Awesome Jev

A curated list of tools, integrations, and experiments built on Jev, the System One model from TypeSafe AI that makes fast, typed, confidence-aware decisions.

Jev doesn't write paragraphs. You give it a question and a set of candidates, and it returns a typed answer with a confidence score, in milliseconds and for fractions of a cent. That makes it a good fit for work that needs many small decisions in a loop: picking the next click in a browser or on a phone, routing a task to the right model, classifying documents, scoring code changes, or deciding whether to trade.

No waitlist needed: Jev is available through Vercel AI Gateway as typesafe-ai/jev.

Within each section, projects are sorted by GitHub stars. Counts are refreshed weekly by a workflow. Stars belong to the whole repository, including projects where Jev is one optional integration.

• Official resources
• Frameworks & integrations
• Browser & computer use
• Mobile & robotics
• Coding agents & developer tools
• MCP servers & agent skills
• Observability
• Data, search & classification
• Trading
• Games & fun
• Apps with Jev inside
• Community SDKs
• Open models & replications
• Other lists

• TypeSafe AI - Product site for System One models and Jev.
• Documentation - Guides, SDK references, patterns, and the HTTP API.
• Quick start - From an API key to your first typed decision in Python or JavaScript.
• Primitives - Choice, Score, and Noul, and when to use each.
• Patterns - Confidence-gated routing, composite scoring, speculative fan-out, intent routing.
• TypeSafe Console - API keys and live request inspection.
• Workflow evals - Published workflows, model comparisons, and methodology.
• skills - Official agent skills for designing TypeSafe workflows from Claude Code, Codex, and similar agents. ⭐ 2.7k
• system-one-adapter-python - Drop-in TypeSafeClient replacement backed by regular LLM APIs, handy for local testing. ⭐ 390
• typesafe-sdk-python - Official sync and async Python SDK. ⭐ 283
• typesafe-sdk-js - Official TypeScript/JavaScript SDK with inferred answer types. ⭐ 275

• ComposioHQ/composio - Python and TypeScript providers that let Jev select tools and bind supported arguments, with partial-call and abstention results. ⭐ 30.5k
• vercel/ai - AI SDK's TypeSafe provider exposes Jev through the experimental evaluation API for typed Choice, Score, and Boolean questions. ⭐ 27.3k
• 0xPlaygrounds/rig - Rust agent framework with an experimental rig-typesafeai crate for typed Jev judgments. ⭐ 8.9k
• ax-llm/ax - DSPy-style framework with TypeSafe/Jev support for boolean and class signatures, plus a native client for probabilities and scoring. ⭐ 3k
• agentjido/req_llm - Elixir library with typed Jev evaluation through TypeSafe or OpenRouter, retaining probability distributions and provider responses. ⭐ 596
• ash-project/ash_ai - Maps Ash action arguments and return types to Jev evaluation requests through ReqLLM. ⭐ 190
• donvito/ai-backends - API server with a dedicated Jev evaluation endpoint and an interactive playground for typed decisions. ⭐ 146

• trycua/cua - Open-source computer-use platform with cross-OS drivers. Its jev-use example lets Jev choose the next action. ⭐ 29.5k
• browser-use/jev-ultrafast - Browser agent that uses Jev to pick target elements and only calls a small LLM when it has to type text. ⭐ 22.7k
• awlevin/typesafe-computer-use - macOS computer use for about $0.0002 a step: OCR the screen, let Jev pick the next click. ⭐ 1.2k
• wy-coliney/jev-browser-use - Jev handles the clicks while Codex plans and verifies, for 5-10x faster browser runs. ⭐ 1.1k
• moritzkremb/jev-voice-browser - Voice-controlled browser. Jev resolves intent and target in about 300 ms per spoken word, and Playwright executes. ⭐ 401
• kitze/unclutter - Browser extension that uses Jev to strip page clutter, with reusable template rules. ⭐ 369
• jkudish/jev-browser - Browser automation driven by Jev. ⭐ 332
• socai-io/socai - Browser and computer-use agent tuned for social media research and content extraction. ⭐ 243
• realZachi/typesafe-adblock - Fun Chrome extension that asks Jev "is this element an ad?" and removes it. BYOK, no backend, not a real ad blocker. ⭐ 89
• jcpsimmons/jev-macos-loop - Native macOS GUI automation on Apple silicon: OmniParser CoreML and Apple Vision OCR find the controls, and Jev picks the action. ⭐ 24
• valentynkit/jev-skip - Browser extension that sends YouTube captions to Jev, visualizes segment classifications on the seek bar, and auto-skips high-confidence segments by default. Requires a user-supplied API key; auto-skip can be disabled. ⭐ 7

• jev-chat/jev-chat-jarvis - Android chat assistant that uses Jev to assess intent and rank draft replies. Can fill the input field, but sending remains manual; chat context goes to configured providers. ⭐ 7.7k
• droidrun/mobile-jev - Agent that drives real Android devices with Jev. Includes live demos, a CLI, and execution traces. ⭐ 440
• RomanSlack/jev-drone - Camera-only autonomous drone in MuJoCo with Jev in the control loop at 2.5 Hz. ⭐ 268
• rokbenko/quackd - One CLI for many robots (Open Duck Mini, LeRobot, ToddlerBot, ROS bases). An LLM does the planning and Jev handles the cheaper steps. ⭐ 258

• tamaratran/fast-jev-compaction - Claude Code plugin that replaces compaction summaries with Jev decisions. Each tool call is scored, stale ones are dropped, and the rest stays verbatim. ⭐ 7.6k
• dzhng/jevgrep - CLI that uses Jev to find relevant files and source excerpts from natural-language queries. Sends selected source content to the configured hosted provider. ⭐ 2.6k
• vercel-labs/ai-cli - Terminal CLI whose ai evaluate command asks typed questions about stdin and returns JSON, using Jev through Vercel AI Gateway by default. ⭐ 824
• thruwire/foreman - Software factory "foreman" built on Jev. ⭐ 742
• devagrawal09/jev-review - Staged code review over git diffs or whole codebases, scoring correctness, safety, reliability, compatibility, and test risk in a local dashboard. ⭐ 683
• gargpratyush/jev-router - Routes each Claude Code or Codex turn: easy tasks go to fast models, hard ones to strong models. ⭐ 588
• coldteadotai/abide - Makes your coding agent follow your project rules: Jev checks every edit or turn against each rule in AGENTS.md and returns a probability. ⭐ 578
• ksenxx/kiss_ai - Agent framework whose optional Jev task classifier decides whether a request needs development work, with an LLM fallback. ⭐ 556
• BillionsBobby/JevRouter - Routes requests among models, tools, skills, and subagents using Jev's typed choices. Returns decisions for the host to execute, with CLI, SDK, and MCP interfaces. ⭐ 459
• notque/vexjoy-agent - Agent toolkit with a Jev-powered /d command for routing requests to agents, skills, and pipelines. ⭐ 441
• WrongStack/WrongStack - Coding agent with opt-in Jev skill suggestions and fleet-dispatch classification through TypeSafe or OpenRouter. ⭐ 374
• lakeday-org/perch - Semantic code linting with Jev. ⭐ 352
• SREGym/SREGym - Benchmark for incident-resolution agents with optional Jev review of Codex diagnostic tests and submissions, disabled by default. ⭐ 339
• yonatangross/orchestkit - Claude Code toolkit with opt-in Jev session classification and workflow routing, including shadow modes and fallbacks. ⭐ 292
• monotykamary/pi-fabric - Pi agent runtime with typed Jev judgments and foreground or background decision loops with explicit budgets. ⭐ 284
• 0xNatoshi/jev-codex-router - Per-turn routing for Codex: Jev picks the model, reasoning depth, and speed mode. ⭐ 273
• NiazMorshed2007/jev-review - Local-first MCP plugin for continuous quality review by AI coding agents. ⭐ 240
• DevMortimer/pi-warden - Guardrails for Pi: Jev flags irreversible or off-task tool calls, stuck loops, and unverified "done" claims. ⭐ 168
• y0usaf/pi-jev - Jev as a decision layer for the Pi coding agent, with a tool-call gate and typed jev_ask. ⭐ 163
• mrnugget/jev-shell-history - Fish-style zsh history autosuggestions ranked by Jev. ⭐ 119
• vinilana/jev-eval-agent - Evaluation agent built on Jev. ⭐ 106
• ellipsis-dev/blink - Codebase search powered by Jev. ⭐ 100
• IAmUnbounded/save-token-jev-clean - Portable context compaction for coding agents: Jev decides which tool calls still matter, and conversation text stays verbatim. ⭐ 80
• EliaAlberti/jev-rules - Claude Code plugin where Jev picks which of your rules apply to each prompt, so Claude only sees the relevant ones. ⭐ 66
• valentynkit/jev-belay - Claude Code Stop hook that reads the transcript for evidence and, only when files changed with no passing check since, spends one four-question Jev call before letting an unverified "done" stand. ⭐ 38
• suenot/codex-jev-router - Chooses Codex subagent model and reasoning effort using Jev Choice and Noul responses, with a Sol fallback. ⭐ 9
• valentynkit/jev.nvim - Neovim plugin that asks the buffer a plain-language question, splits it into functions with Treesitter, and ranks Jev's per-function scores in quickfix. ⭐ 8
• morcoan/JevSeek - Local coding workspace: Jev routes actions and DeepSeek generates arguments. Native tools, persistent sessions, React desktop app. ⭐ 3
• ReallyArtificial/jev-by-example - Runnable JavaScript lessons using Jev for memory reconciliation, tool-result checks, and handoff decisions, with offline fixtures and opt-in live calls. ⭐ 2

• jkudish/jev-mcp - Fast, cheap, typed judgments from Jev as MCP tools. ⭐ 521
• itsmostafa/typesafe-mcp - MCP connector that gives any agent direct access to Jev. ⭐ 349
• kitze/skillbox - Self-hosted, versioned skills library for AI agents with optional Jev recommendations. ⭐ 269
• dbreunig/building-with-jev-skill - Agent skill for writing and improving programs that call Jev. ⭐ 155
• Dicklesworthstone/skillranker - Rust CLI that ranks agent skills for the next step from live session context. Ships Claude Code hooks. ⭐ 132

• Arize-ai/phoenix - Observability platform with documented Python and TypeScript integrations for tracing TypeSafe System One calls through OpenInference. ⭐ 11.8k
• Arize-ai/openinference - Python and TypeScript instrumentation that records TypeSafe SDK calls as OpenTelemetry spans with inputs, outputs, model, and token usage. ⭐ 1.3k

• realZachi/pg-jev - PostgreSQL extension for asking your tables questions in plain language. ⭐ 1.1k
• superagents-lab/jev-search - Web search with Jev handling source selection, query understanding, and relevance ranking. ⭐ 530
• jerryjliu/docjev - Python library, CLI, and local app for document classification and splitting. Uses local LiteParse OCR and hosted Jev decisions, with optional cloud OCR. ⭐ 524
• kyotofin/tax-doc-classifier - Tax document page classifier. Reports 100% strict accuracy on 261 IRS forms at about $0.001 per page. ⭐ 505
• robbyczgw-cla/hermes-web-search-plus - Hermes search plugin with optional Jev checks for news intent, extracted-page quality, and language; disabled by default. ⭐ 426
• jexp/neo4jev - Jev navigates a Neo4j graph by classifying neighbouring relationships. ⭐ 173
• trungdq88/youtube-sponsor-detection - Detects YouTube sponsor segments from live audio and transcript. ⭐ 112
• ChetasLua/jevmeter - Scores every sentence of a video with Jev and renders a live meter on a 16:9 edit. ⭐ 108
• pithings/advocaat - Small, type-safe client for asking questions about your data with Jev. ⭐ 96
• giuliosmall/pg_typesafe - Pre-alpha PostgreSQL extension for categorical classification with Jev. ⭐ 88

• virattt/ai-hedge-fund - Educational investment research and backtesting with a native Jev adapter for investor decisions. The documented workflow does not place real trades. ⭐ 64k
• jarrodwatts/jev-trader - One buy/sell decision per Monad block on Kuru MON-USDC. Runs in dry-run/mock mode by default and makes no promise of profit. ⭐ 3k

Trading bots can lose money. Read the code and run in dry-run mode first.

• rmalde/minecraft-agent - Minecraft agent with an LLM planner and Jev selecting bounded Mineflayer actions. Runs against a locally configured server and requires paid model access and machine-specific setup. ⭐ 584
• fhshaik/typesafe-mario - Jev agent that plays Super Mario Bros. from structured emulator state. ⭐ 448
• dabit3/jev-experiments - Assorted Jev experiments. ⭐ 403
• standardagents/jevpilot - Playable Three.js driving simulator with a Jev-powered autopilot. ⭐ 220
• valentynkit/jev-plays-pokemon-red - Pokemon Red on PyBoy: code owns the route and the arithmetic, Jev picks only at branches, and every battle turn logs a faint prediction scored by Brier against what the RAM says. ⭐ 8

Larger products that use Jev for one part of the job.

• vercel-labs/json-render - Generative UI framework with experimental Jev-based composition from a component catalog. The integration is unreleased and requires a source build. ⭐ 18.6k
• Asymptote-Labs/agent-beacon - Cross-agent session history and memory. An opt-in command uses Jev to evaluate task success, reusable lessons, and supporting evidence; normal trace capture makes no Jev calls. ⭐ 1.8k
• anishfn/shapeshift - Text input that selects UI cards using typed Jev intent decisions. Uses a local keyword classifier by default; hosted Jev is optional. ⭐ 860
• CatCatUncle/openworkbuddy - Local-first AI office agent. Jev checks each acceptance criterion as a yes/no question in goal mode (README in Chinese). ⭐ 309
• MillionSend/millionsend - Open-source email platform on AWS SES. The hosted version scores outbound mail with Jev for content monitoring. ⭐ 227

Unofficial clients for languages the official SDKs don't cover.

• cequence-io/openai-scala-client - Scala client with a native TypeSafe System One module for typed questions and a closed-schema adapter for its OpenAI-style interface. ⭐ 255
• Twister915/typesafe-ai - Rust client with async and blocking backends and observable retries. ⭐ 15
• saibimajdi/typesafeai-dotnet-sdk - .NET SDK. ⭐ 12
• Tangerg/typesafe-sdk-go - Go SDK. ⭐ 9
• joshmn/typesafe-sdk - Ruby client. ⭐ 7
• jonesmelton/verdict - OCaml client built on Eio. ⭐ 0

None of these are the official TypeSafe model. They are independent projects that copy Jev's input/output shape with open weights.

• NandhaKishorM/laya - Fine-tuned open decision model, benchmarked against Jev's published numbers. ⭐ 32.5k
• jaredpalmer/kev - Tiny Jev-like model on Qwen2.5-0.5B that you can train and run on a MacBook. ⭐ 9k
• TheoLeeCJ/SemIf - "Semantic ifs" from open models on a single home GPU (3090). Formerly openjev. ⭐ 4.8k
• TianyuCodings/NanoJev - Nano replica of Jev with parallel decisions, dynamic candidates, and an end-to-end training pipeline. ⭐ 2.5k
• vinnylarouge/jevlike - Trains a Jev-like model on your own data to pick one option from a changing candidate set. More an interesting experiment than a practical tool. ⭐ 1.4k
• feder-cr/jev - jevos: independent local yes/no decision model with a Jev-compatible Noul API, served through llama.cpp with CPU support. The jevos binary model does not support Choice or Score questions. ⭐ 1.3k
• ollaya-dev/ollaya - Local runtime for downloading and serving open decision models, including Laya, through a TypeSafe-compatible API. Independent of TypeSafe. ⭐ 1.3k
• nokia-applied-research/AnyJev - Independent Jev-style decision library for open models, with Choice, Score, and Noul questions, logit readouts, and optional learned heads. ⭐ 1.2k
• receptron/laya - Node.js/TypeScript runtime for the independent Laya model using ONNX, exposing Jev-shaped typed questions and answers locally. ⭐ 900
• Rizzo-AI-Academy/rizzo-flow - Independent local decision server that scores open-model logits for Choice, Score, and Noul questions, with a TypeSafe-compatible System One endpoint. ⭐ 882
• featherless-ai/simple-jev - Turns any open model into a classifier / Jev-style endpoint. ⭐ 602
• ekzhang/openjev-sglang - Jev-compatible, prefill-only API endpoint built on open models with SGLang. ⭐ 333
• hr98w/jev-visual - Educational Jev-like visual inference experiment on Apple Silicon. ⭐ 317
• kshetrajna12/reflex - Small open decision model on Qwen3.5: state plus typed questions in, calibrated probabilities out. ⭐ 165
• iammrduncan/typesafe-ai-benchmark - LLM gateway that imitates TypeSafe's structured output, for benchmarking against the real thing. ⭐ 38
• SamratDuttaOfficial/WaterSheep - Independent open-weight decision model fine-tuned from ModernBERT-base that answers Noul, Choice, Score and multi-label questions with a probability per option, through a TypeSafe-compatible /v1/systemone server or in the browser with ONNX. ⭐ 12

• yibie/awesome-jev - Public projects, integrations, and discussions. ⭐ 2.3k
• Anil-matcha/awesome-jev-by-typesafe - Use cases, patterns, prompts, and starter code. ⭐ 912
• AnotiaWang/awesome-jev - Applications, libraries, and resources for Jev and System One. ⭐ 640
• AbdelStark/awesome-typesafe - Official resources and community projects for TypeSafe and System One models. ⭐ 581
• cobanov/awesome-jev - Source-backed list of projects built with Jev. ⭐ 543
• fatwang2/awesome-jev - Project directory plus a reusable Jev-based GitHub review workflow. ⭐ 225

Found something good? Open a pull request. See CONTRIBUTING.md.

This list is released under CC0 1.0. This project is not affiliated with TypeSafe AI.