Skip to content
Jev Model Guide

Jev GitHub projects and official SDKs

Jev’s launch produced a flood of GitHub projects. We currently list 276 repositories by category. Earlier entries retain their original verification basis; new entries follow the public repository’s automatic inclusion rules. Stars and repository status are refreshed as dated in the table.

Each row shows its latest verification date; stars reflect attention only. “Official” is reserved for first-party repos under the typesafe-ai org; replicas reimplement the interface idea, not Jev’s weights.

ProjectCategoryRelationshipLicense★VerifiedSummary
GaganSD/pi-extensionsagentCommunity integrationMIT312026-10-29MIT Pi extension collection whose pi-web-search can opt into Jev judging, reranking and suppression of cited candidates, separately from retrieval. The host owns model auth and all candidates go out first; no installation, auth access or testing of other extensions/performance.
philrenaud/practical-dataappCommunity integrationMIT32026-11-05MIT table-color extension using Jev for comparability/direction and ordinary code for parsing/painting. Native own-key, author-hosted and Clef paths are distinct; table text leaves the browser. No install, private tables, store-version acceptance or author-metric reproduction.
Tom-R-Main/FootworkagentCommunity integrationMIT22026-10-30MIT computer-use framework combining Jev completion/per-requirement Nouls with local evidence gates and mechanical-action advice under host policy. Native/browser qualification and builds are scoped. No drivers, traces or reliability reproduction; probability is not authorization.
dengyie/decidexagentCommunity integrationMIT22026-10-31MIT community game-decision framework with a native Jev adapter, legal-action gates, heuristic fallback and a 2048 execution example. No gameplay, account-pool access or validation of industrial/speed claims. Mock, other-provider and non-Jev turns remain distinct; quota evasion is not encouraged.
uint4096/hndelegateappCommunity integrationGPL-3.0—2026-10-29GPL-3.0 HN curator batching native Jev Choice/Noul tree questions before gated Telegram delivery. Trees/SQLite require setup; no database/channel connection or independent schedule/quality test.
theroich/opencode-jev-pruneragentCommunity integrationMIT—2026-10-29MIT OpenCode V2 context plugin using two-stage Jev Nouls to prune server/tool/skill descriptions, retaining requests on failure. Not execution authorization; proxy/host untested and fake-jev is not the real model.
lorenzorama/juger-vite-bloquer-justeappCommunity integrationMIT—2026-10-29MIT mail-agent talk demo with Jev ingress/egress Noul gates, shadow/enforced modes and Score evaluation. It creates local drafts without sending mail. Author results are not independent security validation; no user mail or replay execution.
bomliu/decision-driftresearchCommunity integrationMIT—2026-10-29MIT decision-model drift tooling using frozen/repeated, shuffled and paraphrased probes through OpenRouter to report noise, changes and calibration. Data has separate terms/private splits. No private items or ranking reproduction; periodic concurrent budget checks are not a hard cap.
gotalab/s1flowresearchCommunity integrationApache-2.0—2026-10-30Apache-2.0 decision-model evaluation/policy tooling with native TypeSafe wiring, typed conversion, explicit allowed connections and policy gates/feedback. Coverage/accuracy are author samples, not universal claims. No private data, skill installation or model execution.
Vlad0n1m/youtube-jev-filterappCommunity integrationMIT—2026-10-30MIT Manifest V3 extension judging video cards with two native Jev Nouls and a substance Score, then locally hiding/dimming and caching; Shorts handling is local. Real-feed calibration, false removals and cache aging are unverified; no install or personal history.
HangChi/JevRouteragentCommunity integrationMIT—2026-11-01MIT agent framework separating native four-question Jev routing, tool-risk policy, human confirmation, completion verification and bounded execution. Web-search/mail tools are local mocks and fifty-three-case metrics are author reports. Pinned source only; no real-tool grants.
AS214933/astrbot-betterchatappCommunity integrationMIT—2026-11-01MIT AstrBot plugin gating group participation with three native Jev Nouls and local rules. Private merging/interruption is local. Failure defaults to no unsolicited reply with explicit-wake exceptions; streaming and reload limits apply. No install or chat access.
charles-forsyth/jev-decision-wallappCommunity integrationMIT—2026-11-02MIT parallel-Noul experiment rendering two hundred questions as heatmaps, gated readouts and change observations, with recorded/live modes. Missing answers become zero. Author latency is not site measurement; no demo or text submission.
ChanghaoLiao/jev-rpa-assistantappCommunity integrationMIT—2026-11-02MIT Windows support prototype combining vision transcription and hierarchical Jev Choice over fixed replies, with separate authorization/page checks. Xianyu/Edge has historical checks; 1688 is not adapted. Offline demos differ from real sends. No drivers, chats or messages here.
Meerav29/inbox-sweepappCommunity integrationMIT—2026-11-03MIT local mail-batch app using Jev Choice/Noul signals for local auto-archive, filing proposals or review cards. Fake, dry-run, live and Apply rechecks are distinct; dry run still sends content externally. Source review only.
BronOS/cc-cockpit-jev-triageagentCommunity integrationMIT—2026-11-03MIT cc-cockpit source-plugin triage layer: hints/rules precede the existing Jev plugin grid, followed by human overrides and floors. Service calls and consumers reviewed; upstream network integration unqualified, so no full actual-Jev case. Model changes require rescoring.
yy1588133/dsh-sensitive-guardagentCommunity integrationMIT—2026-11-05MIT DSH decision plugin, now packaged as dsh-decision-guard, for optional command review, tool/skill pruning and path ranking. Jev wiring, Noul masks and tool_search restoration were reviewed. Features default off and failure/host-permission fallbacks remain; it does not replace permissions or sandboxing. Not installed or used to reconfigure tools.
omniaura/toolreviewagentCommunity integrationMIT—2026-11-06MIT catalog-review library using Jev Choices and local thresholds for approval classes enforced by the host. Reviews are per tool, with provider-label and stale-verdict fallbacks; it does not replace permissions. No live model/tool run.
cyriusweng/jev-ternagentCommunity integrationMIT (with inherited browser-use/jev-ultrafast notices)—2026-11-06MIT-derived Tern browser loop with reviewed DOM Choices/Nouls and action consumption; a separate chat model supplies text. Box-preview compatibility, completion heuristics and author performance remain untested. No install or key access.
typesafe-ai/skillsOfficial SDKs & toolsOfficialMIT1,4722026-09-22Official agent skills teaching coding agents such as Claude Code and Codex how to call the System One API.
typesafe-ai/system-one-adapter-pythonOfficial SDKs & toolsOfficialMIT2262026-09-22Official adapter that emulates TypeSafeClient on top of ordinary LLM APIs for A/B testing or fallback.
typesafe-ai/typesafe-sdk-jsOfficial SDKs & toolsOfficialMIT2062026-09-22Official TypeScript/JavaScript SDK (npm package @typesafe-ai/sdk) with inferred answer types, built-in retries and error classes.
typesafe-ai/typesafe-sdk-pythonOfficial SDKs & toolsOfficialMIT1782026-09-22Official Python SDK (pip package typesafe-sdk, Python >= 3.10) with sync/async clients and Choice/Score/Noul types; reads TYPESAFE_API_KEY by default.
typesafe-ai/n8n-nodes-typesafe-aiOfficial SDKs & toolsOfficialMIT22026-10-01TypeSafe official n8n community node calls System One Choice, Score, and Noul per item; Evaluate attaches typed answers and Route dispatches items by answer or confidence.
typesafe-ai/typesafe-public-examplesOfficial SDKs & toolsOfficialMIT12026-10-07Official documentation, demo and cookbook materials, with public fixtures and Python dispatch code that uses Choice/Noul for functions and finite arguments before code executes them. Trading charts use synthetic bars; this is not a complete standalone package and was not run here.
browser-use/jev-ultrafastBrowser / computer useCommunity integrationMIT14,8202026-09-22Browser Use's fast web agent: Jev picks each action and target element while a small LLM only types when needed. The most-starred Jev project of launch week.
awlevin/typesafe-computer-useBrowser / computer useCommunity integrationMIT1,0442026-09-28A macOS computer-use beta that asks Jev Choice to select actions and controls from screen state, then executes and checks the next frame; it drives real input, so start with a dry run.
droidrun/mobile-jevBrowser / computer useCommunity integrationMIT4222026-09-29droidrun's Android agent where Jev drives a real phone via the Mobilerun API, with a React studio, CLI, and per-request latency traces; the Uber demo does 9 actions in ~21 s and stops at payment.
moritzkremb/jev-voice-browserBrowser / computer useCommunity integrationMIT3652026-09-29Voice control for a real browser: each partial transcript fires one 9-11 question Jev request (~300 ms) to act, wait, ask, or ignore; text is extracted by code and only picked by Jev.
kitze/unclutterBrowser / computer useCommunity integrationMIT3372026-09-29A WXT browser extension where Jev spots page clutter for removal, with reusable template rules.
jkudish/jev-browserBrowser / computer useCommunity integrationMIT2212026-09-22Browser automation driven by Jev.
openqa-cn/jev-browserBrowser / computer useCommunity integrationMIT1042026-09-29A CodexQA skill that replaces vision GUI models with an in-page index: Jev chooses only among indexed controls while Playwright acts; replay, goal runs, and site exploration share the index.
realZachi/typesafe-adblockBrowser / computer useCommunity integrationMIT872026-09-29A fun Chrome extension where code pre-filters ad-shaped DOM candidates and one Jev noul per candidate decides "is this an ad?"; elements above 0.70 are removed. The author calls it a toy, not uBlock.
Ice-Hazymoon/jevwrightBrowser / computer useCommunity integrationMIT122026-09-30Browser business-flow tests: Jev initially grounds controls and records semantic paths for model-free replay; code decides verdicts from requests and API/database assertions.
sedum-dev/sedumBrowser / computer useCommunity integrationMIT122026-10-15MIT pre-alpha natural-language/TypeScript browser tests using Jev by default for bounded element selection and page claims, with Playwright execution, freshness checks and probability/contradiction flags. CI exit policy is separate; marginal results may pass with flags. Clef is optional; author savings/timing were not reproduced and no third-party sites were tested here.
Friedjof/jev-mobileBrowser / computer useCommunity integrationMIT72026-09-28Uses Jev Choice to select the next valid Android action, with SQLite task state, a mutation journal and post-action observation; requires USB debugging and an external API.
brnyxx/jev-raBrowser / computer useCommunity integrationMIT52026-09-27A browser MCP server and CLI for coding agents: Jev selects an action and element each step, and the client validates the reply before driving Chrome. Speed benchmarks are author-reported.
AndresCarreonDiaz/jevviumBrowser / computer useCommunity integrationMIT12026-10-01Explores Appium paths from mobile acceptance goals with Jev choosing on-screen actions, then generates fixed-selector WebdriverIO tests and replays without a model; goals without assertions need review.
andletenkov/decisiumBrowser / computer useCommunity integrationMIT12026-10-27MIT Gherkin/Playwright testing with native Jev choices over controls/actions, code-comparable assertions and separately marked subjective judgments. Cassette replay makes no model calls. Live steps/page text leave the device; no installation or third-party flow verification.
laihenyi/pi-Jev-browserBrowser / computer useCommunity integrationApache-2.0—2026-09-24A pi browser and macOS desktop agent extension: Jev chooses one action from visible text, controls and the goal, within a bounded loop with human-review gates.
kimjisub/ego-jevBrowser / computer useCommunity integrationMIT—2026-10-12Multi-step decision layer for ego lite: Jev selects bounded actions, elements and caller-provided values, while code acts and rechecks requirements. MIT with inherited jev-browser code; requires macOS/ego. Author paired examples are not general reliability and were not run here.
honeybankstudios/jevblockBrowser / computer useCommunity integrationMIT—2026-10-17MIT MV3 content blocker using Jev over host/site/resource type and user policy beyond seed lists, compiling dynamic rules with user overrides first. First requests may pass, errors fail open, parsing fallbacks are permissive and spend limits are soft. Host/site metadata reaches OpenRouter and external lists have separate licenses. Firefox untested; not installed or applied here.
rmalde/minecraft-agentEmbodied & game controlCommunity integrationunknown5712026-09-29A vanilla-Minecraft speedrun agent where GPT-6 Astra plans and Jev picks each action; best verified run 8m43s with 131 Jev decisions and no rule changes.
fhshaik/typesafe-marioEmbodied & game controlCommunity integrationunknown4212026-09-29An experimental NES Super Mario controller that feeds structured JSON state (no screenshots) to a Jev Choice over seven legal inputs; ships no ROM.
RomanSlack/jev-droneEmbodied & game controlCommunity integrationMIT2252026-09-29A MuJoCo quadrotor flies a five-station course from its onboard camera alone; Jev judges tactics at 2.5 Hz while 50 Hz reflexes and a 500 Hz controller keep the veto.
wingedsheep/argentum-engineEmbodied & game controlCommunity integrationMIT702026-10-06A Jev opponent in a Magic: The Gathering engine: local development mode sends a player view and legal actions to OpenRouter Choice, then validates, corrects, or falls back. This does not establish complete rules coverage or independently measured win rates.
mrubens/jevcraftEmbodied & game controlCommunity integrationMIT32026-10-13MIT Mineflayer game experiment: code builds feasible action menus and survival/food/work claims for Jev to judge, then acts and records failures. Most questions wait during outages, with bodily-physics exceptions. The author has not beaten the game; costs, timing and trials were not reproduced, and pinned source takes precedence over older README descriptions.
jevplays-games/jev-factorio-agentEmbodied & game controlCommunity integrationMIT22026-10-05A Factorio macro-decision agent: Jev answers Choice/Noul/Score questions on goals, next actions, stuck state, and urgency; code filters impossible actions and falls back on low confidence. Offline mock is not Jev inference.
Saik0s/axe-decision-controlEmbodied & game controlCommunity integrationMIT12026-10-15MIT iOS Simulator research controller comparing accessibility, screenshot grids and Jev 1.14 Boxes, gating presence/confidence/freshness before AXe actions and separately checking persisted outcomes. Synthetic FocusBench and a few author recordings do not establish general app accuracy; Box receives an extra target catalog. Requires macOS/Xcode/AXe and experimental vision APIs; not run here.
statico/jev-nethackEmbodied & game controlCommunity integrationMIT—2026-10-01A NetHack 5.0 agent where code reads the terminal and enumerates legal actions, Jev Choice picks one, and a motor sends the corresponding keys; runs are documented but ascension is unverified.
tripodxu/board-gamesEmbodied & game controlCommunity integrationMIT—2026-10-05A seven-game board-play site where Jev scores legal moves with Choice probabilities and a deterministic tactics layer can override them. Mock, random, and Rapfi modes also exist; results were not reproduced.
vinicius-francozo/railroad-routeEmbodied & game controlCommunity integrationMIT—2026-10-08Mine-cart puzzle where one player sentence drives batched Choice/Noul/Score switches, then backend code simulates and scores the route. Includes five levels, reference solutions and replay tests; author API and timing results were not independently measured.
Keyizhan/jev-fabric-annotatorEmbodied & game controlInspired replicaMIT—2026-10-22MIT dual-arm fabric-sorting video annotation pipeline combining local compatible-model shape votes with geometry to draw basket boxes until release. Defaults to a local StartLux/jev-compatible service, not verified hosted TypeSafe use. Weights/robot data are excluded; weights have separate non-commercial restrictions. Not run.
tamaratran/fast-jev-compactionCoding agent pluginsCommunity integrationMIT5,8452026-09-22Claude Code plugin that replaces compaction summaries with Jev decisions on every past tool call; kept content stays verbatim. The approach is debated in the community.
kerpopule/hermes-jev-skillsCoding agent pluginsCommunity integrationMIT1,1302026-10-16MIT community Jev skills/agent integrations for routing, retrieval, choice and separately disabled vision shadow evaluation. Mainline cloud image checks use the experimental header and record bounded-answer comparisons without changing tool results; screenshots are not automatically redacted and need explicit opt-in. Author branch action demos and local Jev-Omni are separate, not official weights. Not run here.
kitfunso/hippo-memoryCoding agent pluginsCommunity integrationMIT7702026-10-04An optional Jev reranker for a local agent memory store: one request scores up to 40 candidate memories with Nouls and falls back to a local cross-encoder. Author ranking tests do not establish better final answers.
thruwire/foremanCoding agent pluginsCommunity integrationMIT6082026-09-29A software-factory foreman: Jev continuously scores Codex/OpenCode workers on completeness, requirements, and test sufficiency while code decides to continue, steer, or stop.
devagrawal09/jev-reviewCoding agent pluginsCommunity integrationMIT4602026-09-22A staged code-review workflow with a local dashboard.
lakeday-org/perchCoding agent pluginsCommunity integrationMIT3502026-10-23MIT semantic code-checking CLI with an optional native TypeSafe System One backend. Code consumes probabilities into issues, rule violations and coverage reports. Default cloud routing supports other models; benchmarks must distinguish backends. No code uploaded, installation or recall retest.
gargpratyush/jev-routerCoding agent pluginsCommunity integrationMIT2992026-09-22Claude Code plugin that routes each task to the cheapest sufficient model.
dzhng/jevgrepCoding agent pluginsCommunity integrationMIT1822026-09-27A coding-agent CLI that uses Jev to judge the relevance of folders, files and source units, then returns line-referenced excerpts; eligible source content may be sent to an external provider.
tamaratran/jev-prunerCoding agent pluginsCommunity integrationMIT1532026-09-29A Claude Code plugin that prunes noisy Bash output after the command runs but before the LLM sees it, one Jev noul per chunk, with documents, errors, and git diff exempt.
mrnugget/jev-shell-historyCoding agent pluginsCommunity integrationunknown1152026-09-29Fish-style zsh autosuggestions where a Jev Choice ranks which of your last 100 distinct commands you are completing, shown in grey with its probability.
GhalebDweikat/winnowCoding agent pluginsCommunity integrationMIT1002026-09-29A calibrated context sieve for Claude Code: every large tool result is split and judged block-by-block; confident-no blocks become stub-plus-summary with a recall key, while uncertain ones stay.
QAInsights/jmeter-aiCoding agent pluginsCommunity integrationMIT652026-10-28MIT Feather Wand JMeter plugin with optional native Jev tool-family routing, confidence cards and full-tool fallback, followed by a separate editing/diagnosis agent. Routing defaults off and is not authorization; other backends are not Jev. No installation, load test or paid run.
jonathanavis96/jev-kitCoding agent pluginsCommunity integrationMIT552026-10-02A Jev kit for Claude Code with tool-call guards, task-tier selection, search, and browser integrations; `airlock/client.py` calls official System One, can fail open on model errors, and has author-reported performance.
darkmatter/adhereCoding agent pluginsCommunity integrationMIT92026-10-17MIT semantic-rule checker using Jev by default for Noul violation/scope judgments and Choice section/line location, with thresholded reports and caches. Deterministic rules belong in conventional linters; probabilities/locations may be wrong, no automatic source fixes occur and source/context reaches providers. Other providers are separate; author results not reproduced and not run here.
ckorhonen/jev-lintCoding agent pluginsCommunity integrationMIT62026-10-04A post-edit fuzzy rule check for coding agents: TypeSafe Jev uses Nouls to flag team-rule violations for agent repair. Clef is optional, failures pass through, and author results and costs were not reproduced.
disler/ten-levels-of-jevCoding agent pluginsCommunity integrationMIT52026-09-28A progressive code lab from one bounded Jev decision to agent hooks, with live TypeSafe/OpenRouter calls, an offline mock, demos and tests; speed and cost figures are author-reported.
Tech-Byte-Frontier/jevgateCoding agent pluginsCommunity integrationApache-2.0 OR MIT32026-09-24Rust code-review tool that asks Jev typed questions about parsed source units, then composes located findings and suggested actions; security rules are opt-in.
SASUKE40/harness2Coding agent pluginsCommunity integrationApache-2.022026-10-21Community Harness² fork with native Jev final-output scoring/selection and aircraft/road-sign vision demos. Jev supplies typed judgments while the surrounding framework refines prompts/tools. Consumer and root license reviewed, not run; not an official Google/TypeSafe Jev integration and photo terms are separate.
Lexus2016/fast-jev-compactionCoding agent pluginsCommunity integrationMIT12026-10-23MIT maintenance fork of tamaratran’s compactor with parallel-result preservation, long-session windows, redaction and timeout/fallback handling. Client/hook reviewed; calls remain by default unless explicitly dropped. No fork npm release, installation or retest; not counted as a duplicate case.
rodrigojager/pi-workspace-searchCoding agent pluginsCommunity integrationMIT—2026-10-03Bounded source search for Pi: ripgrep handles exact/regex, a local index is optional, and explicitly enabled TypeSafe classification ranks current snippets with line references and partial-coverage markers; it does not prove exhaustive absence.
DAXZEIT/Pi-JEV-VCC-memory-relevanceCoding agent pluginsCommunity integrationMIT—2026-10-04A Pi coding-agent memory relevance extension that scores each indexed memory with a Noul and gives the agent ranked hints, using hosted Jev through OpenRouter or a local compatible CLI; ranks need review.
hraness/sys1Coding agent pluginsCommunity integrationMIT—2026-10-05A System One toolkit for coding agents with a CLI, gateway, and typed client; hosted Jev can check diffs and completion claims when explicitly enabled. Preview outbound content.
justmytwospence/opencode-lean-contextCoding agent pluginsCommunity integrationMIT—2026-10-06An OpenCode tool-output trimming plugin: Jev chooses relevant lines using task context, with code prefiltering oversized results. Unavailable or uncertain decisions preserve output; text may leave the device, and author results were not reproduced.
zereight/pi-zereight-routerCoding agent pluginsCommunity integrationMIT—2026-10-08Jev-backed virtual router for Pi that selects backend models and effort by work phase, hands off after a file mutation and falls back to planning on classification failure or uncertainty. Requires OpenRouter and Pi/Cursor integration; account compatibility and cache benefits were not tested here.
jomi-se/system-one-deslopCoding agent pluginsCommunity integrationMIT—2026-10-14MIT technical-writing style research tool using per-sentence Jev Nouls for density, intervals and a 13-post human baseline; experimental editing uses generative models separately. Style metrics do not identify authorship or factual quality. Requires Node 26/API configuration; author quality/cost claims were not reproduced here.
comchatx/cx-jev-tidyCoding agent pluginsCommunity integrationMIT—2026-10-15MIT Pi context-tidying package porting fast-jev-compaction two-Noul logic through a ComChatX loopback proxy to propose context_edit entries. User/assistant prose and original sessions remain; under-25% savings or failure proposes nothing. Requires that platform proxy and sends relevant context to the model. Not run here; its case meaning duplicates an existing compaction entry.
kurowashi/pi-jevCoding agent pluginsCommunity integrationMIT—2026-10-18MIT Pi extensions for content and new-file placement, wiring Noul answers to edit/write blocking; errors allow by default, bash is outside scope, and no model was run.
flupkede/fast-compactionCoding agent pluginsCommunity integrationMIT—2026-10-18MIT privacy-oriented derivative of fast-jev-compaction, adding secret-shape redaction and optional stable identifier substitution before judging. Cloud calls can use Jev; replay defaults to local Ollama. Pattern scrubbing is not a confidentiality guarantee; the base compaction case is deduplicated.
must-lioncho/jev-memCoding agent pluginsCommunity integrationMIT—2026-10-19MIT memory-selection adapter retrieving from local claude-mem before native Jev Nouls and threshold-gated injection. Defaults to one best record and no injection on failure; prompts/summaries leave the device. No hooks installed or user memories read.
mud-the-developer/jev-loopCoding agent pluginsCommunity integrationMIT—2026-10-19MIT source development version derived from pi-loop, combining advice from a separately configured public jev_decide tool with a fixed verifier for bounded-loop completion. Requires a backend and Pi nested tools and runs while Pi is active; live-model/terminal acceptance is unverified, so no end-to-end Jev case is claimed.
shivaylamba/jev-model-router-for-coding-tasks-ossCoding agent pluginsCommunity integrationMIT—2026-10-20MIT coding-router experiment with pinned tools/catalogs, nconnect, and hidden tests, separating all 50 tasks from the post-selected 32 saving cases. Requires external pinned tools and paid APIs; not independently reproduced. The author article is already a case, so no duplicate case is added.
PeesukaGu/jev-adviseCoding agent pluginsCommunity integrationMIT—2026-10-22MIT Claude Code plugin displaying jev CLI probabilities beside bounded options; users choose by default, with explicit auto-select available. Hook/consumer reviewed without installation; author preference-matching rates are unverified and selected context is transmitted.
jpcamara/judgelintCoding agent pluginsCommunity integrationMIT—2026-10-24MIT Ruby/native CLI turning team conventions into paired Nouls with line-range findings and CI results. Native TypeSafe reviewed; other backends remain separate. Refusals/skips are explicit. Judges/rules can still err; no source uploaded, binary installed or tests rerun.
House-elves/pr-wallCoding agent pluginsCommunity integrationApache-2.0—2026-11-02Apache-2.0 Java PR wall asking native Jev about runtime change, test reach and assertions, showing probabilities and local feedback without composite scoring or GitHub writes. Static judgment is not test execution; enabling sends diffs externally. No real PRs or execution here.
kevinyin9/next-steps-jevCoding agent pluginsCommunity integrationMIT—2026-11-04MIT terminal Claude Code plugin ranking model-proposed prompts with Jev Scores and filling drafts on user input. Conversation egress and early host API apply; timeout fallback does not prove cancellation. Source review only, not installed.
wuyoscar/jev-skillMCP & agent skillsCommunity integrationMIT5432026-09-29A collection of Jev use cases, workflows, and agent skills.
kitze/skillboxMCP & agent skillsCommunity integrationMIT2732026-10-25MIT self-hosted versioned skill library with optional scoped Jev Score recommendations. Candidate/version/grant changes invalidate results; failures explicitly fall back to search without execution. Rubric scores are not calibrated probabilities and task/descriptions leave the device. No install, private-library access or quality retest.
jkudish/jev-mcpMCP & agent skillsCommunity integrationMIT2132026-09-22MCP server exposing typed judgments (screening, filtering, ranking) as agent tools.
itsmostafa/typesafe-mcpMCP & agent skillsCommunity integrationMIT1872026-09-22MCP connector giving agents direct access to Jev.
aaddrick/building-with-typesafe-jevMCP & agent skillsCommunity integrationMIT1362026-10-08Unofficial Jev development skill covering bounded questions, confidence handling and code-owned workflows, with pattern sketches and community references. Separate from TypeSafe’s official skill; installation and author evaluations were not run here.
w3cj/jev-chatMCP & agent skillsCommunity integrationMIT1042026-09-29A chat command bar where no LLM writes a word: Jev picks the tool, argument sources, and confirmations while code calls MCP servers and builds replies from tool data.
Brainwires/jevwireMCP & agent skillsCommunity integrationMIT222026-09-28MCP tools, an embeddable DecisionModel and a coding-agent plugin for TypeSafe System One, with request validation and budget controls; plugin advice does not replace host permissions.
typakon4/jev-layerMCP & agent skillsCommunity integrationMIT72026-10-18MIT bounded-decision/MCP adapter with TypeSafe, image, and Box advice wiring. The default is offline demo; vision/desktop surfaces are opt-in and the host retains permissions and execution. Listed as a project, not accepted desktop automation.
Jev-Engineering/cookbookMCP & agent skillsCommunity integrationMIT42026-10-09Unofficial Jev notebook curriculum planning 60 recipes, with five published in the reviewed snapshot. Shares fixture, replay and explicit live SDK backends with mode/model provenance. Separate from the 18 official cookbooks; synthetic metrics are not live Jev results and notebooks were not run here.
waddle-zoo/signal-weaveMCP & agent skillsCommunity integrationApache-2.042026-10-20Apache-2.0 analytics-workflow MCP service combining reviewed cards, source evidence, and native Jev bounded policy judgments. It returns investigation/delivery handoffs and receipts while callers schedule and send messages. Other backends exist; no BI or messaging channel was connected here.
aktonay/snapdecMCP & agent skillsCommunity integrationApache-2.0—2026-10-04An MCP/CLI decision service for coding agents with a TypeSafe Jev System One backend for Choice/Noul/Score; low-certainty results return to the agent. Hardware and accuracy examples are author-reported.
celesteanglm/decision-models-skillsMCP & agent skillsCommunity integrationMIT—2026-10-08Six portable decision skills and a Python CLI for guardrails, routing, reranking, tool gating, confidence gates and output evaluation. The Jev/OpenRouter adapter validates typed responses and returns recommendations only. Initial thresholds are uncalibrated; test and cost evidence is the author’s.
dphbfs/fast-resume-scoringMCP & agent skillsCommunity integrationApache-2.0—2026-10-13Apache-2.0 resume self-check CLI/MCP: Jev provides bounded role-match and experience-gap signals for weighted scores and requirement-coverage evidence. Includes synthetic replay; author errors use a generative-model reference and costs/timing were not reproduced. Scores are not objective ability or hiring accuracy and do not make hiring decisions.
nocoo/falconMCP & agent skillsCommunity integrationMIT—2026-10-20MIT native macOS HTTP/MCP proxy and decision-observation tool with reviewed upstream forwarding and recording paths. Supports direct or self-hosted relay, replay, and export. Not installed or device-tested here; the unnotarized development build’s credential and local-record storage boundaries require user review.
BiteFoundry/switchbiteMCP & agent skillsCommunity integrationMIT—2026-10-23MIT Rust/MCP worker-assignment proposal tool using native Jev 1.14 judgments, deterministic eligibility filters and holds. Host owns permissions/execution; the tool launches no workers. Preview release validation is Windows x64 only; no live model run.
bitskc/jev-agent-kitMCP & agent skillsCommunity integrationMIT—2026-10-28MIT Worker MCP integration combining upstream Jev tools, bearer access and agent skills/hooks, with native TypeSafe/Gateway wiring. Vendor code must be built; guards fail open and some omp interfaces are author-unverified. No hook install, deployment or live 15-tool verification.
yetone/magpieRouting & gatewaysCommunity integrationMIT16,1482026-10-19MIT local AI gateway with System One request/response adaptation, Jev intent and effort judging, and decision-model group failover. Other judges are optional; source review does not establish Jev testing of every gateway feature or performance claim.
BillionsBobby/JevRouterRouting & gatewaysCommunity integrationMIT2882026-09-29Models, subagents, skills, MCP tools, and CLIs become one candidate set: Jev answers a single Choice while the router enforces availability, permissions, risk, and confirmation; decision-only by default.
0xNatoshi/jev-codex-routerRouting & gatewaysCommunity integrationMIT2772026-09-29Per-call model routing for Codex: Jev picks model and thinking effort together (continuations included) from a bounded decision state, failing open on any error.
vinilana/jev-gatewayRouting & gatewaysCommunity integrationMIT2522026-09-29A local gateway for coding agents that hands the "which tool next" decision to Jev while everything else flows to your usual LLM untouched; works with Codex, Claude Code, OpenCode, Kilo, and subscription logins.
yusukebe/hono-jev-routerRouting & gatewaysCommunity integrationMIT452026-09-22Hono middleware that routes HTTP requests by meaning.
Studio1-OSS/nconnectRouting & gatewaysCommunity integrationMIT72026-10-20MIT coding-agent protocol gateway whose Auto mode can explicitly select native Jev, another hosted model, or local Laya. Jev judges task difficulty; explicit user signals take priority and failures fall back to rules. Native client and routing consumption reviewed, without installation or cost reproduction.
Endgame-Labs/understudyRouting & gatewaysCommunity integrationMIT52026-11-03MIT System One proxy and narrow-task student experiment forwarding ungraduated requests to a configurable lead, then auditing/demoting local answers. HTTP wiring reviewed; training, graduation and author metrics unreplicated. Neither TypeSafe official nor open Jev weights.
guanghuang/coding-router-jevRouting & gatewaysCommunity integrationMIT22026-11-02MIT TypeScript/Bun coding router selecting model tier/effort through TypeSafe SDK under option/confidence/capability policy. Sensitivity judgments do not prevent sending original text. Savings, cache hits and Claude/Pi compatibility were not reproduced; no install or credentials.
gebeer/jev-effortRouting & gatewaysCommunity integrationMIT12026-10-06A Claude Code mod maps OpenRouter Jev Score to low/medium/high/xhigh effort. Shadow mode logs only by default; explicit activation changes effort. Recent conversation text leaves the device, and author runs were not reproduced.
prgrmrwy/dsh-auto-model-routingRouting & gatewaysCommunity integrationMIT12026-11-05MIT DSH plugin keeping the main model while Choice adjusts effort, with separate child capability/depth and catalog/allowlist gates. Host-entry limits, global-default synchronization and no automatic feedback learning apply. Source review only, not installed.
justmytwospence/opencode-auto-effortRouting & gatewaysCommunity integrationMIT—2026-10-06An OpenCode per-prompt effort plugin: Jev scores current and recent requests plus a prior-run summary, and code adjusts effort. Timeout or failure holds the level; token savings and answer quality were not verified here.
gbesse/alertmanager-jev-triageRouting & gatewaysCommunity integrationMIT—2026-10-07Prometheus Alertmanager relay that adds finite Jev triage annotations while forwarding every original notification. Low confidence, truncation or API failures become review; downstream failures return HTTP 503. The author exercised the API with synthetic input, without a live Alertmanager instance.
Starttoaster/reflexRouting & gatewaysCommunity integrationMIT—2026-10-13MIT experimental Alertmanager-to-Discord receiver: scrubbed summaries feed a Jev Choice for urgent/info channels, with a logged-only Score. Missing keys, low confidence and timeouts fall back to severity labels; resolutions reuse the channel. Sticky state is in memory, unsuitable as the only critical-alert path; not deployed here.
atvari-eu/opencode-system-oneRouting & gatewaysCommunity integrationMIT—2026-10-18MIT OpenCode System One plugin with bounded judgments and reasoning/model switching; a TypeSafe endpoint can be explicit, while default Console credential/version conditions were not independently accepted.
aen07/jev-grievance-triageRouting & gatewaysCommunity integrationApache-2.0—2026-10-23Apache-2.0 grievance-triage prototype using the official SDK, batched narrow questions, versioned thresholds and deterministic policy for routing/review and audit output. Risks/API failures route to people. Source only; no complaint caches or real rights adjudication inspected.
pbswimmer3/claude-jev-routerRouting & gatewaysCommunity integrationApache-2.0—2026-10-24Apache-2.0 Claude Code mod displaying Jev-based draft model/effort suggestions without switching settings. Merge is default, TypeSafe direct optional, and context is added only for continuation. Obvious-pattern scrubbing is not complete confidentiality. No install or retest.
u1-liquid/hermes-slack-agent-routerRouting & gatewaysCommunity integrationMIT—2026-11-03MIT Hermes Slack routing plugin using native Choices for response admission and exact source scope. Bound to a pinned host, external conversation processing and no inference elapsed-time cutoff; routing is not external-action authorization. Not installed or live-delivery tested.
xerj-org/xerjData / SQL / retrievalCommunity integrationApache-2.014,2352026-10-24Apache-2.0 local-search/MCP engine with explicit optional native Jev reranking. Hosted calls, result consumption and egress boundaries reviewed. Independent local judges are not Jev and probabilities are not calibrated thresholds. No install, index transmission or benchmark retest.
shcherbak-ai/contextgemData / SQL / retrievalCommunity integrationApache-2.02,0162026-10-27Apache-2.0 document extraction via LiteLLM Decisions, including Jev paragraph predicates and Boolean/label/rating concepts, thresholds, probabilities and model fallbacks. Reviewed agreement examples are not legal validation; no installation or model calls.
Arize-ai/openinferenceData / SQL / retrievalCommunity integrationApache-2.01,2542026-10-06OpenInference TypeSafe Python/JS instrumentation wraps System One calls and exports DECISION spans, model identity, and token usage. Input/output masking is available; default payload tracing needs privacy review. Stars belong to the parent repository.
traceroot-ai/tracerootData / SQL / retrievalCommunity integrationApache-2.0 core; proprietary EE8032026-10-16TraceRoot open core can detect trace problems through Jev Noul/Choice questions with state budgets, response validation and error results. Non-ee content is Apache-2.0; all ee directories use a commercial license, including enterprise signal grouping. Traces may leave the machine and truncation can omit evidence. SDK auto-instrumentation is separate; not deployed or called here.
kyotofin/tax-doc-classifierData / SQL / retrievalCommunity integrationApache-2.04832026-09-30An IRS PDF text-page classifier using Jev Choice for form and page kind with a confidence gate for routing; limited to English federal forms and not independently benchmarked here.
superagents-lab/jev-searchData / SQL / retrievalCommunity integrationMIT3632026-09-22Web search where Jev picks sources, interprets queries, and ranks relevance (built on Search1API).
realZachi/pg-jevData / SQL / retrievalCommunity integrationunknown2722026-09-22A PostgreSQL extension for asking plain-language questions over table rows. Its license is not recognized by GitHub as a standard open-source license.
sutro-sh/jev-alignData / SQL / retrievalCommunity integrationApache-2.02582026-09-22Calibrates Jev judgment functions from human feedback using GEPA.
fstandhartinger/jevbenchData / SQL / retrievalCommunity integrationMIT1762026-09-29A benchmark for Jev-class decision models with a live board of 95 systems (Jev 1.13 currently #4): frozen method, sealed decisions, and a SHA-256 manifest; unaffiliated with TypeSafe.
jerryjliu/docjevData / SQL / retrievalCommunity integrationApache-2.01652026-09-22From LlamaIndex: fast document classification and sub-document boundary detection with Jev.
jexp/neo4jevData / SQL / retrievalCommunity integrationMIT1522026-09-29A Neo4j navigation demo where each hop asks one Choice over outgoing edges plus a goal noul in a single round trip, then beam-searches the probabilities; schema is introspected live.
get-convex/convex-evalsData / SQL / retrievalCommunity integrationApache-2.01292026-10-06Convex decision-model evaluations publish a question bank and executable evidence, call native Jev questions, and retain probabilities, costs, and model identity. Some clean-environment replay remains incomplete; this site did not rerun the benchmark.
VectifyAI/jev-doc-searchData / SQL / retrievalCommunity integrationApache-2.0482026-10-03Searches long PDFs with a PageIndex section tree and Jev Choice over nodes/pages plus Noul answerability checks. No vector DB or embeddings, but both PageIndex cloud and TypeSafe API keys are required.
hev/rerankerData / SQL / retrievalCommunity integrationApache-2.0152026-10-05A Jev reranking library asking one Noul per candidate in batches of up to 30, then sorting or pruning by probability. BEIR results are author-reported and unverified here.
typesafe-ai/WorkflowEvalsData / SQL / retrievalOfficialApache-2.0122026-10-01Official workflow evaluation code for invoice, support, agent-trace, and security decisions; compares Jev and other providers on agreement with reference-model actions, cost, and time, which does not establish ground-truth business accuracy.
columnar-tech/jevaroData / SQL / retrievalCommunity integrationApache-2.0102026-09-30A Python batching proxy for TypeSafe Jev: calls the official API once per independent state concurrently, then streams ordered Apache Arrow results, with Python and JavaScript readers; not a native batch API.
vibefilter/filamentData / SQL / retrievalCommunity integrationMIT42026-10-10Laravel/Filament natural-language table filter using batched Jev Nouls over SQL-filtered rows, caching scores and applying a configurable threshold, with confirmation for large unscored sets. MIT plugin; author performance and cost were not reproduced.
SamuelSacco/jev-explorationData / SQL / retrievalCommunity integrationMIT32026-10-23MIT Jev evidence ledger/evaluation tooling with a native client and calibration, baseline and cross-domain research code. Conclusions are limited by samples, model versions and study design, not universal performance facts. No raw email corpora read, experiments run or metrics retested.
youdotcom-oss/risk-analysis-serverData / SQL / retrievalCommunity integrationMIT22026-10-02A profile-based risk monitor using You.com search and Jev Noul/Score/Choice for threat gating, query and evidence ranking, and severity; code owns the investigation budget while Qwen drafts source-linked reports. Outcomes are author-reported.
DataZooDE/anofox-decideData / SQL / retrievalCommunity integrationMIT22026-10-15MIT DuckDB decision extension mapping SQL text/JSON through explicitly enabled TypeSafe Jev into probabilities, choices, scores and distributions. Also supports local models and a clearly constant Stub. Registrations/settings are process-local, hosted text reaches providers and installation channels differ in signing/version. Probabilities/optional calibration need local evaluation; not installed or run here.
Monkeypox966/arxiv-jev-digestData / SQL / retrievalCommunity integrationMIT22026-10-25MIT arXiv digest using native Jev Nouls for interest/exclusion screening, separate institution preferences/Top K and DeepSeek summaries/SMTP. Budgets, preview and pending-delivery records reviewed. Preferences are not objective quality; no workflow, model or email execution.
nothans/ex-regexData / SQL / retrievalCommunity integrationMIT12026-10-09Python semantic text library using typed Jev judgments over candidate source spans for finding, extraction, replacement, splitting and ranking, retaining offsets and uncertain/error outcomes. Includes multiple backends, caching and replay; verbatim spans do not guarantee correct semantic matches or redaction.
dorkitude/skimmerData / SQL / retrievalCommunity integrationMIT12026-10-11Go CLI filtering web/file chunks or list items with goal-based Jev Nouls, preserving source line ranges and input format. Includes batching, concurrency and partial-error markers; unjudged items are kept by default but caps can trim them, and total failure reports an error. Recall and token savings were not validated here.
Zilize/parallax-dshData / SQL / retrievalCommunity integrationApache-2.012026-10-22Apache-2.0 local arXiv set-graph plugin combining keyword recall/set operations with native Jev screening/classification and batch-persisted probabilities/progress. Not installed or timing-retested; local database is not local inference and recall depends on ingested coverage.
ricomanifesto/GRCInsightData / SQL / retrievalCommunity integrationMIT12026-10-31MIT GRC news-report project with an optional three-label Jev relevance pilot, off/shadow modes and uncertain-item fallback. Another model generates reports. Pinned source review proves neither production enablement nor legal applicability; no private articles or jobs were run.
anthony-maio/hn-oracleData / SQL / retrievalCommunity integrationMIT—2026-09-24Screens historical Hacker News comments for checkable predictions with Jev; includes a preregistered pilot, calling code and author-published results, but not a completed full-archive run.
neozhu/jev-auditData / SQL / retrievalCommunity integrationMIT—2026-09-25Compares a baseline contract with scanned-copy OCR text using Jev Noul, Choice and Score questions, separates substantive changes from recognition noise and sends results below the author's cutoff to human review; includes a real API call and consistency tests.
MersivMedia/jev-rag-retrievalData / SQL / retrievalCommunity integrationMIT—2026-10-01A Jev-driven RAG ingestion and retrieval pipeline with paragraph screening, passage relevance/conflict checks, and an answerability gate; supports several vector stores, degrades without a key, and reports author-measured benchmarks.
nicia-ai/admission-decision-evalData / SQL / retrievalCommunity integrationApache-2.0—2026-10-02A reproducible knowledge-write admission study using 85 synthetic cases to compare Jev, Clef, and Kev on contradictions, restatements, and changed meaning, then simulate admit versus review; scores do not establish production accuracy.
NaokiKomura/x-bookmark-digest-templateData / SQL / retrievalCommunity integrationMIT—2026-10-04An X bookmark and tech-source digest template with optional Jev Noul relevance and Choice topic decisions over public text; without a Jev key it records unavailable and uses another workflow.
shopsmartai/ora-jevData / SQL / retrievalCommunity integrationMIT—2026-10-05An Oracle PL/SQL Jev-compatible extension batching row JSON to System One for boolean, probability, choice, and score functions. It needs a network ACL and HTTPS wallet; row data leaves the database.
dreamerron/decision-benchData / SQL / retrievalCommunity integrationMIT—2026-10-05An agent decision benchmark with 450 author-labelled items and a rules baseline; it calls Jev for routing, completion checks, browser-action risk, and more. Reported results were not reproduced here.
BeforeLanding/siftData / SQL / retrievalCommunity integrationMIT—2026-10-08Personal AI digest with a real Jev backend: batched Scores filter items and Nouls gate sources before LLM report generation. Failed or uncertain items are retained, with a separate mock path. Token savings and information recall were not validated here.
darrentmorgan/jev-accountingData / SQL / retrievalCommunity integrationMIT—2026-10-09Local transaction-classification prototype using Jev Choices through Vercel AI Gateway. Low, missing or invalid probabilities retain Unassigned for review before CSV export. Synthetic demonstration only, without database or accounting-system integration; accounting outcomes were not validated here.
armando-token/industrial-selection-benchData / SQL / retrievalCommunity integrationMIT—2026-10-09Pilot comparing structured Jev selection, RAG, scraping, LLM and rules over industrial manuals, with evaluation and evidence interfaces. MIT code and CC BY 4.0 data; only 12 held-out tasks with non-expert gold labels. The author reports low correctness and false approvals; this is not a mature selection system.
SarathChandraBellam/ToolDiscoveryBenchData / SQL / retrievalCommunity integrationMIT—2026-10-10MCP tool-discovery benchmark comparing flat/factored/hierarchical Jev Choice routing and abstention with other models and retrieval baselines. Reports routing performance without proving tool execution or agent success; gold tools were selected by coding assistants, and author timing, costs and accuracy were not reproduced.
pcparts001/pi-jev-compaction-liteData / SQL / retrievalCommunity integrationMIT (package.json declaration)—2026-10-11Experimental Jev compaction extension for Pi Agent, judging tool calls/results with two Nouls before keeping or trimming, with compaction hooks, output budgets and default-summary fallback. package.json declares MIT without a standalone LICENSE; derives from fast-jev-compaction and author retention/savings were not reproduced.
matthelam/seo-jev-analyzerData / SQL / retrievalCommunity integrationMIT—2026-10-14MIT search-intent relevance research app: Jev Scores judge intent and expected terms, with coded product scores and repeat-answer noise bands stored in browser IndexedDB. Text still reaches the model, and ±2 SD does not calibrate rankings or truth. Author 100-run experiments were not reproduced; no installation or production-page evaluation here.
RINNECODER/jev-box-studyData / SQL / retrievalCommunity integrationMIT—2026-10-15MIT Jev 1.14 image-anomaly study and app source, discovering regions and framing whole semantic units before classification/confirmation and localization display. Documents cut-word, cartoon and geometry false flags; v7b did not rerun the old 26-image set. Small handpicked samples/costs are author reports, not AI-origin attribution or production accuracy. Images reach an experimental API; no uploads or execution here.
jhochenbaum/rlm-jevData / SQL / retrievalCommunity integrationMIT—2026-10-16MIT research tools exposing batched Choice classification, Noul judgments and two-stage original-text location as RLM sub-calls, returning probabilities, offsets and usage for root-model code. Supports DSPy/custom loops with estimated Jev-only spend limits; root costs and execution permissions are separate. Small single-run results are author reports, not a production library. Not run here.
ai-ready-future/jagData / SQL / retrievalCommunity integrationMIT—2026-10-16MIT Rust grep reranker using direct Jev Nouls over matched snippets, validated and sorted in code with original paths/lines. Not full-corpus semantic retrieval. Pinned mainline implements search while the README still describes a key-setup-only release, so source is not proof of released packages. Gateway compatibility unverified; text leaves the machine. Not installed here.
etsabary/jev-1.14-comparison-benchmarkData / SQL / retrievalCommunity integrationMIT—2026-10-17MIT decision-model comparison source and records over 1,136 text questions with related variants, fixed options/checkpoints, first attempts, errors and grouped diagnostics. The author presents 19 main-table services from 22 tested, with a post-hoc format-failure exclusion. Dates/routes limit speed comparisons, budgets are soft and scores are not site measurements; not duplicated as an application case.
chadboyda/proseweaveData / SQL / retrievalCommunity integrationMIT—2026-10-17MIT Python prose metrics combining bounded Jev word-class/relatedness/familiarity and Score/genre judgments with local arithmetic/fitted models, flagging departures from author samples for rereading; cadence is separate and local. Data components carry separate licenses, not universal writing-quality judgments. Text reaches the provider and moving-alias caches may age. Author validation not reproduced; not installed here.
LandoCole/document-classifierData / SQL / retrievalCommunity integrationMIT—2026-10-20MIT document-type classification demo using native Jev Choice over 21 types plus other, returning probabilities to a web UI. The 43 synthetic samples are an author demonstration, not independent real-document or auto-filing acceptance.
tancy-920/jev-agentic-wikiData / SQL / retrievalCommunity integrationMIT (code)—2026-10-20MIT-code complete-article Wiki/Graph RAG assistant with an explicitly gated native Jev adapter and optional answer-budget judgments. Static replay and Wiki/graph benchmarks do not establish Jev end-to-end gains. Project listing only; no models or headline metrics were independently tested.
fmegahed/decision-models-for-industrial-image-qcData / SQL / retrievalCommunity integrationMIT—2026-10-23MIT industrial-defect image research with hosted Jev image calls, separate calibration and held-out comparisons, plus a distinct local-model path. R package remains in design; paper metrics are author-reported, data licenses separate, and no data/weights downloaded or paid evaluations run.
MuShan-bit/decision-moderation-benchData / SQL / retrievalCommunity integrationMIT—2026-10-24MIT Chinese moderation evaluation console separating a native Jev provider/four-question thresholds from mock/other providers, with CLI/web metrics. The 825 labels are mapped from public datasets with separate licenses. Preview results were not retested and this is not production-service acceptance.
szia-ai/jev-on-duriansData / SQL / retrievalCommunity integrationMIT—2026-10-24MIT opinion/decision-model research with native Jev’s 18 questions, other-model controls, metrics and Elo charts. The 83 opinions have separate copyrights; samples are small and selected. Rankings/costs are author-reported. No corpus copied, experiments run or weights downloaded.
ankur287/jev-open-calibrationData / SQL / retrievalCommunity integrationMIT—2026-10-27MIT calibration research comparing a scikit-learn interface with actual Jev BANKING77 experiments, including Choice/Noul, out-of-schema and few-shot variants. Data is separately CC-BY 4.0. Metrics are author-reported, not replicated here or open Jev weights.
Dongbumlee/ontosmithData / SQL / retrievalCommunity integrationMIT—2026-10-28MIT evidence-graph tooling separating proposals, Jev judging, thresholds and writes, with budgets and human schema approval. Some reference batch wiring retains an unverified legacy OpenRouter route; listed as related tooling without claiming a verified end-to-end Jev build.
lyhmain/supersonic-ontologyData / SQL / retrievalInspired replicaApache-2.0—2026-11-01Apache-2.0 SuperSonic sidecars with Jev-inspired relevance, conflict and answerability gates, using generated JSON through GLM/DashScope by default and optional System One compatibility. Not verified TypeSafe production use or open Jev weights; fail-open/first-option fallbacks remain limitations. No database connection.
dolou12388/dsh-memory-hybridData / SQL / retrievalCommunity integrationMIT (LICENSE; bundled sql.js MIT)—2026-11-02MIT hybrid-memory plugin with authoritative KV and derived vector index. Rules remain primary; conditional Jev kind/importance Choice adapters and consumers exist, but comments still mark P3 placeholder and live compatibility is unverified, so no actual-Jev case is listed. Root LICENSE includes sql.js MIT notices. No install, sessions or memory changes.
y0usaf/pi-jevSafety & moderationCommunity integrationMIT1512026-09-29A decision layer for the Pi coding agent: one gated request judges destructiveness, exfiltration, scope, and impact, plus an output judge for secrets; shadow mode by default and every error fails open.
openlayer-ai/jevalsSafety & moderationCommunity integrationMIT972026-09-29Evals and guardrails as Jev decisions: eight parallel judgments per trace in one request at a few thousandths of a cent, fast enough for every trace; backends span TypeSafe/Vercel, local Kev/Laya, or a plain LLM.
Ordo-Engine/OrdoSafety & moderationCommunity integrationMIT382026-10-14MIT related rule engine mapping recorded Jev judgments and business facts to tested, versioned outcome labels and threshold-change comparisons. Example trace does not issue refunds; its live OpenRouter URL is not confirmed against the current official API, and some missing fields default to zero. No comprehensive validation or live readiness is claimed; not run here.
datadog-labs/reflexSafety & moderationCommunity integrationApache-2.0312026-10-07Rust control-loop library using Datadog telemetry and optional Toto forecasts. Jev recommends Open/Probe/NoChange; code checks revisions, freshness, cooldowns and probe constraints before applying transitions. Includes a circuit-breaker simulator; production outcomes were not reproduced here.
leepokai/jev-guardSafety & moderationCommunity integrationMIT182026-09-22A safety gate for coding agents: risk-scores every tool call as deny/ask/allow and flags prompt injection across many coding assistants.
win4r/jev-security-scanSafety & moderationCommunity integrationMIT92026-09-22Reviews agent skills and MCP code for suspicious behavior using Jev (Chinese-language project).
madisonrickert/jev-permission-gateSafety & moderationCommunity integrationMIT52026-10-03A Jev permission precheck mod for Claude Code auto mode: eight Nouls and code thresholds decide allow, deny, or defer to the built-in classifier. Commands and recent messages may go to TypeSafe; safety measurements are author-reported.
ismaildasci/claude-refereeSafety & moderationCommunity integrationMIT32026-10-02A Claude Code completion-evidence checker: code parses test and lint results, Jev checks explicit criteria, and local receipts preserve findings; its automatic Stop hook is shadow-only, with active blocking not yet released.
suraj-phanindra/wellposedSafety & moderationCommunity integrationMIT22026-09-27Lints Jev requests with offline structural rules and Jev Noul judgments for semantic defects, then reports suggested fixes; the semantic layer needs an external API.
duberblock/JEValsSafety & moderationCommunity integrationMIT22026-09-30An interactive Jev decision playground comparing a real TypeSafe baseline with an emulator, independent model and judge while saving execution traces; its default hosted emulator receives scenario text.
psygreg/agent-reflexSafety & moderationCommunity integrationApache-2.022026-10-18Apache-2.0 coding-agent plugins; the Pi TypeSafe Noul shell-check/blocking path was inspected and fails open when unavailable. Provider presets and eight-host support claims are not site-run tests.
patronus-protect/decision-model-guardSafety & moderationInspired replicaMIT22026-10-22MIT generic decision guards/audit tooling: scans untrusted content, rejudges erased spans and applies review/abstain/block policies. Pipeline/compatible HTTP adapter reviewed; demonstrations mainly use independent local models, not proof of TypeSafe Jev results. No weights downloaded or safety metrics retested.
ohernandezdev/jevmodSafety & moderationCommunity integrationMIT12026-09-28Uses Jev Noul questions to score community-message risks across bots, CLI, API and MCP interfaces; defaults to flag-only and leaves messages alone on model failure.
coo-quack/vigilant-heronSafety & moderationCommunity integrationMIT12026-10-27Early MIT HTTP reverse-proxy experiment combining fixed rules with native Jev Nouls/category Choice, shadow mode, thresholds and failure policies, plus alternative backends. Real traffic is unvalidated, redaction is not a complete privacy guarantee, and no deployment or protection test was performed.
ravex-io/ravexSafety & moderationCommunity integrationMIT12026-11-04Optional Jev Noul moderation in an MIT Socket.IO chat library, applying local gates before sends/edits. Default fail-open differs from input refusal. Original post and source checked; no chat connection or classifier-effectiveness testing.
JevAdvBench/JevAdvBenchSafety & moderationCommunity integrationMIT (code); CC BY-NC 4.0 (data)—2026-09-27An adversarial Jev decision benchmark with typed questions, single-edit variants, a direct TypeSafe API evaluator and author-published results. Code and data have different licenses; robustness figures were not independently reproduced.
wmy2981/auto-review-siyuanSafety & moderationCommunity integrationMIT—2026-10-05A SiYuan AI tool-call review plugin: Jev assesses authorization, impact, and self-authorizing text; code thresholds then approve, deny, or ask the user. It is not a security guarantee.
OGZamasu/JevBotSafety & moderationCommunity integrationMIT—2026-10-08Discord moderation bot combining rules and Jev context classification. Starts in Observe, sends uncertain detections to review and can warn, delete or time out under configured policies, without automatic bans. Requires Discord/Cloudflare/TypeSafe setup; no live server was tested here.
fullymiddleaged/vulnderSafety & moderationCommunity integrationApache-2.0—2026-10-09Stack-aware CVE tool using Jev on Workers AI for environment, product-fit and internet-exposure judgments. Fixed code matches vulnerabilities and ranks public exploitation signals, retaining base parsing on model failure. Stack URLs are shareable; coverage and risk calibration were not validated here.
earlgreyhot1701D/readme-clew-2Safety & moderationCommunity integrationApache-2.0—2026-10-10README claim-verification prototype using Jev via Glasser for claim-type and candidate-name Choices before confidence gates and deterministic dependency, command, environment-variable and reference checks. Live scan/evaluation modules now exist; the full extension/web app is still in development and paid evaluations were not run here.
nevrs/attention-vaccineSafety & moderationCommunity integrationMIT—2026-10-10Chrome extension using batched Jev Nouls over visible text on permitted sites to show thresholded writing-technique cues, original questions and probabilities, with caching, text limits and daily budgets. Probabilities concern wording, not factual truth; enabled-site text goes to the model backend. Not installed or reproduced here.
EvanProgramming/decilabelSafety & moderationCommunity integrationMIT—2026-10-11GitHub Action classifying Issues/PRs and adding existing labels only, with Jev Noul/Choice probability, margin and fresh-snapshot gates, without executing contributor code. Supports TypeSafe/OpenRouter and a separate model; author quality is not live-verified and 24 examples do not establish calibration.
samiulhuda360/depgateSafety & moderationCommunity integrationMIT—2026-10-11Dependabot/Renovate gate using Jev over release notes, code usage, CI and risk before configurable auto-merge, review or hold policies and GitHub comments/labels with optional notifications. Includes rule/LLM baselines; author evaluations are not an independently verified safe-merge guarantee.
minitru/jev-intent-benchmarkSafety & moderationCommunity integrationMIT—2026-10-14MIT act-level indirect-injection research check using Jev over requests, tool output and a proposed act, with a two-condition flag. Returns flags/readability/probabilities without blocking. Default 400-character acts and 6,000-character content have known padding bypasses. Public attack-set results were not reproduced and are not a security guarantee; not run here.
kallurayaankit/jev-consensus-gateSafety & moderationCommunity integrationMIT—2026-10-17MIT jevlang multi-perspective prototype aggregating safety/authorization/attack questions into allow/human/block. The demo only prints, with no real executor; CI uses fake-backend imports and lint rather than live safety acceptance. Three framings of one model do not establish independent errors or authorization and example thresholds need evaluation. Not executed or called here.
kikijiki/jevmonSafety & moderationInspired replicaMIT—2026-10-19MIT Linux process-monitoring tool inspired by Jev's typed-decision interface, using local Laya/other encoders and heads by default to record safe/warn/crit verdicts for review. Not verified TypeSafe Jev calls or security efficacy; code licensing does not replace model-weight terms.
colesmcintosh/liedarSafety & moderationCommunity integrationMIT—2026-10-19MIT PDF/PNG/JPEG screening CLI and library using native Jev image/metadata requests for fixed signals and JSON reports. Not fraud determination; images are experimental and probabilities uncalibrated. No documents uploaded or independent reproduction performed.
felinae98/cmd-mod-jev-guardSafety & moderationCommunity integrationMIT—2026-10-21Command Code mod reusing jev-guard judgments with tool blocking, interactive confirmation, permission requests and content scanning. Defaults to fail-open and warning-only headless ask; custom backends can be non-Jev. Source-reviewed, without a safety-effectiveness claim.
george-amir/CustosSafety & moderationCommunity integrationMIT—2026-10-31MIT Rust access-log judgment framework batching Jev Nouls/Choices after L1 window filtering and gating expiring bans. Offline mocks and production effects are separate. Source review only: no firewall changes or false-block validation; author load figures are not site measurements.
hierichhsu/your-cto-jevSafety & moderationCommunity integrationMIT—2026-11-02MIT commit/agent-hook tool applying local block/warn gates to Jev Noul/Score advice. Provider failures fail open, distinct from local deterministic rules. Source review only, without hooks, local records or reproducing author safety/latency claims.
Ryankapa/PasteGuardSafety & moderationCommunity integrationMIT—2026-11-04MIT Windows tray prototype holding command-like clipboard text while a Jev Noul and local gate select block/restoration. Configurable error policy; command content leaves the machine. Not run or connected to clipboard; no ClickFix-protection guarantee.
jev-chat/jev-chat-jarvisEnd-user appsCommunity integrationMIT7,0752026-09-29A mobile chat copilot (Android hub plus Windows/macOS siblings) that reads the screen, lets Jev judge intent and mood and rank drafts, then fills the input box — sending stays manual, no platform APIs involved.
anishfn/shapeshiftEnd-user appsCommunity integrationMIT7382026-09-29An input that becomes what you mean: one 14-question Jev call morphs the box into an event card, checklist, bill splitter, or color picker, while dates and math stay deterministic with an offline fallback.
jev-chat/jev-chat-windowsEnd-user appsCommunity integrationunknown6852026-09-29The Windows sibling of the Jev chat copilot: window screenshots plus offline OCR read WeChat or KakaoTalk, Jev judges intent and mood and offers three ranked drafts; sending stays manual.
ChetasLua/jevmeterEnd-user appsCommunity integrationMIT1012026-09-29A live meter for videos: every sentence scored against a preset, dodges and hype flagged, rendered as a postable 16:9 edit; a full debate costs about $0.05.
fazlerocks/jevmailEnd-user appsCommunity integrationMIT902026-09-29Open-source Gmail triage: three questions per message sort five trays with urgency, strictly read-only; 1,000 emails in about a minute for a few cents, with corrections stored locally.
AboveColin/HA-JevEnd-user appsCommunity integrationMIT692026-09-29A Home Assistant integration that turns Noul/Choice/Score questions into sensor entities and automation actions, with 25 blueprints, Assist routing, and daily cost reporting; community-made, unaffiliated with TypeSafe.
retsu-AI/qqEnd-user appsCommunity integrationMIT132026-10-11Optional Jev decision layer in a Rust coding-agent runtime, choosing only among configured authorized models/efforts and disabled by default. Invalid responses or failed plan recompilation retain the configured model. TypeSafe adapter and runtime wiring were inspected without running it or validating task quality/cost.
reindent/jauvexEnd-user appsCommunity integrationApache-2.052026-10-02A desktop voice client for Claude, Codex, and Grok sessions; optional TypeSafe Jev makes bounded routing decisions, with another model as fallback when no Jev key is configured. This site did not run the app.
StellarieX/HeliotropeEnd-user appsCommunity integrationMIT52026-10-10Household load-scheduling prototype using Jev appliance classification and importance scores while code handles urgency, deadlines, capacity constraints and optimization. Low-confidence or unconfigured paths explicitly label heuristic/mixed results. Carbon intensity is a proxy estimate, not measured device emissions savings.
collection-of-things/aggregate-mailEnd-user appsCommunity integrationMIT22026-10-28MIT Emc² mail client with native Jev category/relevance/reason labels, rule/teaching precedence and local fallback, plus desktop/browser targets. No inbox, storage/platform or sending verification; mail fields can leave the device.
kmosher/jevons-talkingEnd-user appsCommunity integrationMIT12026-10-08Experimental predictive keyboard: Jev selects finite words, letters, punctuation and stop keys while code composes text. Includes transcripts, backspace-loop controls and a step budget; answer quality and assistive use were not reproduced here.
kortexa-ai/mappityEnd-user appsCommunity integrationMIT12026-10-09Natural-language map exploration: Jev judges place types, then batches place and street relevance while code combines OSM/Mapillary data, hard constraints and probabilities to highlight matches. Supports a separate local Shingi backend; author demos, speed and cost were not independently measured.
YongLD/Multi-UAV-JEVEnd-user appsInspired replicaApache-2.012026-10-10MuJoCo visual pursuit simulation passing bounded action decisions over structured state to deterministic controllers. Defaults to the separate NeoHorse-Jev-4B model; the current setup has one pursuer and one target, with multi-agent coordination planned. Not TypeSafe Jev open weights or real-hardware validation.
ys118/ra2web-jev-playerEnd-user appsInspired replicaMIT12026-10-10ChronoDivide single-player experiment combining deterministic game logic with System One-compatible judgments and audit/replay. Defaults to local clef-flash, with configurable hosted Jev; current matches are not Jev measurements and were not run here.
daiki-beppu/live-mindmapEnd-user appsCommunity integrationMIT12026-10-19MIT live meeting-mindmap tool using native Jev to judge decisions/action items in new remarks and route Haiku/Sonnet updaters. Judges can be disabled or replaced; no audio/cloud run was performed and Jev does not generate the mindmap prose.
duketopceo/wispEnd-user appsCommunity integrationMIT—2026-10-02A voice desktop assistant for Omarchy/Hyprland: OpenRouter Jev routes transcribed speech to apps, tools, agents, or answers with risk-gated actions; another chat model writes answers, and the author provides a macOS port.
cfdgasman/jev-cfd-copilotEnd-user appsCommunity integrationMIT—2026-10-03A rarefied-gas parallel-plate teaching example: Jev selects flow, geometry, gas, and quantity roles while code gates confidence and runs the physics solver; recorded answers replay offline and solver validation is author-provided.
dante01yoon/ComfyUI-SystemOneEnd-user appsCommunity integrationMIT—2026-10-03ComfyUI System One nodes with TypeSafe Jev and local Laya backends for Choice/Noul/Score workflow branching; visible demos mainly use Laya and are not Jev performance evidence.
SeeYangZhi/clipsieveEnd-user appsCommunity integrationApache-2.0—2026-10-04A creator research tool that builds post evidence from transcripts, on-screen text, and comments, scores fixed rubrics with Jev, then sends a shortlist to Claude. The frontend remains planned; live platform flow was not reproduced.
jimador/bayes-of-our-livesEnd-user appsCommunity integrationMIT—2026-10-12Embabel/Dice city-outing prototype using typed Jev judgments over code-generated tentative plans and a research option. Defaults to local rules with OpenAI/Clef alternatives. Map/site clues and preference memory expose consent and unknowns, without bookings; snapshot dependencies and author observations were not reproduced.
kallurayaankit/jev-email-triageEnd-user appsCommunity integrationMIT—2026-10-13Two-stage email triage: Jev classifies, then selects bounded tone/reply intent; code returns archive, mute, human-review or draft signals. MIT, Python/jevlang example without mailbox integration, prose generation or sending. Thresholds were not calibrated here.
hwong103/jevmailEnd-user appsCommunity integrationMIT—2026-10-13MIT Gmail-triage prototype derived from fazlerocks/jevmail, with inspectable Workers AI/Command Code Jev adapters, persistent corrections and tray rules. Synthetic fixtures work; the author explicitly reports no real Gmail/deployment acceptance and deploy omits the PWA. No complete self-hosting or automatic filing/sending; not run here.
elioli983/TechMedicEnd-user appsCommunity integrationGPL-3.0-only—2026-10-13GPL-3.0-only Windows troubleshooting evaluation app: another model builds state, Jev chooses fault domain/next action and judges diagnosis, remediation and resolution evidence, with coded lifecycle/prerequisite guards. Official TypeSafe and custom gateways are distinct. Offline SQLCipher tests do not prove live-model quality or comprehensive security; no automatic device repair and not run here.
ceokobec-hue/job-drillEnd-user appsCommunity integrationMIT—2026-10-14MIT Korean scenario-response practice: Jev selects accept/defer/decline and scores/checks scenario rubrics, while code computes passes and reports against local baselines. Coding assistants generate scenarios/overall comments. Simulated scales are not actual customer reactions or employee certification; replies reach OpenRouter, with mobile/Windows/persistence limits. Not run here.
tugrulguner/utilitypotEnd-user appsCommunity integrationMIT—2026-10-15MIT bounded conversion/planning utilities where Jev interpretation is reviewed before coded execution. Diagram branches, column maps, dates and other tasks use finite choices validated by a Worker. Source/fixture tests do not prove production deployment or live quality; rate controls are not exact global spend caps and submitted text reaches TypeSafe. Not run here.
WilliamJones/JEV-CityEnd-user appsCommunity integrationMIT—2026-10-16MIT isometric city simulation where Jev selects code-defined feasible actions and deterministic rules govern movement, hazards and effects. Shows probabilities/provenance; recorded decisions are the default, while live mode needs credentials/passcode/budget storage and failures visibly fall back to rules. Author timing and 4,902 valid responses do not establish real-city quality; not run here.
raia-live/jev-senselab-demosEnd-user appsCommunity integrationMIT—2026-10-16MIT Jev/SenseLab decision experiments comparing fixed policy, retrieval memory and feedback memory on synthetic business streams. The model chooses bounded actions; code simulates review, memory updates and reports. Not real customer routing, real losses or Jev weight learning; errors choose the first action and carry an error field. External services need separate access. Author effects not reproduced; not run here.
jjmerri/gspro-commentatorEnd-user appsCommunity integrationMIT—2026-10-17MIT Windows/GSPro simulator commentary booth optionally using experimental Jev images/Boxes/Nouls for minimap facts and coded geometry/threshold branches. LLM/TTS generate speech separately; failures fall back to another vision model. Not real-course safety navigation; screenshots leave the machine and models/voices require separate setup. Author timing/costs not reproduced; not run here.
holy-max/obsidian-systemone-auto-tagEnd-user appsInspired replicaMIT—2026-10-21Obsidian Noul scoring over existing tags, with matrix review before accepted writes. Independent implementation inspired by Kaku’s public UX. API/bucketing/write wiring reviewed without installation or retest; notes leave the device and zero-filled answers/rough costs have limits.
polidog/dot-launcherEnd-user appsCommunity integrationMIT—2026-10-21Android launcher prioritizing strong lexical matches and adding native Jev Choice/Noul semantic candidates for user-confirmed launch; also categorizes apps. Not installed or store-verified; device behavior is author-reported and heuristic thresholds do not guarantee accuracy.
aiden-guan/courtvisionEnd-user appsCommunity integrationMIT—2026-10-22MIT basketball vision/stats/narration prototype with native Jev SDK, bounded coach-card/line-kind judgments and deterministic fallbacks. Live/stub/replay/off are separate. Calls/consumer reviewed without real-game or performance tests; simulations/test counts do not establish field accuracy.
matteoandradeboschi/jev-classificadorEnd-user appsCommunity integration0BSD—2026-10-230BSD Obsidian plugin using native Jev for bounded note classification and wordlist tags, with confirmation for uncertain pairs before frontmatter writes and folder exclusion before reads. No install, note transmission or store verification; classification quality untested.
vllm-project/vllmCommunity SDKs & clientsInspired replicaApache-2.095,3662026-10-25Apache-2.0 inference server with a System One-style structured-decision endpoint. PR #59299 is merged and the current API route was reviewed. This is compatibility tooling for open models, not open TypeSafe Jev weights or equivalent quality. Model/read-strategy limits are version-specific. No weights downloaded, GPU launched or retest.
google/adk-pythonCommunity SDKs & clientsCommunity integrationApache-2.021,9932026-10-25Google ADK native TypesafeClassifier with bounded question types, answer checks and ClassifierNode/callback wiring. Labs APIs are experimental and generic samples default to other models. Source/docs reviewed; no install, model calls or customer-flow acceptance. Not a TypeSafe official project.
juspay/neurolinkCommunity SDKs & clientsCommunity integrationMIT1442026-10-09Community TypeScript library adding typed decide alongside generation and streaming, with direct TypeSafe/Jev and Vercel Gateway transports. Parses Choice/Score/Noul responses for application handling and can degrade without configuration. Author production and cost claims were not verified here.
obie/ruby_decision_modelCommunity SDKs & clientsCommunity integrationMIT522026-09-29The decision-model client for Ruby: one Client over OpenRouter (default) or direct TypeSafe, returning calibrated noul/score answers with zero runtime dependencies.
spring-ai-community/spring-ai-typesafeCommunity SDKs & clientsCommunity integrationApache-2.0512026-10-06A Spring AI community TypeSafe Java client and integrations for typed System One, JevJudge, response-refinement advisors, RAG, and tool search. It is not a TypeSafe first-party SDK; hosted calls and sample quality were not reproduced here.
schainks/system_one_clientCommunity SDKs & clientsCommunity integrationApache-2.042026-10-12Community Elixir System One client returning full Choice/Noul/Score distributions for TypeSafe/OpenRouter Jev, separate Clef and compatible providers. Distinguishes real Req HTTP from Stub, with retries, error tuples and telemetry excluding request content. Apache-2.0, unofficial and not run here.
Engid/question-kitCommunity SDKs & clientsCommunity integrationMIT22026-10-11Community TypeScript System One toolkit and procedural service agent using Jev for bounded intents, menu/slot values and confirmations, while code owns gates, tool calls and templated replies. Pizza demo distinguishes live/fake/replay and uses a simulated register; alpha, without real payment/kitchen integration, and not an official SDK.
diegoag64/jev-devtoolsCommunity SDKs & clientsCommunity integrationMIT22026-10-24MIT experimental React/Node inspector wrapping the official SDK with bounded memory recording, same-origin SSE and condition-flow visualization; capture is off by default. Demos mainly use fixtures and the author has not verified live inference. Tool entry only; no fixture claimed as live Jev, installation or trace access.
mrz1836/go-discernCommunity SDKs & clientsCommunity integrationMIT12026-10-15MIT community Go typed decision layer with explicit TypeSafe Jev/Clef selection and requested/resolved model records. Callers supply credentials; each Call makes one POST without redirect following, with response bounds. No third-party runtime dependencies, without proof of model quality or deployment; not run here and the basic ticket example is not duplicated as a case.
osolmaz/switchCommunity SDKs & clientsCommunity integrationMIT12026-10-27MIT menu-based decision harness with enumerated actions, model choices, evidence traces and replay. A pi-ai classifier supports Jev, with chess and legal-judge domains. Legal grading is not legal advice; pieces are separately BSD-licensed. No runtime or effectiveness verification.
AnthusAI/CyclotronCommunity SDKs & clientsCommunity integrationMIT12026-10-28MIT decision-feedback harness with a native Jev SDK adapter, rubric/few-shot/supporting questions, ML calibration and held-out evaluation, plus other models. README quickstarts use ScriptedModel, not Jev evidence. No event/label stores, optimization runs or model downloads.
OtohaCo/SwiftAgentCommunity SDKs & clientsCommunity integrationMIT12026-10-30MIT Swift Agent SDK with a separate Jev DecisionProvider returning Noul/Choice/Score advice. The example defaults to a fixture and requires live opt-in. RC status and host execution qualification are separate; advice is not tool authorization. No build/install, keys or production-platform test.
BareIQ/Jev.TypeSafe.AICommunity SDKs & clientsCommunity integrationMIT—2026-10-03An unofficial .NET TypeSafe/Jev client with typed Noul, Choice, and Score answers, injectable HttpClient, cancellation, and retries; not a TypeSafe first-party SDK.
deepnoodle-ai/decideCommunity SDKs & clientsCommunity integrationApache-2.0—2026-10-04A Go library and CLI that defaults to TypeSafe Jev for typed Choice/Noul/Score decisions over text, files, or diffs; it also supports compatible services. Demos and timing are author-provided.
dmishra2022/jev-java-sdkCommunity SDKs & clientsCommunity integrationApache-2.0—2026-10-05An unofficial Java 26 Jev SDK with a typed System One client, retries, virtual threads, and an offline ticket sample; its structured-concurrency module requires preview features.
hendrik77/cap-jevCommunity SDKs & clientsInspired replicaMIT—2026-10-22MIT internal System One decision service for SAP CAP, with typed validation, deadlines/retries, Destination and mock support. Its current Cloudflare provider defaults to Clef; compatibility does not establish a verified TypeSafe Jev business call. Related tool only, not installed or npm-publication verified.
weiping/magpie-cf-jevCommunity SDKs & clientsCommunity integrationMIT (package.json declaration)—2026-10-25Magpie adapter for Cloudflare Jev/Clef preserving state/questions and unwrapping nested result envelopes to expose answers. package.json declares MIT; no standalone LICENSE. Author direct/host tests were not repeated. No install, credential access or Cloudflare call; Clef is not TypeSafe Jev.
phpaisdk/typesafeCommunity SDKs & clientsCommunity integrationMIT—2026-10-31MIT community PHP 8.3 adapter mapping predicate/choice/score to native Jev while preserving probabilities and usage. Text/JSON only; images are rejected before HTTP. Source and mock-HTTP tests are available, without dependency installation or live qualification; not an official TypeSafe SDK.
ggml-org/llama.cppOpen replicasInspired replicaMIT130,2382026-10-04Merged PR #29818 adds a compatible `/v1/systemone` API and Choice/Noul/Score probabilities for Laya, Julia-1, Lev, OpenJev, and Kev. This is alternative-model infrastructure, not TypeSafe Jev weights.
NandhaKishorM/layaOpen replicasInspired replicaApache-2.028,2122026-09-29An independent open-source non-autoregressive System 1 decision engine (its own model — neither a Jev implementation nor Jev weights): typed choice, score, and yes/no decisions over any text, served locally or behind compatible APIs, and measured by third-party boards like JevBench.
TheoLeeCJ/SemIf-OpenJevOpen replicasInspired replicaMIT4,5602026-09-29The SemIf author's open-models edition of the semantic-if interface, runnable on a 3090 or in the browser. Independent and unaffiliated with Jev or TypeSafe; not Jev weights.
TheoLeeCJ/SemIfOpen replicasInspired replicaMIT2,9352026-09-22An inspired open-source reimplementation of the 'semantic if' interface on open models, runnable on a single RTX 3090. Formerly OpenJev; the author states it is independent of TypeSafe. Not the Jev weights.
jaredpalmer/kevOpen replicasInspired replicaApache-2.01,9682026-09-22A tiny Jev-like decision-model family built on Qwen3.5 that you can train and run yourself.
TianyuCodings/NanoJevOpen replicasInspired replicaMIT1,7512026-09-22A 0.6B-parameter Jev replica with parallel decisions, dynamic candidates, and an end-to-end training pipeline. Community project, unrelated to TypeSafe.
nokia-applied-research/AnyJevOpen replicasInspired replicaApache-2.09382026-09-29From Nokia applied research: turn any LLM into a Jev-style calibrated decision endpoint without fine-tuning (measurably better order-flip and calibration error), servable via vLLM.
wfzyx/vonOpen replicasInspired replicaApache-2.07672026-09-29An open-source System One decision model: non-autoregressive, sub-15 ms claims, local drop-in alternative. Community work, unaffiliated with TypeSafe.
featherless-ai/simple-jevOpen replicasInspired replicaApache-2.05682026-09-29From Featherless: read open-model logits directly for structured classification and scoring, with a no-login public demo (2k context, 2 RPS) and an eval framework.
razorback16/openjevOpen replicasInspired replicaApache-2.05212026-09-29An open, Jev-compatible decision server built on DiffusionGemma that runs locally. A community replica, unrelated to TypeSafe.
Yinsongxu/LLM2JevOpen replicasInspired replicaApache-2.03702026-09-29Turn local language models into Jev-style structured decision models over text and images. Independent implementation, not Jev weights.
ekzhang/openjev-sglangOpen replicasInspired replicaunknown3322026-09-29Serves Qwen3.6-35B on SGLang behind the TypeSafe/Jev HTTP API: one B200 per container, radix caching, breakable prefill CUDA graphs, plus Modal deploys and a smoke test.
logan-markewich/jeffOpen replicasInspired replicaMIT2642026-09-29A self-hosted, GliFormer-powered drop-in replacement for Jev by a LangChain author; runs on your own machine. Community project, unaffiliated with TypeSafe.
jarrodwatts/jev-traderTrading (research only)Community integrationMIT1,7402026-09-22Makes one buy/sell decision per Monad block (dry-run by default). Research only — not investment advice; live trading carries real risk.
h100envy/crawl-hiveTrading (research only)Community integrationMIT842026-10-14MIT crawler-signal research platform using Jev Nouls to filter low relevance and prioritize scans, with Grok/rules for extraction and separate scanning/human judgment. Inspectable code does not validate token, trading or profit claims. Process-local budgets and author costs/timing were not reproduced; no crawling, wallet connection or execution here.
truongxxxx/jev-polymarket-testTrading (research only)Community integrationMIT—2026-10-12Retrospective Jev evaluation over resolved prediction markets, comparing blind/priced Nouls and category Choices with calibration and market-price baselines. MIT; training knowledge, timing and category confounds remain. Paper results do not prove prospective trading returns; raw model answers are uncommitted and were not reproduced.
yibie/awesome-jevCurated listsCommunity integrationunknown1,4932026-09-24A curated English list of hundreds of projects and discussions across 13 categories; the author warns that same-day batch submissions are unvetted.
heyjunpenn/awesome-jevCurated listsCommunity integrationMIT8952026-09-29A community-maintained, verified catalog of 944 open-source Jev projects with evidence labels.
v-modal/awesome-jev-toolsCurated listsCommunity integrationunknown7322026-09-29A curated list of tools built for Jev.
AnotiaWang/awesome-jevCurated listsCommunity integrationCC0-1.05672026-09-29A curated list of Jev / TypeSafe System One applications, libraries, and resources, published under CC0.
yzfly/awesome-jev-zhCurated listsCommunity integrationCC0-1.0672026-09-24A Chinese curated list of official resources, apps, agent tooling, and replicas, with a Chinese quick-start guide.
mani-aiml/jev-demosCurated listsCommunity integrationMIT72026-10-19MIT community demo collection with reviewed listing-photo Noul/Box wiring and three-way rule scoring, alongside retrieval, judging, and local-model comparisons. Only the named paths were source-reviewed; not an official TypeSafe benchmark or production-policy guarantee.

Using the table

Section titled “Using the table”
  • Adopting an SDK: start in “Official SDKs & tools”; the skills repo teaches Claude Code / Codex-style agents to call the API correctly
  • Building agents: the MCP servers expose judgments as tools; the guard projects implement tool approval (see Tool approval)
  • Local experiments: replicas run a “Jev-style” interface on consumer GPUs — great for understanding the concept, never a basis for extrapolating Jev’s own production behavior

Submitting projects

Section titled “Submitting projects”

This table follows our verification policy: the repo exists, the README matches the description, and the license is confirmed. Report problems with a project to its original repository; to correct this table or recommend a project, email contact@typesafe-jev.com.

The public data and daily discovery record live in awesome-jev-cases.

Section titled “Related pages”

Video tutorials · Open-replica video roundup (Sam Witteveen, title verified) · Use cases

Jev Model Guide (typesafe-jev.com) is an independent community site with no affiliation with, endorsement by, or agency for TypeSafe AI. Opinions here are the editors’ own. See Privacy & ad preferences. Jev, System One, and TypeSafe are names belonging to their respective owner.

Official entry points: typesafe.ai·docs.typesafe.ai·console.typesafe.ai