★ 15

showcase

JevMail

Open-source AI email triage for Gmail with a System One decision path.

Guardrails heat 3 · Mild · ★ 9

showcase

Jev Gatehouse

A Kinde starter that checks every agent tool call server-side: identity and permissions from Kinde, calibrated risk signals from Jev, then allow, ask the user, or stop.

Computer use heat 3 · Mild · ★ 8

showcase

jev-for-chrome

Jev for Chrome: drives the tab you already have open.

Agents heat 3 · Mild · ★ 184

showcase

Jev-Mem

Research code for an agent memory system where Jev makes the frequent memory decisions (memory type, relation links, query routing, search budget, candidate scoring, when to stop) and a separate language model only writes the final answer.

Computer use heat 3 · Mild · ★ 180

showcase

Jev Social

A local-first social research agent: TypeSafe Jev picks the next read-only Instagram/TikTok/LinkedIn operation; the socai CLI runs it in your signed-in Chrome and returns cited evidence.

Routing heat 3 · Mild · ★ 3

showcase

jevsql

SQL with natural-language predicates; Jev helps decide row/filter intent.

Guardrails heat 2 · Cool · ★ 1

showcase

typesafe-ai-firewall

Shadow-mode validation harness around model / tool outputs with Jev checks.

Agents heat 2 · Cool · ★ 100

showcase

jegrep

Rust semantic grep — describe what you want; Jev scores live files/ranges without an embedding index.

Ops heat 2 · Cool · ★ 4

showcase

jevmetrics

Experimental OTel Collector processor: Jev assesses metric metadata → annotate/filter retention.

Routing heat 2 · Cool · ★ 34

showcase

Orbs

Self-hosted multi-bot group chat: an @mention wakes that bot, and with no mention Jev scores which bot should reply; a daemon on your own machine runs each turn through the Pi agent CLI.

Routing heat 2 · Cool · ★ 1

showcase

jev-logtriage

Jev decides whether a batch of logs is noise, alert, or needs a human.

Agents heat 2 · Cool · ★ 28

showcase

skimmer

A Go CLI that splits web pages, files, grep hits or JSON lists into chunks and asks TypeSafe Jev one yes/no question per chunk, so an agent reads only the parts that help with its goal.

Guardrails heat 2 · Cool · ★ 24

showcase

jev-permission-gate

Claude Code mod: Jev answers 8 yes/no questions per auto-mode tool call, then the mod allows, denies, or defers to the built-in classifier.

Agents heat 2 · Cool · ★ 23

showcase

JevDo

Jev-first agent loop for DeepSeek Harness: Jev picks the next step from reusable, verified Actions; the main model is only one of the options.

Platform heat 2 · Cool · ★ 23

showcase

NetworkChuck Jev Guide

MIT companion repo for the NetworkChuck Jev video: Playground steps, curl/API calls, coding-agent skill install, and a Bun model router tested on jev-1.14.0.

Agents heat 2 · Cool · ★ 22

showcase

Jevspresso

XState + Jev simulated espresso bar: the state machine lists what is physically possible right now and Jev picks one event from that closed set. Live demo.

Routing heat 2 · Cool · ★ 1

showcase

n8n-nodes-jev

Community n8n node helper for calling Jev inside automation graphs.

Computer use heat 2 · Cool · ★ 18

showcase

Jevboard

Windows Zhuyin / Bopomofo IME correction PoC: read the IME's own candidates per character, ask Jev which positions are wrong, suggest fixes in a popup.

Agents heat 2 · Cool · ★ 14

showcase

jev-compaction-plus

Claude Code compaction plugin, fork of fast-jev-compaction: Jev keeps still-needed tool results word for word and moves the rest to a drawer file. Author says ~0.5 s vs ~35 s built-in.

Routing heat 2 · Cool · ★ 13

showcase

Jev x AI SDK Form Router

A Vercel Labs Next.js starter where three forms use Jev to route submissions by context, and gpt-6-luna-fast handles uncertain or failed evaluations.

Guardrails heat 1 · Quiet · ★ 11

showcase

Taste Lint

A commit-time design and copy linter: mechanical checks block without a key; optional Jev review via Vercel AI Gateway scores under-review UI/copy risks with probabilities.

Media heat 1 · Quiet · ★ 8

showcase

jevcraft

A mineflayer Minecraft bot that beat the Ender Dragon on fresh Survival worlds using only TypeSafe Jev for every judgement — code lists priced options, Jev picks one.

Platform heat 1 · Quiet · ★ 6

showcase

Jev DevTools

A React / Next.js inspector for TypeSafe Jev that captures systemOne calls, shows returned probabilities, and maps the paths your app takes.

Retrieval heat 1 · Quiet · ★ 6

showcase

jev-doc-search

Long-document page search via PageIndex tree + Jev Choice (optional Noul check); no vector DB.

Agents heat 1 · Quiet · ★ 5

showcase

Jauvex

An Electron desktop client that runs Claude Code, Codex and Grok Build side by side by voice, and uses TypeSafe Jev for the fast decisions when a key is present.

Guardrails heat 1 · Quiet · ★ 5

showcase

jev-engineering

Measured coding-agent tools on Jev (review-router + jev-gate) with public eval kits.

Agents heat 1 · Quiet · ★ 5

showcase

jev-job-search

Claude Code job-application filler: Jev makes every fixed-answer decision, Claude writes the sentences, code drives Chrome, and a human approves before anything is sent.

Realtime heat 1 · Quiet · ★ 5

showcase

Multi-UAV JEV

A MuJoCo urban airspace research platform where a local Jev-style decision model picks one of six flight actions from camera-derived text observations, with a live cockpit and episode archives.

Realtime heat 1 · Quiet · ★ 3

showcase

Jev Soundboard

Listens to a live call (FaceTime, Discord, Zoom), transcribes it every 1.2 s, and asks Jev two questions per slice: a Noul for whether a sound right now makes the moment funnier and a Choice for which clip fits; code plays a clip only above a trigger level.

Retrieval heat 1 · Quiet · ★ 3

showcase

Jevector

Profile matching that encodes each profile's criteria as 64 or 256 numeric decision values with Jev or Cloudflare Clef, retrieves candidates with HNSW, checks mandatory conditions in both directions in code, then ranks by preference coverage.

Routing heat 1 · Quiet · ★ 3

showcase

JevImpact

Python script that scores a vulnerability report with Jev: `is_vulnerability`, `impactful`, and a 0–1 `impact_score`; accepts structured JSON or a Markdown finding.

Ops heat 1 · Quiet · ★ 2

showcase

DeciFlow

Classifies log chunks as Normal, Warning, Error, Security or Other with self-hosted Laya or hosted Jev, and shows live category charts for human review.

Media heat 1 · Quiet · ★ 2

showcase

Jev Highlight Cutter

Agent skill plus Python CLI that cuts long interviews and podcasts into a highlight reel of a target length: Jev scores transcript candidates against your written taste, code picks non-overlapping segments within the budget, and local FFmpeg renders the rough cut.

Guardrails heat 1 · Quiet · ★ 2

showcase

jev-mcp-router

MCP gateway: Jev answers one yes/no per server and per tool, so the LLM only ever sees the schemas Jev said yes to.

Retrieval heat 1 · Quiet · ★ 2

showcase

Tweet Index

Local search over your own X likes and bookmarks: local embeddings shortlist 120 candidates, one Jev Score per candidate rates relevance, and results are ranked on relevance, substance and similarity.

Retrieval heat 1 · Quiet · ★ 2

showcase

arXiv Jev Digest

A GitHub Actions daily digester that screens new arXiv papers with hosted Jev (or a System One-compatible backend), summarizes with DeepSeek, and emails up to 15 matches.

Agents heat 1 · Quiet · ★ 1

showcase

Jev Harness

An open-goal agent harness where a primary model authors tools and checks, and Jev repeatedly chooses the next tool and arguments under code-enforced bounds and independent verification.

Media heat 1 · Quiet · ★ 1

showcase

Jev Slop Detector

Unofficial Chrome extension that labels posts on your X feed with one Jev Noul, 'does the post substitute unsupported hype for concrete substance?', through a local relay that holds the API key; a middle band shows no label.

Guardrails heat 1 · Quiet · ★ 1

showcase

JevTrace

Offline inspector for coding-agent traces that flags redundant reads, unjustified retries, loops and premature completion claims; rules mode runs locally, optional Jev mode adds semantic judgments.

Media heat 1 · Quiet · ★ 1

showcase

sanity-plugin-jev-fields

A Sanity Studio plugin that attaches Jev questions (Noul, Score, Choice) to content fields, shows each answer as a chip under the field, stores it next to the field and can warn or block publishing.

Media heat 1 · Quiet · ★ 1

showcase

Jev Ad Pattern Lab

A local Portuguese panel that captures active Meta Ad Library creatives and classifies each one with hosted TypeSafe Jev across seven ad dimensions plus audience.

Agents heat 1 · Quiet · ★ 1

showcase

jeval

A framework for decision-model harnesses: your domain offers short next-step menus, the model (including typesafe/jev-latest via pi-ai) picks one until it can answer typed questions, with full traces.

Platform heat 2 · Cool · ★ 0

showcase

@rbxts/jev

A typed Jev client for roblox-ts: game servers ask many questions about one state and get typed choices, scores and probabilities back, for example an NPC reading what a player wants.

Retrieval heat 1 · Quiet · ★ 0

showcase

Abstract Sift

A MIT local CLI that screens research abstracts against a research-interest text with OpenAI Decisions, TypeSafe JEV, or both, then exports CSV/JSONL and an offline HTML explorer.

Ops heat 1 · Quiet · ★ 0

showcase

Access Desk

Triages employee access requests: one Jev call answers typed questions, a code policy auto-approves only low-risk, well-justified, time-boxed requests and sends the rest to the system owner or manager with a recommendation.

Platform heat 1 · Quiet · ★ 0

showcase

aisdk/typesafe (PHP)

A MIT native TypeSafe Jev decisions provider for the framework-agnostic PHP AI SDK (aisdk/core): maps portable predicates to Noul and choices/scores to native criteria against jev-latest.

Media heat 1 · Quiet · ★ 0

showcase

AstrBot emoji like (Jev)

A MIT AstrBot continuation fork that can switch group emoji reactions from LLM JSON to hosted TypeSafe Jev Choice (or hybrid) over configured emotion labels, with jevmodel.org / api.typesafe.ai gateways.

Realtime heat 1 · Quiet · ★ 0

showcase

astrbot_plugin_delay_response

A MIT AstrBot plugin that waits on incomplete user bursts: a System One/Jev completeness score shortens or lengthens delay before merging consecutive messages into one reply (OneBot v11 / weixinoc).

Realtime heat 1 · Quiet · ★ 0

showcase

Audience Moderator (Jev)

An Android event helper that counts people and raised hands on-device, with optional hosted TypeSafe Jev round insight and experimental photo second opinions.

Media heat 1 · Quiet · ★ 0

showcase

BetterChat (AstrBot)

A MIT AstrBot plugin that debounces private-chat bursts into one reply and uses hosted TypeSafe Jev (Noul) probabilities—policy kept in code—to decide whether a group message deserves an unsolicited answer.

Agents heat 3 · Mild · ★ 142

showcase

Building with TypeSafe Jev

Unofficial agent skill (142★): teaches Claude Code/Codex/etc. to design typed Jev decisions and points at 150+ community project shapes.

Ops heat 2 · Cool · ★ 0

showcase

Calendar Curator

A Home Assistant custom integration that asks Jev whether each calendar event matches your criteria prompt and mirrors the matches into a target calendar.

Ops heat 1 · Quiet · ★ 0

showcase

Called It!

A MIT Claude Code / Codex skill that grades annual predictions at strict, standard and lenient bars, using hosted TypeSafe Jev for calibrated probability grades over committed evidence rows.

Platform heat 2 · Cool · ★ 0

showcase

cap-jev

A plugin for SAP CAP Node.js apps that adds an internal `system-one` service: choice, score and noul questions against TypeSafe Jev or any Jev-compatible System One endpoint, with typed errors and bounded retries.

Ops heat 1 · Quiet · ★ 0

showcase

cc-cockpit Jev triage

A MIT cc-cockpit plugin that triages mail/Jira/chat/calendar items into priority and kind via bronos.jev (hosted TypeSafe Jev choice grids), with project rules, human overrides, and an improve loop.

Ops heat 1 · Quiet · ★ 0

showcase

ci-triage

Reads a failed CI run, classifies the failure with Jev, re-runs flaky and infrastructure failures once, and posts one short PR comment for real failures.

Platform heat 1 · Quiet · ★ 0

showcase

classifier-bench

A benchmark harness that sends the same labelled requests to Jev, OpenAI Decisions, Cloudflare Clef and Claude models and reports match rate, latency tails, reliability, cost, calibration and cascade trade-offs.

Agents heat 1 · Quiet · ★ 0

showcase

Claude JEV auto-effort

A MIT Claude Code mod that picks per-request effort from plan step, prompt, and stage, using hosted TypeSafe Jev (TYPESAFE_API_KEY → api.typesafe.ai) as judge with scrubbed payloads and word-list fallback.

Media heat 1 · Quiet · ★ 0

showcase

d20 Oracle

A MIT one-page D&D 5e class oracle in Go that sends name, profession, and a crypto/rand d20 to hosted TypeSafe Jev as one Choice over twelve classes, then rolls a local d6 quirk.

Platform heat 1 · Quiet · ★ 0

showcase

Decision Wall

A MIT zero-dependency demo that fires ~200 yes/no TypeSafe Jev questions in one parallel call, paints a live heatmap, and lets code thresholds re-sort without re-asking; ships a no-key public app and measured width latency notes.

Agents heat 1 · Quiet · ★ 0

showcase

depgate

A GitHub Action that reads release notes and how your code uses a package, asks Jev six typed questions in one call, and then auto-merges, sends for review or holds each Dependabot or Renovate pull request.

Ops heat 2 · Cool · ★ 0

showcase

DmLab Catalog Enrich

A Magento 2 module that fills empty filter attributes (material, sleeve, seats, Yes/No) from product descriptions: Jev picks a value from the attribute's existing options, asked twice in opposite option order, with 'Not stated' always available.

Retrieval heat 1 · Quiet · ★ 2

showcase

Docision

Jev-inspired open multimodal decision model for documents: one image, many candidate-answer questions with per-option probabilities, plus Docision-1.0 weights and a published doc-decision-bench.

Guardrails heat 1 · Quiet · ★ 0

showcase

dsh-decision-guard

A MIT DeepSeek Harness plugin (package dsh-decision-guard) that uses hosted TypeSafe Jev for sensitive-command verdicts, per-session tool schema restriction, skill-catalog pruning, and glob relevance ranking—all off by default with fail-open native fallbacks.

Agents heat 1 · Quiet · ★ 0

showcase

dsh-jev

A DeepSeek Harness (DSH) plugin that exposes TypeSafe Jev as an injectable ctx.jev service plus a jev_ask tool, with keys locked to the plugin settings.

Agents heat 1 · Quiet · ★ 1

showcase

dsh-jev-mcp

A MIT, dependency-free Node client and stdio MCP server that exposes TypeSafe Jev as jev_check / jev_route / jev_score / jev_evaluate for DeepSeek Harness (DSH) native and code modes.

Routing heat 1 · Quiet · ★ 0

showcase

email-cleaner

A personal self-hosted Outlook/Gmail inbox cleaner that learns readable rules, classifies leftovers with TypeSafe Jev via the official SDK, and keeps client mail protected.

Guardrails heat 2 · Cool · ★ 0

showcase

English Net

A prompt-injection detector whose 7 neurons are English yes/no questions answered by Jev in one call, combined by logistic regression so every verdict names the questions behind it.

Media heat 2 · Cool · ★ 0

showcase

Facet

Native macOS/Windows photo culling app: custom yes/no labels and face search via TypeSafe Jev (or OpenAI vision), with live threshold preview.

Media heat 2 · Cool · ★ 0

showcase

FrameFind

A local Python app that finds moments in your videos from a plain-language query: FFmpeg samples every tenth frame, Jev checks each sampled frame against the query, and matches above a fixed cutoff become playable, exportable snippets.

Platform heat 1 · Quiet · ★ 0

showcase

HappyHackingSpace/jev

A MIT unofficial Go SDK, CLI, MCP server, and agent skill for TypeSafe's System One API (noul/choice/score/box), wired to TYPESAFE_API_KEY and api.typesafe.ai defaults.

Routing heat 1 · Quiet · ★ 0

showcase

hermes-jev-routing

A MIT Hermes Agent plugin that routes each turn with TypeSafe Jev via OpenRouter Alpha Decisions (~typesafe/jev-latest), mapping task kind, complexity, capability and deep-reasoning need onto a custom tier table.

Agents heat 2 · Cool · ★ 0

showcase

hermes-loopguard

A Hermes Agent plugin that stops cron and kanban automations from paying frontier models for unchanged state: a zero-token change gate, then a typed Jev gate that routes items to agent, kanban, review or drop.

Media heat 2 · Cool · ★ 0

showcase

Hotdog Not Hotdog (Laravel)

A Laravel 13 playground for Jev image input: one request asks a Noul question for HOTDOG / NOT HOTDOG / NOT SURE and a Box question to draw where the hot dog is.

Guardrails heat 2 · Cool · ★ 0

showcase

image-moderation-bench

A benchmark of Jev 1.14, Cloudflare clef-flash, Claude Haiku 5.5 and Amazon Rekognition on all 10,146 UnsafeBench images, with per-image outputs you can rescore.

Ops heat 2 · Cool · ★ 0

showcase

inbox-jev (litebox)

Gmail triage product: scheduled passes ask TypeSafe Jev to label, archive and auto-unsubscribe, with Docker, tests, categories.yaml and a plain-language ops guide.

Computer use heat 1 · Quiet · ★ 0

showcase

Jarvis

A MIT macOS voice/typed command tool that resolves skills with local patterns first, hosted TypeSafe Jev for natural phrasing, and an LLM only to fill details or write drafts—never to invent runnable code.

Media heat 1 · Quiet · ★ 0

showcase

Jev Ad Block

A Chrome Manifest V3 ad blocker that blocks ad requests with network rules and sends unclear page elements to TypeSafe Jev to decide whether they are ads, with local rules as the fallback.

Media heat 2 · Cool · ★ 0

showcase

Jev Cards

Visual quiz game against Jev's vision answers: each picture card is one typed choice or box question, with sprint, versus, streak and live-match modes.

Media heat 1 · Quiet · ★ 0

showcase

Jev Content Filter

A Firefox/Chrome MV3 extension that batches hosted TypeSafe Jev noul questions over classified-ad cards and hides or dims listings that do not match your free-text criteria.

Retrieval heat 1 · Quiet · ★ 0

showcase

Jev evidence triage

A MIT CLI that screens PubMed abstracts for one systematic-review question with hosted TypeSafe Jev (typesafe-sdk / jev-1.13.0), asking six narrow questions per abstract in a single call, with a mock path for offline rehearsal.

Platform heat 2 · Cool · ★ 0

showcase

Jev Field Test

An independent field test of jev-1.13.0 on 63 public datasets in 27 domains (33,817 requests), with stress tests and a static site that shows where Jev works as-is, needs a cut-off, or is not usable zero-shot.

Agents heat 1 · Quiet · ★ 0

showcase

Jev Gate Lite

A MIT Claude Code/Codex plugin that uses cloud Jev (no local Laya) for hybrid tool-call compaction, long tool-output pruning, and advisory MCP tools jev_evaluate / jev_apply_policy, falling back to built-in summarization when reduction is too small.

Ops heat 1 · Quiet · ★ 0

showcase

Jev Inbox Triage

Experiment that reads each customer email of a small online store once with eight parallel Jev questions (a topic Choice, urgency and mood Scores, five risk Nouls), then plain rules send it to Human now, Human queue, Automated or Archived.

Routing heat 1 · Quiet · ★ 0

showcase

Jev Inference Router

Observable decision layer for local GPU inference: Jev chooses the handler (local Qwen, document retrieval, or clarification), Python applies the policy, and local Qwen writes the answer.

Retrieval heat 2 · Cool · ★ 0

showcase

Jev job-matching experiment

An independent experiment where Jev scored 400 real Kaggle job postings against 30 fictional résumés (12,000 evaluations), with a top-3 check, a "none" cutoff and a frank list of seniority mistakes.

Platform heat 1 · Quiet · ★ 0

showcase

Jev Lab (Nijal AI)

A MIT FastAPI workshop that lets participants probe hosted TypeSafe Jev against a chat LLM on phone or PC—missions, duels, calibration tables—without writing code.

Retrieval heat 2 · Cool · ★ 0

showcase

Jev Literature Classifier

A local workbench that sorts research PDFs into a taxonomy you define: Jev picks one category per paper with a Choice question and keeps every category's probability, with an explicit 'none of the above' option, human review and optional Zotero import.

Guardrails heat 2 · Cool · ★ 0

showcase

Jev Moderation for Laravel

A Laravel package that classifies any Eloquent model with Jev and routes each item to passed, flagged for human review, or blocked using per-category probability thresholds.

Media heat 1 · Quiet · ★ 0

showcase

Jev Prompt Reversi

A MIT Reversi lab that feeds board features as state and a natural-language POLICY to hosted TypeSafe Jev (jev-1.14.0 via the TypeSafe SDK) to pick the next move, with a code clone player for comparison.

Retrieval heat 2 · Cool · ★ 0

showcase

Jev RAG Chinese Benchmark

A reproducible Chinese reranking benchmark for Jev on C-MTEB/T2Reranking with hard negatives: pinned model ids, cost caps, full per-call logs, and a 2026-10-19 spot check of Jev 1.14 against 1.13.

Ops heat 1 · Quiet · ★ 0

showcase

JEV SOC Alert Triage

A MIT Python pipeline that classifies exported Windows security events with TypeSafe Jev (plus a rules baseline that runs without credentials) and routes uncertain or high-risk rows into a review queue with JSONL decision trails.

Guardrails heat 1 · Quiet · ★ 0

showcase

Jev ToS Checker

A MIT Chrome extension that splits an on-page terms-of-service into clauses and asks hosted TypeSafe Jev whether each clause looks abusive across eight categories, highlighting matches at ≥70% probability.

Computer use heat 1 · Quiet · ★ 0

showcase

Jev Voice

Talk to your own Chromium profile: Claude plans the goals, Jev picks one page action at a time, and a side-panel extension clicks without chrome.debugger.

Ops heat 1 · Quiet · ★ 0

showcase

JEV WiFi Analyzer

2.4 GHz WiFi site-survey device (ZimaBoard 2 + Alfa adapter): Python and Scapy turn captured 802.11 frames into per-channel telemetry, and Jev scores coverage and picks a congestion level and an optimization action in one request.

Guardrails heat 1 · Quiet · ★ 21

showcase

jev_antispam_bot

A MIT grammY Telegram anti-spam bot powered by hosted TypeSafe Jev, with Bun tooling, optional Postgres stats, and admin delete-message requirements documented in the README.

Platform heat 1 · Quiet · ★ 0

showcase

JEV_Langflow

A MIT Langflow component pack for TypeSafe Jev and local KEV System 1: visual tables for Choice/Noul/Score, parallel evaluation, and confidence gating inside Langflow canvases.

Routing heat 1 · Quiet · ★ 0

showcase

jev-agent-router

Benchmark of three support-message routers behind one typed interface: keywords, Claude Haiku, and Jev. Author reports 98% vs 99% accuracy on 111 hard messages, with Jev about 5x faster at the median.

Realtime heat 2 · Cool · ★ 5

showcase

jev-box-tracking

Realtime object tracking that fuses slow Jev Box answers with per-frame optical flow in a Kalman filter, rewinding history when a late answer arrives.

Platform heat 1 · Quiet · ★ 0

showcase

jev-calibration-lab

A MIT independent calibration harness that compares TypeSafe Jev probabilities to human-labeled public datasets versus Claude and a TF-IDF baseline, with a GitHub Pages leaderboard; Jev/Claude numbers in-repo are still synthetic dry-run until keys run.

Routing heat 2 · Cool · ★ 0

showcase

jev-cascade

Stdlib Python kit: Jev answers first, only the uncertain band goes to a stronger model, and a per-stage report checks that stage two really is stronger.

Agents heat 2 · Cool · ★ 1

showcase

jev-claud

Python library + Claude plugin: keep free-form text for people, route Choice/Score/Noul decisions to TypeSafe Jev (PyPI, CI, CLI).

Computer use heat 2 · Cool · ★ 0

showcase

jev-click-bench

Jev 1.14's box tool on ScreenSpot-v2 and ScreenSpot-Pro with official scoring and every click logged: 84.36% and 18.66% raw, 42.76% on Pro with a 2x zoomed second look.

Guardrails heat 1 · Quiet · ★ 0

showcase

jev-content-mod

A content-moderation demo that asks TypeSafe Jev for harm Score, category Choice and clarity/human-review Nouls, then applies asymmetric thresholds so removal needs higher confidence than approval.

Platform heat 1 · Quiet · ★ 0

showcase

jev-daemon

An Apache-2.0 Rust per-user daemon (jevd) plus jev CLI and tray that holds TypeSafe keys, pairs local programs with explicit jev allow, and shape/rate/secret-checks each request before forwarding to hosted Jev (Claude Code plugin included).

Guardrails heat 1 · Quiet · ★ 0

showcase

jev-decision-harness

A Claude Code plugin and CLI that adds typed TypeSafe Jev decisions at prompt, shell-command, tool-result and stop hooks, while deterministic code keeps permissions and Jev never grants an allow.

Routing heat 2 · Cool · ★ 5

showcase

jev-demo

Three real TypeSafe Jev teaching demos (Jira triage, pre-flight function scan, PR model router) where typed Choice/Noul/Score answers drive code thresholds and Claude only explains what Jev flags.

Platform heat 2 · Cool · ★ 6

showcase

jev-docs-benchmarks

An independent study repo that benchmarks TypeSafe Jev against Microsoft Decision, OpenAI Decisions, Clef, Laya and other decision models on rule-following, probability behaviour and cost, with English and PT-BR suites.

Ops heat 1 · Quiet · ★ 0

showcase

jev-eval

Command-line tool that runs Jev over your own labeled JSONL data and reports per-question accuracy, calibration (ECE and a reliability table) and risk-coverage, then recommends the confidence cut-point to gate on.

Agents heat 2 · Cool · ★ 4

showcase

jev-gc

Context garbage collection for LLM agents driven by OpenTelemetry spans: plain rules settle most spans, and Jev judges only the ambiguous ones to keep, compress to a pointer, or archive, with nothing hard-deleted.

Platform heat 2 · Cool · ★ 0

showcase

jev-gibbs

Replication of Large Language Gibbs (arXiv 2606.19264) with Jev as the per-claim conditional on TruthfulQA, plus controls showing most gain comes from context, not label consistency.

Agents heat 3 · Mild · ★ 83

showcase

jev-kit

MIT kit that wires TypeSafe Jev into Claude Code: tool-call/tier guards, file search, browser agent, review, belay, compaction and installers behind one TYPESAFE_API_KEY.

Retrieval heat 2 · Cool · ★ 0

showcase

jev-lab (feed sieve)

Two stdlib-only Jev experiments: a feed sieve that judges a week of posts from 500+ RSS, Atom and YouTube feeds against a reader profile and writes a weekly reading list, and a bookmark re-filing test with agreement, calibration and cost.

Agents heat 2 · Cool · ★ 7

showcase

jev-layer

Portable System-1 decision layer for agent harnesses (Hermes, OMP, Codex, generic MCP): Jev picks from a host-supplied set while the host keeps permissions and execution, with receipts and offline replay.

Guardrails heat 1 · Quiet · ★ 3

showcase

jev-pii-checker

A MIT CLI that finds PII with hosted TypeSafe Jev category Nouls and sensitivity scores, then regex/segmentation for spans, with an explicit warning that scanned text leaves the machine.

Agents heat 2 · Cool · ★ 0

showcase

jev-plan-review

A Claude Code plugin that checks a plan against what you said while it was prepared: Jev sorts your statements into requirements, constraints, decisions and so on, then answers two yes/no questions per item: does the plan reflect it, and does it conflict with it.

Retrieval heat 2 · Cool · ★ 0

showcase

Jev-Reranker

A head-to-head test of Jev and BGE as RAG re-rankers on 300 multi-hop HotpotQA questions, with the same search, generator and questions for every variant.

Routing heat 1 · Quiet · ★ 0

showcase

jev-senselab-demos

Four business-decision demos where TypeSafe Jev chooses actions and SenseLab runs a self-improvement loop around it, with committed sample reports that rebuild without API keys.

Ops heat 3 · Mild · ★ 663

showcase

jev-seo

A live SEO audit from one homepage URL: it crawls the site, checks 52 rules tied to Google Search Central, asks Jev typed questions about every page, and ranks fixes into PDF, XLSX and Markdown reports.

Guardrails heat 2 · Cool · ★ 0

showcase

jev-threshold-lab

Fit act / review / escalate thresholds for each Jev question on your own labelled data, with a calibration readout, a debugging ladder and an offline demo.

Ops heat 2 · Cool · ★ 1

showcase

jev-triage

Spring Boot benchmark of Java rules, Jev and Claude Haiku 5.5 on warehouse goods-receipt remarks: category, block-from-putaway and severity.

Realtime heat 2 · Cool · ★ 0

showcase

Jevaga

A Galaga-style browser shooter where Jev picks the ship's move up to four times a second, and NVIDIA Nemotron writes one new rule after each death.

Guardrails heat 1 · Quiet · ★ 0

showcase

jevban

A MIT Free Pascal fail2ban-style log watcher that hard-rules clear probes, then asks a decision model—default in-process OpenJev, or hosted TypeSafe Jev API—whether ambiguous clients are attacking before banning them.

Ops heat 1 · Quiet · ★ 0

showcase

jevcells

A Google Sheets add-on whose formulas ask TypeSafe Jev per row: rank, fuzzy match, multi-tag, pick-from-candidates, plus yes/no, choice and score.

Platform heat 1 · Quiet · ★ 0

showcase

Jevelin

A MIT community C# client for hosted TypeSafe Jev that sends state plus named Choice / Score / Noul questions and returns typed probabilities for .NET callers.

Agents heat 1 · Quiet · ★ 0

showcase

jevellan-harness-skills

A MIT Claude Code skill pack that turns a session into a coordinator: TypeSafe Jev places each step's model/effort/provider, workers run via claude -p or codex exec, and an append-only ledger keeps the story recoverable.

Agents heat 1 · Quiet · ★ 0

showcase

Jeverb

A MIT Muse Code / MCP layer that wires hosted TypeSafe Jev bounded judgments into hooks and tools while keeping admission, graph execution, and closure behind deterministic policy and verifier receipts (experimental; JevWire-compatible npm identity).

Retrieval heat 3 · Mild · ★ 50

showcase

jevfetch

A MIT web-reading CLI/agent tool that splits a page into blocks and asks hosted TypeSafe Jev which blocks answer a question, returning verbatim passages with heading paths instead of a paraphrase.

Media heat 1 · Quiet · ★ 0

showcase

JevGAG

A research repo that uses hosted Jev 1.14 as an off-device oracle for adaptive temporal granularity in causal video generation, with audited claim-to-artifact results.

Ops heat 2 · Cool · ★ 6

showcase

jevgate

A Rust CLI that reads stdin, asks TypeSafe Jev a yes/no question, and either greps the text through or runs a command after -- when P(yes) clears a threshold.

Platform heat 1 · Quiet · ★ 0

showcase

Jeview

A MIT zero-dependency local gateway/visualizer that sits between your code and api.typesafe.ai, proxies /v1/systemone calls with your TypeSafe key, stores them in SQLite, and draws a live call map (plus Pixel Knight and support-inbox demos).

Guardrails heat 1 · Quiet · ★ 0

showcase

Jevify

A MIT skill pack for Claude Code / Cursor / Codex that teaches agents to refactor chat endpoints into a three-tier fast path (regex → LRU → System One Jev) plus quota exemption, with templates and an author case write-up.

Platform heat 1 · Quiet · ★ 0

showcase

Jevlar

An Apache-2.0 self-hosted Go gateway for System One decision-model APIs: one TOML file, client auth, multi-provider key failover, retries, and Prometheus metrics without a database or UI.

Ops heat 1 · Quiet · ★ 0

showcase

Jevline

Incident-scoping PoC: start from one confirmed-malicious process and Jev asks, process by process, whether each other execution belongs to the same incident; linked ones form a reviewable timeline. Browser demo.

Ops heat 1 · Quiet · ★ 0

showcase

jevlint

A MIT model-based linter (npm @indxjp/jevlint) that combines AST metrics with hosted TypeSafe Jev (or other decision providers) to rank which files to refactor first, explain findings, and hand prompts to Claude Code/Codex—including PR/commit delta reviews.

Media heat 2 · Cool · ★ 7

showcase

jevman

An AGPL-3.0 Pac-Man lab where characters are driven by System One models—default hosted TypeSafe jev (typesafe/jev-1.13.0) via Opper or direct TypeSafe—with a public demo at jevman.chadda.se and on-board odds overlays.

Media heat 1 · Quiet · ★ 0

showcase

jevmoji

A MIT HTMX demo that scores a 200-emoji catalog with hosted TypeSafe Jev Noul in one POST /v1/systemone call (jev-latest / jev-1.14.0) and shows Slack-like reactions above a confidence gate.

Retrieval heat 1 · Quiet · ★ 0

showcase

JevNER

A MIT Python NER library that uses hosted TypeSafe Jev (TYPESAFE_API_KEY, Noul detection + Choice typing over sliding windows) to return nested EntityNode forests with optional gender and domain packs.

Ops heat 1 · Quiet · ★ 0

showcase

JevOps

A control plane that sends each proposed AI-agent action to TypeSafe Jev for typed semantic judgment, then applies deterministic policy rules to return ALLOW, RETRY, HUMAN_REVIEW or BLOCK.

Retrieval heat 1 · Quiet · ★ 0

showcase

JevQuery

A TypeScript library that turns a sentence into a typed query plan by asking Jev to pick schema parts; Drizzle, Prisma or a SQL builder then runs validated identifiers and bound values.

Media heat 3 · Mild · ★ 50

showcase

jevreel

A MIT MCP server/skill that lays video onto labelled contact sheets and asks hosted TypeSafe Jev which windows match a query, then re-checks top hits on full frames—no embeddings stored.

Media heat 1 · Quiet · ★ 0

showcase

JevSense

A MIT personalization reference that keeps browse history in the browser, sends only allowed page/action signals to a Node or Cloudflare Workers API, and asks hosted Jev (Bearer JEV_API_KEY, default jev-1.14.0) which registered offers to show.

Platform heat 1 · Quiet · ★ 0

showcase

JevVsLLM

A classification benchmark harness that compares TypeSafe Jev with low-cost LLMs and other OpenRouter decision models on five public datasets × 2,000 examples, reporting accuracy, cost and latency.

Realtime heat 1 · Quiet · ★ 0

showcase

Keeping an NPC on quest

Companion code for an Old Nain YouTube episode: a terminal text game where a local model voices an NPC, TypeSafe Jev classifies each player message, and four Jev Nouls check every drafted line for spoilers, invented facts and broken character.

Agents heat 1 · Quiet · ★ 0

showcase

MaiBot-JevPreprocessor

A GPL-3.0 MaiBot SDK2 plugin that calls hosted TypeSafe Jev in maisaka.planner.before_request to drop irrelevant heuristic memories and mark external facts before the Planner runs.

Media heat 2 · Cool · ★ 0

showcase

Nature Lens

Unofficial demo: upload a nature photo; Jev returns per-category Nouls, a main-subject Choice, person Noul, and optional Jev 1.14 Box.

Media heat 1 · Quiet · ★ 17

showcase

netizen

A MIT spike that simulates Indonesian netizen reactions to a draft post with hosted TypeSafe Jev persona/interest judgments, anchors against labeled real cases, and ships a BYO-key web demo.

Agents heat 1 · Quiet · ★ 0

showcase

next-steps-jev

A MIT Claude Code terminal plugin where the session model proposes up to three next prompts and hosted TypeSafe Jev scores each against your goal so the best suggestion is listed first; never auto-submits.

Platform heat 1 · Quiet · ★ 0

showcase

omadia decision-jev

A MIT omadia plugin that installs hosted TypeSafe Jev (default jev-1.14.0) as decisionModel@1 so each chat message is gated for LLM need, data-source choice, and learned direct plans before any language-model call.

Media heat 2 · Cool · ★ 10

showcase

Onion or Not?

Type a headline and see hosted Jev, AnyJev L0 on local Qwen3-8B and raw Qwen3-8B judge satire vs real news side by side, each with a confidence and an optional swapped-order check.

Realtime heat 1 · Quiet · ★ 0

showcase

open-annie

Open-source browser 3D character: GPT-Live-1 does the talking over WebRTC while Jev reads the live transcript and picks her face, gestures and body language, with every answer passing a confidence gate.

Agents heat 1 · Quiet · ★ 0

showcase

opencode-auto-question

An OpenCode plugin that auto-answers the question tool for autonomous agents using Jev Choice/Noul (Zen or OpenRouter) with a confidence-gated two-pass context gather.

Agents heat 1 · Quiet · ★ 0

showcase

opencode-jevcode

A MIT OpenCode v2 plugin that adds a jevCode primary agent: hosted TypeSafe Jev chooses coding model, skills, tool groups, and relevant history while OpenCode's LLM writes code; fail-open if Jev is down.

Agents heat 3 · Mild · ★ 50

showcase

optim-jev

A MIT Claude Code (and partial Codex) plugin pack that uses hosted TypeSafe Jev for context compaction, per-subagent model routing, and skill-catalog pruning, with fail-open defaults.

Guardrails heat 1 · Quiet · ★ 0

showcase

PasteGuard

A MIT Windows tray agent that watches the clipboard for command-like text, scores danger with hosted TypeSafe Jev, and clears malicious ClickFix payloads before you can paste them into Win+R or a shell.

Routing heat 2 · Cool · ★ 0

showcase

pgdocket

Durable, versioned, calibrated AI decisions for Postgres: builds on pg-jev to store answers as data, calibrate them on your labels, and give thresholds and counts with stated guarantees.

Media heat 2 · Cool · ★ 2

showcase

Photo Ferret

A photo gallery you browse with sliders: Jev's image input rates every photo on dozens of scales (sky, water, snow, sky drama, wildness, scale, quality), and a static site lets visitors filter by those ratings.

Agents heat 1 · Quiet · ★ 0

showcase

pi-jev-ext-router

A pi coding-agent extension that asks TypeSafe Jev (via OpenRouter) which lazily loaded extension a prompt needs, then activates it before the model turn so the agent skips search/activate round trips.

Agents heat 1 · Quiet · ★ 0

showcase

plan-anchor

A MIT Claude Code/Codex skill that captures planning directives as numbered anchors, tunes typed Jev questions, and gates both the plan and delivered work against your own quoted words via hosted TypeSafe Jev.

Platform heat 1 · Quiet · ★ 3

showcase

ProbabilityBench

MIT benchmark of synthetic worlds with exact class probabilities so reported scores from Jev and other decision models can be compared to ground-truth posteriors (arXiv 2610.17016).

Computer use heat 2 · Cool · ★ 0

showcase

quick-qa-helper

Hands an Android UI QA task to a fast loop: Claude Haiku plans each step, TypeSafe Jev picks the element (Box on the screenshot when the UI tree has none), and an independent Jev verifier must agree before a run passes.

Retrieval heat 1 · Quiet · ★ 0

showcase

Recall

A MIT Chrome extension that indexes browsing history locally for fast search and optionally reranks a shortlist with hosted TypeSafe Jev after opt-in, with a no-key synthetic demo HTML.

Realtime heat 1 · Quiet · ★ 0

showcase

ROOMMATE.EXE

Seven-day pixel life sim where you share an apartment with a Jev roommate: the engine offers 6–8 legal, reachable actions and Jev picks one with a Choice question; pathfinding and consequences stay in code.

Retrieval heat 1 · Quiet · ★ 0

showcase

seo-jev-analyzer

A web app and MCP server that scores how well each page matches a target search intent, using TypeSafe Jev Score questions for term ratings and intent fit, with a measured noise floor.

Computer use heat 2 · Cool · ★ 0

showcase

siteeye

Construction CCTV pipeline: YOLO summarizes recent motion, TypeSafe Jev 1.14 returns near-miss Noul, risk Score and action Choice, with optional image/Box fallback.

Guardrails heat 1 · Quiet · ★ 0

showcase

Social Risk · Qwen3-VL + Jev

Research project on Chinese social-media image-and-text risk detection: a local Qwen3-VL LoRA writes structured evidence, Jev makes a fixed-policy decision into low / medium / high risk bands for human review, and the README reports where Jev did not help.

Media heat 1 · Quiet · ★ 0

showcase

Spoiler Danmaku Filter (JEV)

Zero-dependency Chrome extension and CLI that asks Jev to rate every Bilibili danmaku comment on a five-level spoiler scale and drops the spoilers before the player renders them.

Platform heat 3 · Mild · ★ 55

showcase

spring-ai-typesafe

Spring AI Community Java SDK and starters for TypeSafe System One: TypeSafeClient, JevJudge, guardrail/self-refine advisors, RAG document filter/reranker, tool index, and runnable examples.

Guardrails heat 1 · Quiet · ★ 0

showcase

Strayguard

A git pre-commit hook that asks TypeSafe Jev whether each new file belongs in the repo, and blocks likely scratch scripts, dumps, build output and local-only files.

Media heat 1 · Quiet · ★ 0

showcase

Sunagaku Brand Check

A MIT local app that keeps human brand criteria separate from two independent evaluators—OpenAI Decisions and hosted TypeSafe Jev (TYPESAFE_API_KEY)—and reproduces synthetic comparison runs into reports/latest.json.

Routing heat 1 · Quiet · ★ 0

showcase

Support Ticket Router

Service that asks Jev three questions per support ticket (team, urgency, churn risk) in one call, then multiplies the chance of each mistake by its cost to decide between auto-assign, quick human review and full human handling.

Media heat 1 · Quiet · ★ 0

showcase

swaraga

Classifies Hindustani and Carnatic ragas from vocal audio: signal processing extracts pitch telemetry and shortlists ragas, then ten batched Jev questions pick the raga and its parent scale family, combined as a hierarchical beam score.

Agents heat 2 · Cool · ★ 0

showcase

Switchbite

A Rust library and native MCP server that recommends a target from your worker or model pool; the online judge is Jev 1.14, and the host keeps authorization and execution.

Platform heat 2 · Cool · ★ 0

showcase

System One for Delphi

A single-unit, RTL-only Delphi client for System One models: Choice, Score and Noul questions against TypeSafe Jev or Ollama, plus an adapter that puts the same questions to any OpenAI-compatible chat model.

Platform heat 2 · Cool · ★ 2

showcase

system-one (desplega.ai)

A Bun workspace with `@system-one/ai`, one typed API over TypeSafe, OpenRouter, Vercel AI Gateway, OpenAI and Cloudflare decision models, and an arcade where decision models play games against each other and get ranked.

Ops heat 1 · Quiet · ★ 2

showcase

system-one-reviewer

A MIT local pull-request reviewer that keeps plumbing in deterministic Python and every judgment on a System One model (hosted TypeSafe Jev or local Laya), with a built-in precision/recall evaluation harness.

Platform heat 1 · Quiet · ★ 0

showcase

systemone-mcp

A MIT MCP server that proxies POST /v1/systemone typed decisions (Choice/Score/Noul/Box) to any System One provider—including TypeSafe Jev and Bailian—with loose usage.output_tokens normalization so providers that omit the field still work.

Realtime heat 2 · Cool · ★ 0

showcase

THE SWARM

WebGL demo of 1,000 agents whose strategy is chosen live by TypeSafe Jev (Choice + Score + Noul) about every 1.5 seconds.

Guardrails heat 1 · Quiet · ★ 29

showcase

toolgate

A MIT measured tool-call firewall for AI agents (Claude Code mod/hook, OpenCode plugin, MCP proxy) that asks hosted TypeSafe Jev seven policy questions before allow/ask/deny, with published evaluation notes.

Agents heat 1 · Quiet · ★ 0

showcase

Topic Gate

A MIT agent skill that asks hosted TypeSafe Jev whether a turn is same / subtopic / new_topic, then resets the session (Hermes /clear or Claude Code /new) only when drift is clear, saving a handoff note instead of dragging dead context.

Platform heat 1 · Quiet · ★ 3

showcase

train-jev

Open training stack that reproduces JEV-style typed decisions (Noul/Choice/Score) on Qwen with LoRA + a 24-slot head, KL distillation from TypeSafe Jev 1.13, calibration, and a System One-compatible /v1/systemone server.

Routing heat 2 · Cool · ★ 0

showcase

tsjev for DuckDB

DuckDB extension that turns Jev decisions into SQL functions (tsjev_label, tsjev_classify, tsjev_if, tsjev_score, tsjev_tags, tsjev_ask) with row dedupe, memory and disk cache, and 32 parallel requests.

Routing heat 2 · Cool · ★ 0

showcase

TypeSafe (Jev) × Salesforce

Salesforce DX project that calls Jev from Apex for Service Cloud case triage and screens guest messages to an Agentforce agent, with thresholds in custom metadata.

Platform heat 2 · Cool · ★ 0

showcase

typesafe_rs

Async Rust SDK for TypeSafe Jev: Tokio + reqwest + Serde client with Choice/Score/Noul, images, and experimental Jev 1.14 Box (v0.2.0).

Guardrails heat 2 · Cool · ★ 0

showcase

typesafe-gates

A TypeScript library of decision gates for LLM pipelines: each gate asks Jev typed questions about one item, and your code owns the threshold, the keep rule and an explicit fail-open or fail-closed policy.

Retrieval heat 1 · Quiet · ★ 4

showcase

Vibefilter for Filament

A MIT Filament PHP package that turns a plain-English statement into a zero-shot TypeSafe Jev yes/no score per table row and keeps rows above a threshold, with a public demo.

Ops heat 1 · Quiet · ★ 0

showcase

Vigil

An e-commerce fraud scoring service that sends each masked checkout to TypeSafe Jev with four typed questions (fraud probability, pattern, risk level, action) and turns the answers into allow, step-up, review or block.

Computer use heat 1 · Quiet · ★ 0

showcase

webhand

A MIT Claude Code browser agent plugin: Claude plans goals/steps while hosted TypeSafe Jev picks each click and entry from learned service runbooks (or from the live page when unlearned).

Realtime heat 1 · Quiet · ★ 0

showcase

wsy-jev-uav-track

A MIT UAV tracking prototype that rolls out semantic behaviors, applies hard safety filters, then asks hosted TypeSafe Jev Choice (api.typesafe.ai/v1/systemone) which next behavior to take.

Media heat 2 · Cool · ★ 2

showcase

YapSketch

Mac app: record your screen, talk and point; Jev splits your words into points, matches each to what you showed and picks the clearest frame, then puts numbered text and pictures on the clipboard.

Media heat 1 · Quiet · ★ 0

showcase

YouTube Jev Filter

A MIT Chrome extension that hides YouTube clickbait, hype, and Shorts by asking hosted TypeSafe Jev three questions per card (clickbait Noul, hype Noul, substance Score 0–3) and applying code-owned soft/normal/strict thresholds.

Media heat 1 · Quiet · ★ 0

showcase

YouTube Sponsor Skip (Jev)

A Chrome and Safari extension that finds spoken sponsor segments in Russian YouTube captions with TypeSafe Jev and skips them, with an undo control and a daily spend cap.