All Jev builds
Every project in the directory by name: 756 builds in 21 use cases, in alphabetical order inside each. One line says what a build does, and its page has the source and the figures its author published.
For the same builds as cards, newest work first, go to the feed. For the repositories alone, see Jev GitHub repos.
Jev computer use and browser agents
Agents that pick the next action: a link, a tool call, a click. 77 builds
- 18K stars in a weekjev-ultrafast splits every browser decision into operation, target and value.
- 200× faster and 400× cheaper than a plain loopLet Claude reason, let Jev decide, split the two.
- A computer assistant that listensLocal Whisper hears everything; Jev decides which action to take, in real time.
- A Jev quick-start checklistFrom the API to jev-browser.
- A study group week on JevHow it differs, where it fits, how to put it in a harness.
- Agent DesktopDesktop control from the accessibility tree, with Jev picking the operation.
- Agentic browsing in ChromeA Chrome extension where agents browse and click for you, with Jev deciding.
- An agentic pipeline with Jev at the coreAn architecture where Jev makes the pipeline decisions.
- AskJevTalk to Claude, and Jev moves your real browser on any site.
- AugustusAn agent skill for designing systems around Jev’s judgments.
- Auto-filling external link infoJev fills in the details for a pasted link.
- Browser control on a cloud machineThe popular browser scenario, run on a cloud box for AI employees.
- browser-aiA Jev browser agent in pure Go, attached to a real Chrome.
- Computer use without screenshotsOn-device segmentation and OCR turn the screen into text for Jev.
- CortexA local MCP server that cuts computer use from 23s to 0.8s.
- dsh-jev-toolsJudgment, not generation, inside DeepSeek Harness.
- dwimType what you want, and the right menu item runs in any Mac app.
- ego-jevOne ~0.4 s typed decision per DOM step, instead of an LLM turn.
- fastbrowseJev picks each action; an LLM plans; every claim cites the page.
- GPTBots.ai integrates JevA two-layer AI with Jev on the decisions.
- Headless Chromium agentJev drives headless Chromium so agents can browse quickly.
- InterlockA gate on every agent tool call, with Jev as one of its sensors.
- Jev agent with top-k logits proposalsA research design for a Jev-native agent system.
- Jev books an Uber by itselfAn Android phone controlled by Jev, from app open to payment.
- Jev BrowserJev chooses the control, Playwright acts.
- Jev Browser UseA Codex skill: Jev clicks, Codex thinks and checks the result.
- Jev decides, Grok actsA cleaner agent architecture.
- Jev driving real Chrome from Grok BotBots use Jev for the clicks instead of a slow look-and-click loop.
- Jev Engineering rewrites the agent graphThe agents create the paths; Jev decides which survive.
- Jev inside a harness, from a study groupNotes from a gakuse.ai session on mechanisms and overseas cases.
- Jev is the semantic if in your agent loopThe LLM writes, Jev answers typed questions, your code branches.
- Jev on the WebMCP benchmarkJev picks the tool, a fast LLM fills the arguments: 49 of 49 tasks solved.
- jev-browserBrowser automation where an LLM plans and Jev decides.
- jev-browser (Playwright)One Jev and Playwright core behind a typed SDK, a CLI and an MCP server.
- jev-browser in ClineA marketplace plugin that gives Jev a browser inside Cline.
- jev-browser, a general browser skillGive it a site and a task, and Jev decides every click.
- Jev-cuA computer-use repository built around Jev.
- jev-cuaA local computer-use fast path for Codex and Waku.
- jev-desktopJev picks the action inside Codex Computer Use.
- jev-harnessAn LLM proposes, Jev answers, code decides, every step leaves a receipt.
- jev-mcp for coding loopsAn MCP server that puts Jev on the coding loop in Cursor and Codex.
- jev-mobileAndroid control loops with Jev and Mobile MCP.
- jev-raBrowser use for coding agents, with Jev deciding each step in ~300 ms.
- jev-reflexClaude thinks, Jev reacts: browser tasks at about 100 ms a decision.
- jev-superpowersA skills framework for coding agents, with Jev gates and vetting.
- jev-ultrafastBrowser Use’s browser agent, with Jev choosing each next step.
- jev-useSay it, and your Mac does it, with no vision model.
- jev-use (voice)Voice control for macOS where Jev is the fallback, not the router.
- jev-voice-browserA browser you control by voice, from Moritz Kremb’s Jev tutorial.
- JevHarnessLLM-written Jev harnesses, with reward reflection over the full trajectory.
- JevLoopAn agent loop where the decisions cost no LLM call.
- jevwireA decision layer for agents: MCP server, library and Claude Code plugin.
- Launching agents on a canvasOrchestration without waiting for a slow LLM loop.
- Learn Jev end to endA free course: 13 agent use cases with a fast brain and a slow brain.
- macos-computer-use-kitAX-first computer use on macOS with optional Jev semantic guards.
- Mobile JevA mobile agent on a real Android phone, with Jev making each decision.
- pi-heedChecks every side-effecting tool call against what you asked for.
- pi-jev-skill-pickerRanks the Pi agent’s skills for the task in front of it.
- Planning while Jev picks the next actionA 105-second demo of the async split between the two models.
- playjevTyped browser automation: Playwright acts, Jev decides.
- prompt2jevTurn a prompt, or the code that runs one, into a Jev decision.
- Pydantic AI adds Jev on day oneOne string change to switch a Pydantic AI agent to Jev.
- skillrankerA Rust CLI where Jev ranks agent skills for the next step.
- SniffslopA browser extension that classifies text with Jev.
- solari-reflexA speed layer for computer use: one Jev decision per step.
- Stagehand on a remote browserBrowser tasks for about a tenth of a cent each.
- super-jevA small decision-to-action harness for Jev.
- The Jev and Claude Code stack, run wideOne prompt lets agents run wide instead of long.
- The Jev stackOne builder’s agent stack, shared.
- TipTourA pointer companion for the Mac, with Jev driving the cursor.
- TypeSafe agent skillsThe official skill for building with Jev, installable with one command.
- typesafe-computer-useComputer use on macOS, with one Jev decision per step.
- typesafe-mcpAn MCP connector that gives your agent direct access to Jev.
- unclutterA browser extension where Jev decides which parts of a page are clutter.
- Voice-controlled computer use on a MacThe app opens before you finish the sentence.
- Website to AppPaste a URL and Jev decides how to rebuild the site as a native mobile app.
- WeChat automation with Jev and DeepSeekThe accessibility node tree read first, screenshots only as a fallback.
Jev for context engineering and agent memory
Compaction, pruning and memory: what the model is allowed to read. 27 builds
- A memory gate for Claude CodeOne question per message: is this worth remembering?
- Agent memory is usually one growing fileLoaded whole on every run, and Jev could cut that cheaply.
- Beacon fixes a coding agent’s memoryJev on the memory decisions in a coding agent.
- cairn-jev-labTest what your AI should remember, with Jev deciding admission.
- codex-context-dietA Codex plugin where Jev trims bulky tool results.
- compact-adviserTells you when your session is at a safe point to compact.
- dsh-jev-pruneContext compaction for DeepSeek Harness, judged by Jev.
- fast-jev-compactionClaude Code compaction that deletes instead of summarising.
- How Jev and Laya help support agents prepare contextBoth models tried on an agent’s context step.
- Instant Claude Code compactionJev compacts the context the moment a redeploy makes it stale.
- Instant compaction for ClaudeJev scores every tool call and drops the ones that no longer matter.
- Jev SiftClassify first, read selectively: an agent plugin and MCP tool.
- jev-harnessEvery tool result is filtered by Jev before the model reads it.
- Jev-MemA paper on System-One-controlled agentic memory.
- jev-prunerTrims long Bash output with Jev before Claude reads it.
- jev-rulesJev picks which of your rules apply, so Claude only sees those.
- Let Jev pick the memories, not the answerCandidates selected for Codex or Claude Code to read.
- perfectrecallAgent memory where Jev decides what is worth recalling.
- pi-jevA Pi extension: Jev picks the file excerpts, a local model writes the code.
- pi-jev-compactionAutomatic context clearing for Pi, without losing the conversation.
- pi-jev-context-curatorA context curator for the Pi agent, built on Jev.
- Skill suggestion against context bloatOn-demand loading for a folder full of custom skills.
- Skill suggestion injects one skillJev decides which skill is relevant, so the rest stay out of context.
- tab-jevA Jev-like model paired with a tabular foundation model.
- The state packet you put in front of JevState is evidence, not a second prompt.
- winnowA context sieve for Claude Code: Jev judges each tool result before it lands.
- YoshiA context-pruning proxy for Claude Code and Codex.
Jev for code review and coding agents
Pull requests, linters, tests and code search with a decision inside. 32 builds
- A coding agent that asks Jev where to lookCursor, one worktree per feature, Jev for the narrow decisions.
- A launch inspector for vibe-coded projectsA score, the problems, the evidence, and a fix prompt.
- A syntax highlighter on JevJev names the language, colours every word and fires nine lint rules.
- blinkCodebase search by sending a hundred walkers through the file tree.
- CannyA warden that will not let a coding agent finish on a claim.
- DiffJuryPaste a public PR link; Jev says if it is safe to merge or needs review.
- End-to-end tests run by agentsAn open-source framework for agent-run e2e tests on web and mobile.
- HunchCode review against rules you write in plain English.
- Jev PR LabelerGitHub PR labels chosen by Jev, with size measured as scope, not lines.
- Jev ReviewA staged code-review workflow with a local dashboard.
- jev-auto-approveA pull request approver that asks Jev before it approves.
- jev-e2eThe same eBay test flow: 47 s on Jev, 79 s on Claude Sonnet 5.
- jev-judgmentAn agent skill that sends a coding agent’s closed judgments to Jev.
- jev-oas-sentinelOpenAPI compatibility checks, structural and semantic.
- jev-oxlintTurn an agent skill into an Oxlint plugin that asks Jev.
- jev-reviewAn MCP plugin that scores a coding agent’s work while it runs.
- jev-review-actionA GitHub Action that reviews and classifies PRs with Jev alone.
- jev-semgrepgrep by meaning, and across languages.
- jev-specChecks generated code against the specification that asked for it.
- jev-test-filterScore every test against a git diff, then emit the runner arguments.
- jev-turn-analysisReads a finished coding session and judges each turn.
- jevcumberCucumber tests from the .feature file alone. No step definitions.
- jevgrepSemantic code search for agents: find behaviour, not strings.
- JevoptCompiler optimisation decisions made with Jev.
- JevPRRisk review of a pull request, decided rather than written.
- patdownA fuzzy linter: Jev gives your code an ocular patdown.
- Scoring codebase relevanceA Jev tool that scores how relevant code is.
- siftrSemantic code search for coding agents, benchmarked on SWE-bench Lite.
- taste-lintCatches AI slop in a diff before it ships.
- Teach Codex to call JevOne skill install, and code judgement goes to Jev instead of the LLM.
- The Jev-enatorThree Claude Code hooks that each ask Jev a question.
- Ticket triage with Jev in the loopJev asks where to look, then flags test gaps after the diff.
Jev model routers and tool routing
Which model, which tool, which skill handles the request. 42 builds
- 20,000 events a day, 4% reach the writerThe model bill is decided before the model is called.
- A chat bot with no LLMJev picks the tool and its arguments, so answers are instant.
- A chief of staff for botsJev reads the task, picks the agents and gives each one a model.
- A model router on JevJev picks which model fits each request, then sends it there.
- A prompt box that fills itself inJev picks the agent, model, computer and folder for each prompt.
- A side panel that drives any site's WebMCP toolsOn each keystroke Jev picks the relevant page tool.
- A Slack agent, twice as fastJev picks the skill, tool and parameters before the agent starts.
- A smart 404 pageCorrects typos and redirects the visitor.
- A structured decision with a probabilityUseful for model routing and for agent loops.
- A usage router for Grok BotJev decides which route an agent takes before it spends anything.
- An agent with a Jev model routerJev picks which model handles each request.
- codex-jev-routerRoute Codex work through a Jev decision.
- First JEV experience: OpenRouter plus a Grok botA token from OpenRouter, plugged into a bot.
- Grok-Jev ReflexAn open-source router that combines Grok 4.7 with Jev.
- hermes-jev-skillsModel routing, memory, compaction and skill selection for Hermes agents.
- Intent-RouterTyped IntentSpec contracts as the input layer for routers.
- jcrTurns a vague agent request into a deterministic command.
- Jev Codex RouterJev picks the model and the thinking effort for every Codex turn.
- Jev decides whether Cursor wakes the modelA Noul, Choice and Score layer inside the Cursor loop.
- Jev inside CodexRouterA more optimised routing workflow, with a custom harness next.
- Jev on OpenRouter and 2,000 Monid toolsThe agent picks the tool; the reported speed-up is 30x.
- Jev picks the reasoning level firstA project where Jev chooses how much reasoning runs before generation.
- Jev Router on OpenRouterA cache-aware model router that picks the model and reasoning effort per request.
- jev-agent-skill-routerRoutes which agent skill runs, with typed Jev decisions and confidence.
- jev-dsh-decisionA structured decision plugin for agent harnesses, built on Jev.
- jev-flash-routerAn MCP server so coding agents stop spending tokens on small choices.
- jev-harness-routerOne batched call picks model tier, tools, skill and effort for each turn.
- jev-mcp-dispatcherPlain-language MCP tool calls with no LLM at all, only Jev.
- jev-routerRoutes each Claude Code task to the cheapest model that can do it.
- jev-router (skill)A Claude Code and Codex skill that picks the model for a prompt.
- jev-skill-routerSkill routing for an agent, decided by Jev.
- jev-tool-routerPicks the MCP tool for Codex, instead of listing them all.
- jev-tool-runnerJev selects the developer tools; Codex writes the code.
- On-device audio, no LLM in the loopJev makes about 20 decisions in one call and picks which local model runs.
- pi-jev-routerA Pareto-optimal OpenRouter router for Pi, decided by Jev.
- pi-jev-router (win4r)Model routing at task boundaries, with caching and failover.
- Real-time agent routing with JevA routing demo for developers building agents and AI tools.
- skill-pickerRanks more than thirty agent skills against a task.
- switchloomDeterministic model routing for coding agents, with a skill for Codex.
- TanStack AI subagents, chosen by JevWhich models or agents run, in parallel, sequence or mixed.
- Watching 25 customer groups on WhatsAppJev reads the groups and only wakes the LLM when it matters.
- Which agent answers in a group chatMentions, an LLM orchestrator and Jev, compared on the same job.
Jev SDKs, CLIs and MCP servers
Clients, CLIs, MCP servers and gateways in every language. 41 builds
- 16 paying customersJev’s question API already has buyers.
- A spoiler blockerA Chrome extension that prevents spoilers with Jev.
- any2jevPut a Jev-shaped interface in front of another model.
- AnyDecisionModelTyped decisions from language models, as a Swift package.
- Building with JevA skill for writing and fixing programs that call Jev.
- discernUncertainty-aware pattern matching and control flow for Effect.
- duckdb-jevAsk a question about every row in SQL, and get a real SQL type back.
- go-jevA Go SDK and CLI for typed Jev decisions.
- gutElixir control-flow helpers for LLM and Jev judgment.
- hearimA Jev-compatible, multi-backend System One gateway in Go.
- hermes-typesafe-jevChoice, Noul and Score as first-class MCP tools.
- jev CLIA typed judgment CLI: questions in, answers and exit codes out.
- Jev goes live on the B.AI APITypeSafe’s first System One model, available to B.AI developers.
- Jev is live in New APIChoice, Score and Noul through the same TypeSafe SDK.
- Jev joins MindsHub’s free tierTypeSafe’s decision model available on the free tier.
- Jev on Cloud RunAbout 47 s cold start and about 117 ms end to end on a cloud GPU.
- jev_elixirJev from Elixir.
- jev-axiA shell CLI for agents: pick, rate, check, rank, triage and guard.
- jev-cliAsk Jev from a shell, and get an exit code back.
- Jev-CLIJev from the command line.
- jev-feelsSemantic decisions as ordinary Ruby: feels?, decide, score.
- jev-foundation-modelsA Swift 6 bridge from Apple's Foundation Models to Jev.
- jev-mcpAn MCP server with claim checks, screening and ranking built on Jev.
- jev-recipes80+ TypeScript recipes for agents, retrieval and verification.
- jev-rsA Rust, Jev-compatible /v1/systemone engine, one prefill per judgement.
- jev-spring-boot-starterJev as a Spring Boot 4 starter, over Spring MVC and RestClient.
- jev-studio v0.2.0One pip install, MCP tools for Choice, Noul and Score, and a CLI.
- jev-workflow-builderBuild a Jev workflow visually.
- jevalynA Rails-native decision layer: typed answers inside control flow.
- jeviewA local visualiser: every Jev call your code makes, live.
- jevkitA Rust CLI for Jev, with a linter that checks questions before you pay.
- LangChain's Jev tool-call gateTool output stays out of the classifier's input.
- Laravel TypeSafe JevAn unofficial Laravel package for Jev, with typed responses and test fakes.
- mazaA local MCP gateway with Jev tool discovery and a dashboard.
- mcp_jevA local MCP server that runs ready-made Jev question packs.
- MercatorA pipeline for paid services, turbocharged by Jev.
- pg_typesafeA PostgreSQL extension that calls Jev from SQL.
- Routing agents with Jev and Laya in LangChain4jSystem One decisions inside a Quarkus workflow.
- Speeding chat answers with pre-structured SQLA brainstorm on Jev plus pre-structured queries for faster replies.
- system-oneA provider-neutral System One runtime for TypeScript and Pi.
- typesafe-sdk-goA Go client for the TypeSafe API.
Open source alternatives to Jev
Open-weight alternatives to Jev, and the runtimes that serve them. 123 builds
- 805 verified Jev examplesExamples indexed by the decision each one makes.
- A Jev-style local model for imagesTyped decisions, but for image inputs.
- A Jev-style model on a laptopRun it locally.
- A local classifier instead of JevJust as fast for the job, with a Python server holding the model.
- A local JevA Jev-style model running locally, with room to get faster.
- A local open-source version of JevRun a Jev-like decision model locally.
- A WeChat reply assistant with Jev and DeepSeekWindows and Apple Silicon builds, with local OCR and human control.
- AgentJev-0.6BState in. A distribution out. Nothing decoded.
- An LLM distilling JevTypeSafe notes an LLM distilling Jev.
- An open alternative appears: Laya421M parameters, built to decide, runs locally.
- An open typed-decision classifierYes/no, a score or a choice, with no chat and no generation.
- AnyJevTurn any LLM into a Jev-style decision model, with no training.
- AppitStudio/awesome-jevCurated Jev resources and runnable examples for typed decisions.
- APUS-OpenJev-v1Decision models for browser agents with a selectable compute depth.
- arbiterServe your own typed-decision model behind a Jev-shaped API.
- autotrust/JEV-9BAn open student of Jev 1.13 with System 1 and System 2 in one set of weights.
- awesome-jev-surveyAn evidence survey of Jev and Jev-like typed decision models.
- awesome-typesafe-jevA source-backed field guide to TypeSafe’s System One model.
- Beating Jev with open modelsAn attempt to beat Jev’s accuracy, speed and cost with open models.
- Bringing Laya to Insight O MateA Jev-inspired move to a local model.
- Clef and Clef-flashCloudflare's open-weight decision models, Jev API compatible.
- Clef on Hugging FaceThe Apache 2.0 weights for Cloudflare's Jev alternative.
- clef-flash running locally on an M5 MaxCloudflare's new decision model can see, and it runs on a laptop.
- CLM-8B claims to be faster than JevAn open model that its release claims is up to 9x faster.
- CLM-8B, a new Jev competitorAn 8B decision model positioned against Jev.
- CUA-S1-FORMSAn open 706k-parameter model that fills a form in one 50 ms pass.
- decider-2bAn open reproduction of the System One model class on Qwen3.5-2B.
- Decisive judgment agents that don’t chatJev against Laya.
- deepseek-v4.1-flash-jevAn open model made to answer like Jev with a scoring endpoint.
- djevJev-style structured decisions on DiffusionGemma.
- Eikos-27BAn open typed-decision model for finance and trading rules.
- Eleven trending Jev reposThe Jev repositories moving fastest on GitHub, in one post.
- From Jev to Laya in an AI video pipelineA developer moves the decision layer from Jev to Laya.
- headroomA local-first brain-budget meter for Codex, scored by Laya.
- Hundreds of tools built on Jev, collectedA GitHub collection treating decisions as their own primitive.
- Jeff 1An open typed-decision model with a guide to training your own.
- Jev against Laya, a smoke testAccuracy, calibration, latency and cost per decision, side by side.
- Jev vs. Laya-MLXA head-to-head comparison.
- Jev-CompatibleMake an existing SGLang or vLLM deployment answer as a decision service.
- JEV-CPUSemantic ifs from an open model on a laptop CPU, no GPU.
- jev-dataopsA workbench for data selection, quality scoring and LoRA training.
- Jev-like decisions from open LLMsTurn a low-cost open model into a fast decision model, without training.
- jev-liteA QLoRA adapter that turns Gemma 4 E4B into a System One model.
- Jev-OmniA decision classifier for text, images, audio and video.
- Jev-Style 0.8B Decision v3Jev-style decisions on a laptop, 0.53 GB in 4-bit.
- jev-suiteFour decision-quality tools on Jev, with code keeping the final say.
- jev-treeA probability tree and graph runtime for verifiable decisions.
- jevelryOpen code for Jev, shown on Hacker News.
- JevEmbedTurn embeddings into choices, scores and judgments.
- jevflowProbabilistic decisions as composable backend primitives.
- jevgraphKnowledge graphs built from typed relation decisions, with the evidence kept.
- JevHubA hot list of the Jev ecosystem.
- jevk5An open-weight alternative to Jev, Apache-2.0.
- jevlikeA Jev-like architecture, reverse-engineered, to train your own.
- jevperA Jev-shaped classification wrapper over OpenAI-like clients.
- JevTunerTuning for Jev-style decisions.
- Julia 1A 144.3M-parameter decision model that edges Jev on typed decisions.
- KaLM-JevA local judgement engine in Nano, Small and Large sizes.
- kevA tiny Jev-like model on Qwen2.5-0.5B that trains and runs on a MacBook.
- KevAn open-source System One engine with calibrated probabilities.
- LayaOpen weights that answer typed questions in one forward pass, 33 ms on a T4.
- Laya compared to JevAn open-source model next to the official one.
- Laya on a 12-year-old laptop CPUSub-second decisions on an i5-5200U with no GPU.
- Laya VisionTyped decisions about an image plus text, in one forward pass.
- laya-appleA correctness-validated Laya runtime for Apple Silicon.
- laya-browserLaya fine-tuned into the decision head of a browser agent.
- laya-candleA Jev-like Rust port using Laya, fully local and with no Python.
- laya-coremlA faster, local Laya-like model for the Mac.
- laya-fastAn Apple Silicon runtime for Laya: Neural Engine plus GPU router.
- laya-goA Go server plus agent CLI and MCP for open System 1 models.
- laya-goishLaya GGUF inference and an HTTP server in Rust.
- Laya-MLX, taken apartA repo to understand an encoder-only decision architecture.
- laya-mpsJev-style typed decisions on a Mac, on little RAM.
- laya-pilotReplayable Excel test cases from live browser pages.
- laya-serverA self-hosted API for Laya, compatible with the Jev API format.
- laya-ultrafastjev-ultrafast, with Laya in place of Jev.
- Laya, 421M parameters, local86.5 decisions a second on about 1 GB of inference memory.
- Laya, 50× faster, on open alternativesA benchmark of the open-source Jev alternatives.
- lejudge-jev-jepaNatural-language constraints for JEPA planning, judged by a decision model.
- Local Laya against cloud Jev, on latencyAbout 45 ms deciding locally against about 300 ms in the cloud.
- Local Laya beats JevA local decision model coming out ahead.
- LogJevText, images or audio in, choices and scores out of the logprobs.
- mini-JevA frozen Qwen3-0.6B plus a 1.1 MB head for tool selection.
- NanoJevA 0.6B replica: parallel decisions in one forward pass, no decoding.
- NeoHorse-Jev-4BPrefill-only decisions for agent workflows.
- Open Alternative to JevTyped, calibrated decisions from any open-weights model, on your own GPU.
- Open JEV DeBERTa V3 LargeA text classifier that returns typed decisions, not labels alone.
- Open Jev PlaygroundOne place to try every open-source alternative to Jev.
- open-jev (browser TypeScript)Typed decisions in the browser: one state, any number of options.
- open-jev (PyTorch)A from-first-principles rebuild of the ideas behind Jev.
- Open-Jev-9BA LoRA adapter and decision head for Qwen3.5-9B.
- open-jev-layaBrowser-only Laya demos powered by Transformers.js.
- open-spark-jevLocal System One models on Qwen3, for an NVIDIA DGX Spark.
- OpenJevAn open attempt at a Jev-class decision model.
- OpenJevAn open Choice, Noul and Score API on open models.
- OpenJev (open weights)Typed decisions about text, web pages and screenshots, within 2 points of Jev.
- OpenJev (Verdict)An open 151M decision model on ModernBERT, with a WebGPU playground.
- openjev on Qwen 4BAn MLP trained on top of Qwen 4B that works like Jev.
- OpenJev Verdict 2.0A 151M non-autoregressive decision engine, with its numbers published.
- openjev-sglangA public Jev-compatible API on an open model: 64 tasks in under a second.
- OpenThai-SystemOneAn open Thai and English System One decision model.
- Post-train your own judge on your agent's tracesA 4B, 8B or 27B judge that beats Jev on human agreement.
- Predicting queries before submissionA community test with Jev and Laya.
- QwenJevA Jev-style conversion of Qwen-3.5-4B, at 0.169 s per frame.
- reflexA small open decision model: state and typed questions in, probabilities out.
- rizzo-flowThe open, local take on Jev: typed decisions without generating a token.
- Simple JevAn open-source library that turns any Hugging Face model into a Jev-style endpoint.
- Simple Jev, reusing the KV cacheShared context cached, only the candidate logits read.
- skillboxA self-hosted skills library, with optional Jev recommendations.
- snapTyped decisions from unstructured state, with no generated text.
- solar-mini4-jevJev-style decisions on Solar Mini 4.
- stuntdA local proxy that answers typed LLM decisions with a Laya head.
- sys1A System One compatible API for open decision models, in Rust.
- System One scorerQwen3.5-4B with a scoring head, type-safe by construction.
- system1-agentsJev, Laya and Cua-S1 as the brain for agents.
- Tev1 0.8BA tiny Jev-like classifier running locally on a Mac.
- Tiny-JevA 0.6B System One model: structured state in, typed decisions out.
- Trying to beat Jev on a DGX SparkAn open System One model, with the data already collected.
- Turn any LLM into a Jev-like modelA trained KV cache bank, with a live demo.
- TypeLLMAn open-source model that reasons before returning type-safe output.
- VejA Jev clone that runs locally, built with Claude in a night.
- visual-jevTyped decisions over images, not text descriptions of images.
- vonA sub-15 ms local drop-in alternative to Jev.
Jev benchmarks, evals and calibration
Measurements, calibration audits and side-by-side comparisons. 83 builds
- 154 ms against 860 ms, and the scope of itThe routing result holds for 200 synthetic classification cases.
- A browser step from 2 s to 0.2 sTen model calls became two, and accuracy fell six points.
- A correction about generating haiku with JevChoice does not generate hiragana, and the earlier claim was wrong.
- A decision brain, not an assistantJev's CEO places Claude Code in the assistance era.
- A faster, cheaper path for business modelsJev as an alternative for business logic.
- A free tool that scores how rigorous a paper isJev reads an academic paper and reports how rigorous it is.
- A Jev idea that hit a wallA developer asks TypeSafe to fix a limit.
- A Jev speed test app, made publicOpened up because the usage fees are small.
- A number two open-weight Jev-like modelAdded to the Decision Index, strongest for its size.
- A review of Jev for AI workflowsFast, structured decisions, with the vendor's own figures quoted.
- A structured decision layerOn Jev as the decision layer in AI.
- A sub-100 ms decision engineHow the launch was described.
- An agentic classification loop, 7x fasterThe loop replaced with a single Jev call, and the numbers published.
- An Astra-made explainer for JevNot understanding the hype, the author had a model explain it.
- An ELI5 for JevThe simplest explanation of TypeSafe’s Jev.
- An eval suite for System One modelsBuilding the tests Jev-like models need.
- askjev.aiAsk Jev anything. It won’t answer, it will judge.
- Auto-generated backgrounds for a Jev explainerDirecTune produces the visuals and the video around them.
- awesome-jev-robustnessTests, calibration audits and failure-mode studies of Jev.
- Calibrated confidence in 70 to 500 msYes/no, choices and scores, with the published price per token.
- ChatJevDriving the classifier as a next-token predictor, one token at a time.
- Coverage is not product valueJev judges rather than writes, and that is where its value sits.
- daf-jevA Python toolkit: question builders, confidence gates and calibration.
- Decisions inside the system, not in a chatWhat makes Jev different from a conversational model.
- Early access since September 15A System One model for agent decision loops, in typed options.
- eveVercel’s open agent framework, with Jev as its default evaluate model.
- Every’s editorial vibe check37 documents, 21 questions each, 1,709 judgments for under a cent.
- Fast, structured decision-making for AI workflowsA wire piece on Jev.
- Five dollars a month was not enoughReal tasks forced reading the skills docs and using state properly.
- Hard-coded rules moved to JevIlias Ism uses Jev instead of regex in aiseotracker, linkdr and genppt.
- How Jev works, in a one-hour masterclassThe founder explains.
- Is a quick classifier worth building?The traps that come with building one fast.
- J++An experimental language for composing Jev questions.
- Jev 1.13.0 against Hopper on Benchmark HeavenHopper wins three columns, Jev keeps Intelligence.
- Jev against Fable 5.1, 100 buildersThe same post scheduler, built twice, in front of an audience.
- Jev against Llama and DeepSeek235 ms median latency and 82% accuracy on the same labelled inputs.
- Jev and Laya on a medical diagnostic testSixteen of sixteen against thirteen, on authored notes.
- Jev as an observer of agent runsGood for progress and cost-to-finish, not for guarding anything.
- Jev Capability AtlasWhere Jev holds up and where it breaks, with real API receipts.
- Jev catches the Opus 5.5 effort dropThe default effort moved from high to medium, and agents without settings lost reasoning.
- Jev chooses, the LLM writesA short explainer with a 1x demo, no speed-up.
- Jev Engineering as the brain of an agentAn answer to unstable results from free-form prompting.
- Jev for next best actionTesting Jev’s accuracy on predicting the next action.
- Jev in tool filtration testsPraised for filtering tools.
- Jev on a movie recommenderTesting Jev on recommendations.
- Jev vs. Sol, on categorisationA text categorisation showdown.
- Jev_StarA Jev and System One related repository.
- jev-acentoA pre-registered audit of how Jev handles Spanish.
- jev-alignTurns human labels into calibrated Jev classifiers with GEPA.
- jev-architectHelps you find, design and evaluate a Jev decision loop.
- jev-as-judgeA refund agent graded by Jev, recorded as an Opik experiment.
- jev-calibrateTune your Jev questions against your own labels, then confirm on held-out data.
- jev-designA console whose design system is chosen by Jev at runtime.
- jev-evaluationNine experiments and 28 predictions, all fixed before any data.
- Jev-in-the-LoopResearch on which LLM-decision tasks Jev actually speeds up.
- Jev-QuantumA sub-microsecond System 1 model whose accuracy is a Gaussian.
- JEV, por qué ha revolucionado todoA Spanish guide to what Jev is and how to try it.
- jevalWorks out what a Jev confidence score is actually worth.
- JevAnyCalibration-aware reinforcement learning for decision systems.
- JevBenchA reproducible benchmark for typed decision models.
- jevchatA chatbot built out of a model that only answers typed questions.
- JevGuardA decision runtime with zero-token caching and a calibrator.
- Keystroke oracleA predictive launcher, the first of Nader Dabit’s Jev experiments.
- LeJudge: JEPA × JevA JEPA world-model planner with Jev as the judge.
- Metask-JevCalibrated typed-decision models, one forward pass each.
- Not a ChatGPT competitorThe question it answers is whether to act now.
- Not every model needs to generateSome models should only get very good at deciding.
- OpenRouter’s Ori EvalJev judged a 30-way labelling task faster than every model tested.
- Reading the System One blog with an assistantGemini Enterprise used to break the articles down.
- Separating signal from noise with Jev and OpusA pattern can still be noise, so ask what would change your mind.
- State in, auditable decisions outA short summary of what Jev returns.
- tenbinSplit a judgment into Choice, Score and Noul, lint it, then measure it.
- The best explanation of JevA pointer to a clear Jev explainer.
- The decision-making brain for agentsJev framed as an agent’s decision brain.
- The first public System One modelState and explicit questions in; choices, scores and probabilities out.
- The System One paradoxOn System One models and where Jev goes next.
- Tokens saved, AI employees multipliedWhy the author treats Jev as a skill worth learning properly.
- Trust the ordering, not the confidence numberWhere to place the decision model inside an agent harness.
- TypeSafe introduces System One models and JevA recap of the launch.
- TypeSafe launches JevA millisecond, low-cost decision model for agents.
- TypeSafe’s blueprint for building with JevThe founder’s approach to building on Jev.
- TypeSafe’s System One, for decisionsA recap.
- xor joins the Jev Decision Index 0.2A new open-weight entry, eighth on text alone.
Jev for semantic search and reranking
Search and ranking by meaning, over a site, an inbox or a vault. 26 builds
- 1,066 reaction videos, searchableGhostfeed labelled its reaction clips with Jev and Gemini, then opened the library.
- A bookmark organizerJev sorts your bookmarks.
- A forum database, searched by JevJev finds the most relevant posts in a forum database.
- A RAG loop where Jev decides what is realHybrid search, one Jev scoring call and a claim check, with agents around them.
- A visual reference finderOne prompt, 100 images from Cosmos, NASA and The Met.
- Automatic note taggingEvery note tagged by Jev, built for Obsidian users.
- enskAn English–Icelandic dictionary with Jev reranking the results.
- Filtering WIP todos by meaningFinds the todos where revenue moved, or the work got stuck.
- Flight search with Browser UseA flight search run by a browser agent in about seven seconds.
- Intent-based search in a Chrome extensionSearch by what you meant, with Jev picking the match.
- Intent-based search in GmailSearch an inbox by what you mean, not the words you type.
- Jev for RAGTrust semantic matching more than dot-product similarity.
- Jev-powered Obsidian searchSearch an Obsidian vault with Jev deciding relevance.
- jev-rankerJev-powered reranking for RAG.
- jev-searchOffline Obsidian search, with an optional Jev rerank you approve first.
- jev-search-mcpPlain-language web search where Jev picks the sources.
- jev() for PostgreSQLSearch a table in plain language, with no index and no embeddings.
- JevRelevanceRetrieverAn open-source LangChain retriever built on Jev.
- laya-jev-GraphRAGGraphRAG with swappable System One models.
- Margin, bookmark searchBookmarks found fast, where a large model was too slow.
- Natural-language search over ZillowThousands of listings sorted by what the site has no filter for.
- neo4jevWalking a Neo4j graph one hop at a time, each hop a typed choice.
- Search and tagging on keep.mdJev on Cloudflare Workers, tested against keep.md’s current search.
- Semantic BookmarkA Chrome extension that files bookmarks by rules you write.
- SemDecidegrep for meaning and jq for judgment, with a Unix exit code.
- YC IndexorSearch 6,241 YC companies in about a second, by meaning, look or logo text.
Jev for OCR, documents and classification
Reading a large collection and returning one typed answer per item. 22 builds
- 1kpapers1,018 AI papers sorted by topic for $0.08 and published as a site.
- 2,300 AI papers in about 83 secondsA small-team case, with a caveat about where the cost came from.
- 24,000 Hacker News posts in two minutesJev on the Venice API, sorting a day of Hacker News into 12 buckets.
- 900 images in 40 secondsFayaz Ahmed’s image classifier: OCR reads each image, Jev sorts it.
- A 12-page PDF on building a Jev harnessFrom the Jev founder.
- A 26-sheet plan set in 2.9 secondsCivil and building plan sheets classified for half a cent.
- A thousand papers sorted, then gradedJev does the 24-topic sort, Opus 5 grades the labels.
- captureA private Mac voice diary: local transcription, Jev sorting, Notion library.
- Classify rows in DuckDBHamilton Ulmer’s DuckDB extension: Jev classifies the rows of any table or file.
- Company invoices, classified for the booksEvery invoice reviewed and given an accounting category in seconds.
- DocJevClassify and split document packets with Jev. Open source and free by default.
- Geo PulseClassifying geopolitical news in real time with Jev.
- hermes-jev-curatorTyped relation governance and safe archive plans.
- How to Give JEV EyesJev takes text, so turn the picture into text first.
- jev-curateA Rust dataset sifter that keeps clean rows verbatim and drops the rest.
- JEV-Paper-RadarJev reads every new arXiv paper each morning and keeps the few you need.
- jev-papersA thousand arXiv papers, one Jev decision each, checked by an LLM judge.
- jev-reviewerSystematic-review data from trial reports, every answer a verbatim quote.
- Nitpicky, zooming into AI imagesFaces, fingers, text, numbers and poses checked one at a time.
- Tocsin22.8M log lines grouped into patterns, then one question each.
- Which exam questions come back80 real exam questions and 297 practice ones, ranked in 80 seconds.
- 大声读 (dasheng)Streaming speech recognition listens, and Jev marks each word.
Jev for guardrails, moderation and security
Guardrails, malware checks, fraud and moderation. 22 builds
- A CAPTCHA rebuilt with JevThe browser measures how you fill the form. Jev decides.
- A Jev gate belongs in a hook, not MCPMCP tools the agent can skip are not a gate.
- A phishing message detectorTypeSafe Jev checks messages for phishing.
- Agent ChaperoneScreen an agent's tool calls with Jev before they run.
- An agent-agnostic permission gateJev and hooks, so a dangerous command asks first.
- BouncerA judgment layer that checks every Claude Code tool call against your policy.
- Every rule checked in 348 msJev reads each Claude Code reply and edit against the rules you wrote.
- firehose-judgeThe live Bluesky firehose, judged post by post, with a lane for humans.
- Fraud detection with Jev and Kimi K3Jev sorts 100 emails in 1.42 s and sends the unsure ones to Kimi K3.
- is-maliciousA codebase scanner that helps you not run malicious code.
- JcyberA bug-bounty chain where Jev gates every step.
- Jev and Laya on security queriesTwo decision models make different calls on the same security queries.
- Jev Anti-Spam BotA Telegram bot that deletes only the spam Jev is sure about.
- Jev as an agent safety monitorChecks each agent action first: most attacks caught, almost no false blocks.
- Jev blocks a $50,000 transfer95% irreversible risk and 94% sensitivity, so it refused.
- Jev finds the sensitive files in a shareA security researcher picks out risky files with Jev.
- Jev Moderation BotA Discord moderation bot with Jev making the calls.
- jev-crawlersBug-hunting crawlers built from five Unix-style primitives.
- jevmodModeration with a probability per category and thresholds you set.
- mastra-jev-moderationInput moderation for Mastra agents in one file.
- Testing Jev for prompt injectionA demo that shows how to test prompt injection against Jev.
- UXRayA screen overlay that flags manipulative interface patterns.
Jev for email triage and support tickets
Email, tickets and chat threads sorted into buckets. 13 builds
- 500 emails for 3.5 centsRiley Brown classified 500 emails in seconds.
- A WeChat assistant, open-sourcedJev behind a WeChat assistant.
- Batch ticket analysis with CowAgentOne call per ticket returns seven classifications with probabilities.
- crush-monitorReads a WeChat thread and scores emotion, intent and reply quality.
- Fifty emails in under two secondsAn email classifier that stays fast across a region hop.
- Jev inside a telemarketing AIBuilt on the GPT Live API, with the decisions moved to Jev.
- jev-chat-jarvisA reply co-pilot for phone chat apps: it reads the screen and drafts, you send.
- jev-chat-jarvis-macA macOS overlay that judges message intent and risk, then drafts.
- jev-chat-windowsA WeChat reply helper for Windows: screenshot, local OCR, three drafts.
- jev-inboxA Chrome extension that finds and labels your important mail.
- jev-wechat-liveAn abandoned WeChat overlay, published with the full post-mortem.
- Support answers in a Mac appJev reads the built-in manual and picks the article that answers the user.
- Triage across 1,500 emailsAbout 1,500 emails sorted in one run.
Jev for ads and marketing research
Competitor ads, creative testing and campaign research. 16 builds
- 1,891 ads in 19 secondsA whole brand ad library, labelled by funnel stage and style.
- 178M tokens spent testing marketing ideasNearly 48,000 requests through Jev for one campaign.
- 430 competitor ads a secondEvery live ad tagged by hook, format and survival rate.
- 700 live ads in 40 secondsA marketer scanned more than 700 live ads with Jev for four cents.
- 723 ads, 30 personas, 22 centsA scroll-or-stop focus group run entirely as Jev decisions.
- 724 competitor ads, broken down724 live ads from 37 brands, analysed in about 40 seconds.
- A shortlist from a competitor swipe fileAds broken down by hook, angle, offer and format.
- Ad creatives from filtered assetsJev filters content and picks assets; DeepSeek and Higgsfield turn them into ads.
- An A/B test against 4,000 personasTwo launch tweets, voted on by survey respondents, in twelve seconds.
- Frevana decodes a category's ads in 20 seconds294 TikTok ads across 68 brands, for about two cents.
- Grok thinks, Jev decidesSix questions asked, and only the good ones opened.
- hookmeter-jevMillisecond viral hook telemetry, in a Chrome extension.
- Jev for content marketingJev watches every video in your niche and judges it before you see it.
- Scoring competitor ads with Opus 5.5 and JevJev makes the ad judgments.
- Tearing down Meta ads and their landing pagesWhat page type, which offer, and how fast the creative ships.
- Winning patterns out of the Meta Ad LibraryLong-running ads mined, search terms filtered, landing pages scored.
Jev for lead scoring and sales
Scoring leads, outreach and candidates before a human sees them. 14 builds
- 400 companies matched to one candidatePredicts which jobs a candidate is most likely to get, for $0.0005.
- 700 leads scored in 40 secondsPredicts how each outreach message will perform, for $0.09 in total.
- A candidate for agent brainsCould Jev be the brain of an agent?
- A lead qualification workflowTriage what the agent can decide, then decide the rest.
- A skeptical take on JevWhere the modeling choices lead to real failure modes.
- call-coach-aiA sales-call coach where Jev scores what was said.
- clay-jev-people-rankerLead scoring in Clay, ranked by Jev instead of a prompt.
- Jev RecruiterA LinkedIn recruiting agent that browses profiles and saves the evidence.
- jev-job-hunterA job crawler that finds the Careers page and scores each role against your profile.
- Only 47% of leads fit the ICP605M tokens of live campaigns screened with Jev.
- Qualifying candidates with JevAn internal tool, and the limits of a rubric.
- Sales CopilotListens to the call live and tells the seller what to say next.
- Scanning Reddit for leads with JevTwo days of building, and no replies.
- Which outreach signals book demosJev read thousands of outreach messages and found the signals that booked demos.
Jev for SEO and GEO workflows
Page audits, internal linking and AI citation work. 9 builds
- App Store keywords from the real competitionJev drops the lookalikes before they pollute the keyword set.
- Internal linking toolFree tool: Jev classifies the pages and picks which ones link to which.
- jev-seoAn agent-first SEO and GEO CLI in Rust, with DuckDuckGo and Jev.
- jev-seoA live SEO audit for any site, with Jev judging every page.
- jevseoTyped SEO, AEO and GEO judgements, with the rules staying in code.
- LurkFind and monitor Reddit threads to get cited by AI, for free.
- NotraTurns your work into content, with Jev classifiers for GEO.
- SEO and GEO fixes, 90% cheaperRyze AI’s agents audit and fix a client’s SEO and GEO with Jev.
- Seven more SEO and GEO workflowsScores competitor pages, then names the element to change on each.
Jev for ecommerce and shopping agents
Listings, shopping agents and the paperwork behind an order. 4 builds
- A second-hand shopping agentReads about 26 listings a minute and decides on each in 406 ms.
- A shopping conciergeJev decides as you shop.
- jev-weekend-shopping-chromeWhile you shop, Jev guesses the seller's working hours culture.
- One-click invoice finderFinds the billing page on any website and lists every invoice.
Jev plays games
Games and simulations where the answer has to arrive within a frame. 74 builds
- 1v1 Jev — Quickscope ArenaA browser shooter where the opponent is Jev, first to five kills.
- A 3D character that reacts while you talkFace and body responses chosen by Jev, including while watching YouTube.
- A leveling agent that gets cheaper as it runsA guide-directed WoW agent designed to spend less the longer it plays.
- A Minecraft bot that builds and fightsHouses, towers, lakes and castles on command, mobs handled live.
- A Monument Valley-style gameAfter Snake and Tank, a 3D game built with Jev and Astra.
- A Nintendo DS game built with JevPerry the Platypus added, with Jev in the build loop.
- An endless web game collecting glassesJev plays forever; players submit their own screenshots.
- An ultra-fast futsal simulatorShown on 23 September at the Generative AI Anything Exhibition.
- Beat JevA penalty shootout against Jev, on Render Workflows and Postgres.
- Beating the Ender Dragon in MinecraftJev and Astra finish Minecraft in 8 minutes 43 seconds, for under a dollar.
- Both hands on a piano, in real timeJev sees the note waterfall and controls ten fingers.
- Can Jev steer music?Jev picks the plan; code renders the sheet music, audio and MIDI.
- Chess, once free and once filteredLeft to decide alone it lost; with an attack-defence filter it won.
- clash-jevA Clash Royale bot with no trained policy at all.
- Clue MeA word game where you describe a word and Jev guesses it.
- Driving pedestrians in a multi-agent simulationJev at the centre, with concurrency limits.
- Fighter jets flown by Jev agentsA GTA-style contest between Jev-powered agents.
- Game levels generated in real timeJev builds game levels as you play.
- game-coachA chess coach where Stockfish owns truth and Jev owns judgment.
- hermes-and-jev-play-minecraftHermes plans, Jev picks bounded actions, Mineflayer executes.
- Jev and Laya control Flappy BirdTwo decision models on one game.
- Jev and Laya play chess instead of benchmarksReal interaction in place of a leaderboard.
- Jev plays chessJev picks from the legal moves, compared with reasoning models.
- Jev plays DoomTypeSafe’s launch demo: Jev picks Doom moves about ten times a second.
- Jev plays PokémonPokémon Red, played one typed turn at a time.
- Jev plays Pokémon ShowdownA Codex-built harness, a saved replay, and mixed results.
- Jev plays Puyo PuyoFour questions a move: which phase, then the best landing.
- Jev plays Slay the Spire 20.7 seconds per move, where GPT-6 Astra was slow.
- Jev plays Smash Bros. against itselfJev controls all four characters and picks each move in a fraction of a second.
- Jev plays Subway SurfersAt superhuman speed, and 50 games at once, for under a cent.
- Jev plays Super Mario Bros.Real-time play with fast, structured answers.
- Jev plays TetrisTetris on its hardest mode: 357 pieces and 134 lines in two minutes.
- Jev PongPong where the ball moves one step per model decision. Jev against the LLMs.
- Jev runs a city’s trafficA traffic simulation where Jev controls a whole city.
- Jev vs the LLMs: TetrisJev plays real-time Tetris against Claude Haiku or Gemini Flash.
- Jev vs. Laya, over a chessboardBoth played poorly, so the next experiment is a different approach.
- Jev vs. Laya, over chessAnother chess experiment between the two.
- Jev-miceA mouse colony simulation with Jev deciding and a deterministic engine acting.
- jev-pianoJev can’t write a note, but with the right questions it improvises piano.
- JevMinesweeperJev plays Minesweeper, one square at a time.
- JevPilotA Three.js driving sim where Jev picks the path, up to four times a second.
- JevPokerBenchA Texas Hold'em benchmark for decision models, with leaderboards.
- jevs-flyA Three.js first-person flight, steered by Jev.
- JevTownA Jev-based AI town simulation.
- Kiru Hai CoachA solo mahjong coach: Jev picks the tile to discard. In Japanese.
- Laya beats cloud Jev at TetrisAn open-weights System One model on a MacBook Air against cloud Jev.
- Laya plays Flappy Bird on a CPUA 421M open-weights System One model, real time on a laptop CPU.
- laya-vs-jevLocal MLX and hosted decisions playing T-Rex side by side.
- laya-vs-jev-arenaTwo decision models race in Snake, then fight in an arena.
- LLM Chess: jev-latestJev played 80 games in a chess benchmark and made no illegal move.
- Local Laya beats cloud Jev at TetrisAn open decision model on a local machine.
- Luna plans, Jev plays pokerA planner and an actor, taking the pot.
- Magic Jev BallA magic eight ball answered by Jev.
- Magic Jev Ball for code reviewShake it on a pull request and Jev approves or rejects in 200 ms.
- Mario 1-1, with the round trip measuredIt failed on the first enemy, and the timings are published.
- Mario Never DiesEvery death forks the VM into four timelines and tries again.
- minecraft-agentAn Astra planner with Jev as the controller, playing Minecraft.
- OpenDecisionA 400M local model doing zero-shot decisions, and playing Doom.
- Play chess against JevA new way to use Jev: as an opponent.
- Real-time agent battles in gamesJev drives agents that fight in real time.
- Realtime virtual try-onTalk, and Jev picks the next outfit from the closet in about 620 ms.
- San Francisco rebuilt in Unreal, run by JevEvery person, pet and car in the city powered by Jev.
- Shareable results in the Jev chess appThe game can be shared now, and Jev won again.
- Snake, four waysPlay alone, play with Jev, veto its moves, or race it.
- Stanford Town, run by JevAI residents that live on their own.
- T-Rex Run: local MLX Laya vs. cloud JevA game showdown between the two.
- Tetris with JevA game of Tetris where Jev picks each placement.
- The next move in Minecraft and in a browserOne short video, two loops, with OpenRouter benchmarks.
- Two bots playing 2048 and Flappy BirdA spectator view of judgement-only AI in a game loop.
- TypeEvacSafeA fire evacuation simulation with a Jev-style decision for each person.
- TypeSafe MarioJev picks NES controller inputs from telemetry, never from pixels.
- typesafe-jev-dojoA graphical dojo: routing, Tetris, swarms and a gauntlet.
- Where the latency actually goesThe HTTP round trip is a fraction of the state-to-action time.
- WikiracingGet from one Wikipedia page to another by picking links.
Jev trader and trading bots
Buy and sell decisions made at the speed of a price feed. 25 builds
- $10,000 in Jev’s handsAbol gave Jev $10,000 and let it trade.
- $17,831 in the first nightA Jev Polymarket bot built by a student in Japan.
- 40,000 wallets screened in four secondsA Jev filter targeting mercenary farmers before a TGE.
- A Jev and Grok copy-trading botNow open.
- A Jev trading bot on TelegramNine coins, live probabilities, and a 10,000 USDT wallet.
- A Kalshi and Solana bot on Grok, Jev and LayaSelf-improving and auto-calibrating, still on paper.
- A Kalshi BTC bot on a 15-minute clockA Bitcoin trading bot on Kalshi built with Jev.
- A stock scoring engine, on a first investmentNon-mathematical stock data, scored by Jev into a decision.
- A trading loop with Jev and GPT-6 AstraThe signal becomes a trade, then the risk limit is checked.
- Cheap per call, but it adds upThe cost of running Jev inside a trading bot.
- Clawby re-scores four dimensions every ten minutesTechnical, liquidity, sentiment and risk, re-judged on a ten-minute loop.
- Crypto tagging, 20× fasterDopamyn’s account tagging is about 20× faster and cheaper with Jev.
- Fading a trading bot live on PolymarketA build using the last of the Jev credits.
- fedjev-benchDoes Jev hear a rate rise coming in an FOMC statement?
- Fewer inputs, better performanceA Jev backtest on cutting the input down.
- Hawk or dove?Fed press conferences scored word by word, Jev against chat models.
- jev-as-quantTyped System One decisions as the judgment layer of a quant stack.
- Jev-powered AutoScalper passes testnetAn 82 per cent average win rate on the one-minute chart.
- jev-traderA trading bot that places an order in every 300 ms block.
- jevocksA stock terminal: Valyu brings the evidence, Jev gives the status.
- Live data into one ID, with a hard risk capA Jev integration for risk.
- News sentiment in a trading terminalJev reads the news and returns the sentiment.
- Prism Liquidity AgentAn LP rebalancing agent that routes every Jev call through one gate.
- Real-time decisions in algorithmic tradingDynamic calls where hardcoded rules used to sit.
- Scoring Morpho vault risk with JevFifty vaults, 22.6 seconds and about $0.006.
Jev robotics, drones and devices
Control loops on hardware and in simulation. 14 builds
- A dual-arm robot with Jev in the middle layerDecisions to Jev, inverse kinematics and physics to code.
- A family-only LINE bot for the houseLights and AC by chat, with Jev keeping the cost down.
- A voice assistant with no wake wordIt decides from context whether to interject at all.
- EmbodiedJevA MuJoCo workbench for robot decisions, with Jev in the loop.
- home-assistant-typesafeHome Assistant routes what you said through Jev, not an LLM.
- Jev in embodied robot controlWhere a typed decision fits next to VLA and diffusion policies.
- Jev-as-PolicyOne-click simulation setup with Jev as the control policy.
- jev-droneA simulated drone that asks Jev for a judgment 2.5 times a second.
- OmniJevMultimodal Jev for robot decision-making.
- RoboJEVTwo-stage Jev control of a Franka Panda arm in MuJoCo.
- Strike Robot wires Jev into a Physical AI stackAn autonomy architecture change, not a model release.
- unawair v3, a rule engine after JevTrying Jev led to a rule-based engine with better temperature response.
- Voice control of a MacThe task runs as soon as the command is spoken.
- Voice control over Fusion 360Jev decides only one thing: was that a command or chatter?
Jev for UI
Interfaces that pick the next screen, look or result from a catalogue. 8 builds
- Auto-tagging in the interfaceGustav Ekerot: the interface tags the item while you are still typing.
- Dynamic UI generated by JevJev detects the input intent and builds the interface on the fly.
- Hover explanations for about $0.006A cheap demo of Jev explaining what the pointer is over.
- jev-design-testA shadcn block generator with Jev picking the block.
- JevFormA form that branches by meaning instead of if-then rules.
- json-renderVercel Labs’ generative UI framework, with a Jev composer behind a flag.
- When a designer gets access to JevType a phrase, and the matching objects float up from the pile.
- X, made cuteJev restyles the X interface in realtime.
Apps and utilities built with Jev
Small apps and utilities that fit no narrower use case. 48 builds
- $0.042 per 1M input tokensOutput is free.
- $5 for 123M tokensA user runs out of free credits and is surprised by the cost.
- 12 million source updates every six hoursA research pipeline that used to cost over a million dollars.
- A canvas you control by pointing and speakingJev turns pointing and speech into drawing actions.
- A coding interview prep use caseJev for interview preparation.
- A community thread on Jev use casesBuilders compare what they use Jev for.
- A data center optimizerJev at the centre of a data center optimization build.
- A Downloads folder that sorts itselfA macOS app that files downloads by your rules, with no LLM besides Jev.
- A neural network pickerJev picks the network.
- A real-time ad blockerJev checks every element on the page and removes the ads.
- A site scanner built with Opus 5.5 and JevOpus writes the app, Jev makes every decision.
- A walkable 3D street view, classified by JevSF and Bay Area startups and VCs, sorted by Jev.
- An always-on assistant with no wake wordJev tells a command for the computer from ordinary talk.
- An efficient chatbot with JEVA developer shares the build.
- An open-source chat copilotBuilt at the JEV hackathon.
- Copy and paste, with a decision in betweenWhat you copied and where you are pasting decides what arrives.
- EmogeeA Jev example from youens.
- evokeSoftware by reflex: a sentence picks a small program and runs it.
- Finding errors in AI draftsJev used to spot potential mistakes in a draft.
- fork-castWhat to eat, decided without any labels, on probabilities.
- Fun apps on a Jev backendA developer’s small builds.
- GIF DeciderA GIF picker with Jev making the call.
- Guess the age from a Korean nameA small React app where Jev guesses an age band from a name.
- guessworkFish-style zsh autosuggestions, ranked by Jev instead of by prefix.
- Jev apps and how they use confidenceAn observation on confidence in Jev builds.
- Jev builders, share what you’re working onAn open call to the Jev community.
- Jev CalcA calculator notebook that understands plain-language terms.
- Jev Doors takes uploaded genresWebMCP support lets an agent generate them by itself.
- Jev for real-time speechUse cases for speech, decided by Jev.
- Jev in a chat, no installTry Jev online without setting anything up.
- Jev OmniA demo showcase.
- Jev’s prompt format and inference codeA guide to the model’s prompt format and how to run inference.
- jevguideCurated Jev showcases from X, by category, with media previews.
- JevinciPaintings made from Jev’s pixel probabilities.
- Kill My IdeaDescribe a startup idea. Jev says kill it, fix it or ship it.
- live-jevControl Ableton Live with one short sentence.
- Port CleanupA macOS utility that tells you which listening ports are safe to stop.
- Predictive spreadsheetsName a column and Jev fills in every row, about 100 ms each.
- Put Jev in a chatTry it online, with open-source code.
- Real-time ClippyA Clippy that wakes up only when it thinks you are stuck.
- Real-time sentiment with Jev and ElevenLabsJev decides, ElevenLabs speaks.
- Scoring educational video segmentsJev scores segments of educational video.
- ST-jevedA SillyTavern extension that measures each reply before it lands.
- talkrSpeak for 30 seconds, and Jev scores the pauses and filler words.
- The unglamorous Jev jobsPerformance checks after a ship, chat compaction, prioritising work.
- Twenty Jev builds in one threadA curated roundup of twenty builds, each with its skills.sh link.
- TypeSafe AI playground110 use cases, games and dilemmas, with editable prompts and A/B tests.
- YouTube sponsor skipperA Chrome extension that detects sponsor segments and skips them.
Jev for social feeds and slop detection
Scoring posts, filtering timelines and catching slop. 36 builds