Jev report · in production

Jev in production

532 products and tools that run TypeSafe AI's Jev for real users, sorted into 14 uses. 209 publish a measured result, which leads their row; 374 are open source. Last entry added 2026-10-11; the list refreshes twice a day.

Application of the week · 2026-10-11

Jevscan

Jevscan classifies every blockchain transaction live, fast enough to flag attacks before they finish.

Jevscan is a chain monitor that sends every on-chain transaction's fact sheet to Jev for a classification of attack or benign. The vendor reports catching 56 of 60 known hacks in its benchmark, with a 0.058% false positive rate and a decision time of 2.7 seconds per block. Monitoring every transaction on a live chain requires a judgment that keeps pace with block production; a generated-text model would be too slow and too costly to run on each transaction, which would force sampling or delay. Because Jev returns a typed decision instead of prose, Jevscan can screen the full transaction stream in real time and flag suspicious activity before a hack fully executes, rather than reviewing it after the fact.

4securitydata

What it is used for

Every entry sits under the one job Jev does in it. Pick a job to see only those.

  1. 01Model and agent routing62Jev picks which model, reasoning tier, agent, handler or cached answer takes a task, prompt or turn, usually to cut cost or latency before a larger model runs.22 with a published number
  2. 02Message and ticket triage21Jev sorts incoming emails, chat messages, support tickets, issues, alerts, logs or customer feedback into categories, owners or priorities so the right person or queue acts on them.10 with a published number
  3. 03Agent tool and action selection61Jev picks the next tool, skill, command, click or operational move an agent or automated system executes, including browser and computer use, trading orders and warehouse steps.27 with a published number
  4. 04Agent guardrails and approvals46Jev judges whether an agent's tool call, shell command, edit, claim or turn end is safe, in scope and rule-compliant, then allows, denies, nudges or auto-approves it.16 with a published number
  5. 05Agent context and memory36Jev decides which tool output, conversation history, repo files, skills or stored memories an agent keeps, recalls or drops, to shrink context and keep answers grounded.27 with a published number
  6. 06Code and security review28Jev judges source code, diffs, pull requests, API specs or security configurations against written rules or risk, inside linters, PR gates and scanners.12 with a published number
  7. 07Evaluation and testing29Jev grades LLM and agent outputs, semantic test assertions, test selection or end-to-end test steps, or is benchmarked as the judge inside an eval harness.20 with a published number
  8. 08Search, ranking and matching50Jev reranks or filters search hits, RAG passages, listings, catalog items or form fields by how well each one fits a query or stated need.20 with a published number
  9. 09Moderation and content filtering30Jev screens user posts, incoming requests, reviews, listings and social feeds for spam, abuse, prompt injection, ads, slop or policy violations, then blocks, hides or flags them.8 with a published number
  10. 10Document and record classification39Jev labels documents, receipts, invoices, tax or trade forms, papers, web pages, applications or activity records into categories or flags inside a business, research or personal process.14 with a published number
  11. 11SQL and data pipelines24Jev runs as a SQL function, dataframe column, shell-pipeline step or bulk labeler that answers one typed question per row across a dataset.10 with a published number
  12. 12Writing and content scoring20Jev scores drafts, posts, comments, transcripts and video for tone, clarity, stance, quality or hook strength, to give the writer feedback or the analyst a read rather than to block content.9 with a published number
  13. 13Chat bots and games16Jev decides when and how a chat bot, messaging assistant, companion or game character replies or acts, and judges player input inside games and puzzles.5 with a published number
  14. 14SDKs and workflow integrations70Language SDKs, MCP servers, gateways, CLIs, agent frameworks and workflow-builder nodes that expose Jev's typed decisions for whatever job the developer defines, rather than one fixed job.9 with a published number

The entries

More filters

532 of 532 shown

01Model and agent routing62

  1. ~0.4 s per turnmeasured

    AI infrastructureBetaModel and agent routing

    Hermes Jev Skills

    Skill pack for Hermes, Claude Code and Codex where Jev picks the model per turn, selects skills, ranks search results and memory passages, and sorts mail. About 0.4 s per routing decision (author).

    1,081 stars · Python · Open-source tool

  2. ≈ −60 %vs a baseline

    Developer toolsBetaModel and agent routing

    Jev Codex Router

    Local proxy for Codex where Jev picks the model and thinking effort for each model call, including continuations after tools. About 60% lower list-price cost than a full-frontier baseline in a 237-turn historical replay (author).

    273 stars · JavaScript · Open-source tool · archived

  3. 86%vs a baseline

    Developer toolsAnnouncedModel and agent routing

    Quicksilver

    Claude Code skill routes bulk judgment calls to Jev, cutting Claude tokens 86% and running up to 20x faster across a 12-task benchmark (Quicksilver).

    120 stars · JavaScript · Open-source tool

  4. 13.0×vs a baseline

    AI infrastructureBetaModel and agent routing

    Grok Bot Jev Router

    Python router where Jev classifies each Grok Bot request before research, browser or retry work and returns an action such as reuse cache, stop retry or ask a human. 13.0× shorter wall time in one ranked-research comparison (author).

    104 stars · Python · Open-source tool

  5. 4,000measured

    AI infrastructureBetaModel and agent routing

    Jevstiller

    Jevstiller distills a repeated Jev classification call into a local model over time; a demo shows the local model taking over after about 4,000 requests (tomerglick57).

    66 stars · Python · Open-source tool

  6. ~230 msvs a baseline

    Developer toolsIn productionModel and agent routing

    jev-use

    Plugin and MCP server for Claude Code, Codex and pi that routes agent steps needing no text output to Jev (check, pick, rate) and gates shell commands via a PreToolUse hook. Dangerous commands denied in ~230 ms (author).

    56 stars · JavaScript · Open-source tool

  7. 36%vs a baseline

    AI infrastructureBetaModel and agent routing

    DuoMind

    A local dual-system LLM server pairs a 3B model with Jev; author's benchmark shows tokens per second up but latency up 36% (author).

    16 stars · Python · Open-source tool

  8. 394msvs a baseline

    AI infrastructureIn productionModel and agent routing

    jevcache (kushals256)

    Local OpenAI-compatible proxy in which Jev judges whether a new prompt has the same intent as a cached one and serves the stored answer on a match, failing open to the model. A Jev-confirmed hit returned in 394ms, against 3244ms for a miss (author).

    12 stars · TypeScript · Open-source tool

  9. 350msmeasured

    AI infrastructureIn productionModel and agent routing

    Anchor (QpiAI)

    Extracts facts and a confidence score from policy requests with Jev, falling back to GPT for about a third of cases while the rest finish in 350ms (Anchor).

    10 stars · Python · Open-source tool

  10. no published number

    Developer toolsBetaModel and agent routing

    claude-subagent-router

    claude-subagent-router uses a Jev classifier to pick Sonnet or Opus per Claude Code sub-task; the author measured Sonnet at high effort passing the same hidden tests as Opus at max effort (stas4000).

    10 stars · Python · Open-source tool

  11. 14%vs a baseline

    Developer toolsBetaModel and agent routing

    opencode-jev-router

    opencode-jev-router asks Jev how much reasoning each OpenCode step needs; the author measured the same 44/44 pass rate as fixed high effort with 14% fewer tokens (robertn702).

    10 stars · TypeScript · Open-source tool

  12. a few hundred millisecondsmeasured

    Developer toolsIn productionModel and agent routing

    Hermes Switchyard

    Jev answers skill-routing, reasoning-effort and model-choice questions for Hermes Agent sessions, returning a decision in a few hundred milliseconds per call (author).

    7 stars · Python · Open-source tool

  13. 90% vs 88%vs a baseline

    Developer toolsIn productionModel and agent routing

    langchain-skill-router

    Jev picks which skill handler takes each turn in a LangChain deepagents router; author reports 90% vs 88% accuracy at 4x fewer tokens per turn. (deyna256)

    6 stars · Python · Open-source tool

  14. 434msvs a baseline

    Developer toolsBetaModel and agent routing

    ego-decision-layer

    Replaces the per-step LLM turn in ego lite browser automation with one Jev decision to pick the next action, cutting median latency to 434ms (author).

    5 stars · JavaScript · Open-source tool

  15. 89.7%vs a baseline

    AI infrastructureIn productionModel and agent routing

    Miser

    Jev classifies prompt complexity to route requests in an open-source AI gateway; a held-out benchmark reports 89.7% exact accuracy versus 83.6% for its own heuristic fallback (Miser).

    5 stars · Rust · Open-source tool

  16. 6.5-25msvs a baseline

    Developer toolsBetaModel and agent routing

    Design OS Generative UI

    Generative UI engine routes layout decisions to Jev instead of streaming LLM tokens; author claims responses of 6.5-25ms versus seconds-long token streaming (author).

    3 stars · TypeScript · Open-source tool

  17. 100-400msvs a baseline

    FinanceBetaModel and agent routing

    finance-agent (AmishKakka)

    Jev replaces the LLM router in a personal multi-agent finance app, cutting routing latency from a full LLM call to about 100-400ms (author).

    2 stars · Python · Open-source tool

  18. 11% (7 of 63)vs a baseline

    ProductivityIn productionModel and agent routing

    Agentic Hat (Heartbeat)

    Agentic Hat's Heartbeat feed watcher asks Jev whether each new RSS or Atom entry matches the watch request and calls the main model only for matches or uncertain entries. Only 11% (7 of 63) of scheduled runs reached the main model (company).

    Commercial product

  19. 0.06 centsmeasured

    AI infrastructureIn productionModel and agent routing

    Best Fit

    Jev runs a Choice, Score and Noul call per query to rank AI models for a task, each call costing 0.06 cents. (Best Fit)

    Commercial product

  20. 65% cheapervs a baseline

    Developer toolsPilotModel and agent routing

    Fuel

    Fuel classifies the first message of a new chat to pick the intelligence tier that chat needs; on its routing benchmark Jev beat GPT-OSS-120B at the task. 65% cheaper (author).

    Write-up

  21. up to 42%measured

    Developer toolsIn productionModel and agent routing

    jev-router (kalowery)

    Local proxy classifies coding-agent requests with Jev to route them to different Claude models, cutting cost by up to 42% (author).

    Open-source tool

  22. sub-25msmeasured

    AI infrastructureIn productionModel and agent routing

    JevProxy

    Hosted reverse proxy for coding agents that intercepts intermediate tool calls, routing and guardrail checks and answers them with Jev instead of a frontier LLM, escalating on low confidence. Sub-25ms decisions (company).

    Commercial product

  23. 30-40% lower cost per correct answervs a baseline

    AI infrastructureBetaModel and agent routing

    Zabio

    Jev classifies each incoming request to route it to a model; claimed 30-40% lower cost per correct answer than OpenRouter's auto router (company).

    Commercial product

  24. no published number

    AI infrastructureIn productionModel and agent routing

    Agent Squad

    Multi-agent orchestration framework, formerly awslabs/agent-squad, ships JevClassifier in TypeScript and Python, which sends agent descriptions to Jev as one choice question and routes each turn to the winning agent with its confidence.

    7,790 stars · Python · Open-source tool

  25. no published number

    Developer toolsAnnouncedModel and agent routing

    JevRouter (BillionsBobby)

    An early-stage open-source router lets Jev pick models, tools and subagents for agent harnesses; author reports fast community growth but no metrics yet.

    458 stars · TypeScript · Open-source tool

  26. no published number

    Developer toolsAnnouncedModel and agent routing

    System One Connector

    MCP connector gives Claude Code, Codex and other agents typed access to Jev for routing and urgency decisions (System One Connector).

    349 stars · Go · Open-source tool

  27. no published number

    Developer toolsBetaModel and agent routing

    keel

    macOS coding workspace where opt-in hosted Jev, or local Laya by default, picks an eligible route for each fresh unpinned task and a tool focus inside the embedded DeepSeek loop; the app validates each choice before applying it.

    332 stars · Rust · Open-source tool

  28. no published number

    Developer toolsIn productionModel and agent routing

    GoEventBus

    Go event bus library adds an optional decision layer where Jev chooses which event type to enqueue when no deterministic rule matches, alongside rule-based and cached routing paths (GoEventBus).

    73 stars · Go · Open-source tool

  29. no published number

    ProductivityIn productionModel and agent routing

    Light-Whisper

    Windows dictation app where, in Auto mode, Jev decides whether AI polish, screen context or web search is needed for a transcript before the LLM runs; optional Jev checks also flag meaning changes after polishing.

    69 stars · Rust · Open-source tool

  30. no published number

    Developer toolsIn productionModel and agent routing

    Pi-Bifrost

    Model-routing extension for Pi with an optional Jev classifier that picks a configured routing tier before generation; Bifrost then selects the model from that tier's pool by health and strategy.

    69 stars · TypeScript · Open-source tool

  31. no published number

    Education and researchIn productionModel and agent routing

    ZengLian AI Tutor

    K-9 education RAG tutor app uses a dual-channel failover design where Jev decides which model or path handles each request, shipped as downloadable APK releases (ZengLian AI Tutor).

    62 stars · JavaScript · Open-source tool

  32. no published number

    ProductivityIn productionModel and agent routing

    ClickyX

    Jev decisions judge the focused window in a chat co-pilot added in version 0.2.3.

    25 stars · Rust · Open-source tool

  33. no published number

    AI infrastructureIn productionModel and agent routing

    BrighTO Router

    BrighTO Router, a Rust LLM gateway, routes traffic to SystemOne/Jev decision backends alongside OpenAI and Anthropic models, with team keys and budgets (author).

    21 stars · Rust · Open-source tool

  34. no published number

    Developer toolsIn productionModel and agent routing

    Gatekeeper

    Answers which handler owns a request as a typed Choice with confidence bands using Jev, after simple requests are routed directly by rules, inside a Claude Code hook (Gatekeeper).

    20 stars · Python · Open-source tool

  35. no published number

    Developer toolsIn productionModel and agent routing

    Switchboard

    Local proxy for Claude Code and Codex sends each new task to Jev in one request for a capability tier and per-model reasoning effort with confidence; a user-editable policy maps the answers to a model and pins it for the conversation.

    19 stars · TypeScript · Open-source tool

  36. no published number

    Developer toolsAnnouncedModel and agent routing

    Helm

    Jev-based skill routes each coding task to the best installed agent (Claude Code, Codex, Cursor, Gemini CLI) and judges task completion.

    17 stars · Python · Open-source tool

  37. no published number

    Developer toolsIn productionModel and agent routing

    Jev Router (hajdu-patrik)

    Picks model, reasoning effort, parallel agent count and skill before every prompt to Claude Code, Codex or Antigravity, using Jev (hajdu-patrik).

    15 stars · Python · Open-source tool

  38. no published number

    Developer toolsIn productionModel and agent routing

    Vybrid

    Optional Jev routing classifies each turn's task type and dispatches it to a prep, coding or cheap chat model inside this Rust coding CLI.

    12 stars · Rust · Open-source tool

  39. no published number

    Developer toolsBetaModel and agent routing

    pi-jev-model-router

    A Pi extension asks Jev four questions per prompt to pick a budget-aware model tier before each turn, with automatic tier fallback.

    11 stars · TypeScript · Open-source tool

  40. no published number

    Developer toolsAnnouncedModel and agent routing

    jev-pilot

    Claude Code plugin asks Jev typed questions each turn to set reasoning effort, subagent model and skill.

    10 stars · TypeScript · Open-source tool

  41. no published number

    Developer toolsAnnouncedModel and agent routing

    pi-model-router

    Adds optional Jev advice to a four-tier model router inside the Pi coding agent, picking which model and reasoning effort handles each turn.

    6 stars · TypeScript · Open-source tool

  42. no published number

    Developer toolsIn productionModel and agent routing

    coderelay

    CLI router for local coding-agent CLIs whose opt-in jev mode sends the task prompt and installed candidates to Jev to pick whether Claude Code, Codex, pi or omp runs it; the default mode uses local rules and scoring.

    5 stars · TypeScript · Open-source tool

  43. no published number

    Developer toolsBetaModel and agent routing

    jev-router (FrancoisChastel)

    Jev judges which coding-agent turns the cheapest model can finish, routing each turn and logging cost against always-cheap and always-frontier baselines.

    5 stars · TypeScript · Open-source tool

  44. no published number

    Developer toolsBetaModel and agent routing

    opencode-auto-jev

    OpenCode plugin adds an Auto (Jev) virtual model so Jev picks which configured real model handles each prompt turn.

    5 stars · TypeScript · Open-source tool

  45. no published number

    Developer toolsAnnouncedModel and agent routing

    pi-pignon

    Pi coding-agent extension uses local or hosted Jev to judge prompt difficulty and pick the LLM tier from a user-defined routing table.

    5 stars · TypeScript · Open-source tool

  46. no published number

    Developer toolsIn productionModel and agent routing

    Alloy

    Skill and CLI that runs a panel of local AI coding CLIs in parallel; opt-in Jev routing sends the task text to Jev to choose which CLI model runs it, using model profiles, evidence and subscription quota.

    3 stars · Python · Open-source tool

  47. no published number

    Developer toolsAnnouncedModel and agent routing

    Jev Agent Control

    Jev routes requests between OpenCode V2's Plan, Build and custom agents, handing off unfinished work at completion boundaries.

    3 stars · TypeScript · Open-source tool

  48. no published number

    Developer toolsAnnouncedModel and agent routing

    Coding Router Jev

    Jev routes Codex, Claude Code, and Pi coding-agent turns, auto-selecting the model and reasoning effort for each turn.

    2 stars · TypeScript · Open-source tool

  49. no published number

    Developer toolsBetaModel and agent routing

    JevRouter.nvim (Mawfyy)

    Neovim plugin that sends the prompt to Jev as one choice question through the OpenRouter Decisions API and routes it to a file-read, code-edit, terminal or Q&A handler.

    2 stars · Lua · Open-source tool

  50. no published number

    AI infrastructurePilotModel and agent routing

    kfchow-llm-value-router

    Jev judges whether a turn is routine enough to route to a free model tier inside a Hermes Agent plugin, currently shipping in shadow log only mode and not yet live.

    2 stars · Python · Open-source tool

  51. no published number

    AI infrastructureBetaModel and agent routing

    Atoma

    Picks which agent takes a task and whether a plan is approved in a multi-agent work platform; early and unaudited, no numbers published.

    Commercial product

  52. no published number

    ProductivityIn productionModel and agent routing

    Balabol

    Jev detects a message's dominant source language to disambiguate text before a translation LLM runs.

    Commercial product

  53. no published number

    ProductivityIn productionModel and agent routing

    Filelathe

    Free browser file viewer in which Jev decides, for a format no built-in tool handles, whether to invent a mini-app or open a generic inspector, returned as a typed choice with a confidence.

    Commercial product

  54. no published number

    AI infrastructureIn productionModel and agent routing

    Jev Router (OpenClaw plugin)

    OpenClaw plugin that asks Jev to classify each task's model and thinking needs and choose a cost and quality profile, then overrides the run's provider and model where the host's before_model_resolve hook allows it.

    Open-source tool

  55. no published number

    Developer toolsIn productionModel and agent routing

    Jevia

    Assistant-ui's open-source router scores coding tasks with Jev to pick a capability tier and caches outcomes locally; used internally, no published numbers (company).

    Open-source tool

  56. no published number

    AI infrastructureAnnouncedModel and agent routing

    OODA AI

    Hosted AI platform with 150+ models; a vendor post says it runs Jev for model routing, context preparation, guardrails and agent-loop decisions. The linked site does not describe the Jev integration.

    Commercial product

  57. no published number

    AI infrastructureAnnouncedModel and agent routing

    OpenRouter Jev Router

    OpenRouter launched typesafe/jev-router, a cache-aware model router that uses Jev to pick the model and reasoning effort for each request across LLM calls.

    Commercial product

  58. no published number

    Developer toolsIn productionModel and agent routing

    Photoshop MCP

    Jev picks fast in-app edits instead of invoking a full LLM call inside an MCP server for Adobe Photoshop.

    Commercial product

  59. no published number

    General-purposePilotModel and agent routing

    Procurement Mind

    A public procurement assistant ran Jev live alongside its LLM planner for a first week; author reports promising latency and cost on a still-small sample (author).

    Write-up

  60. no published number

    ProductivityIn productionModel and agent routing

    Shapeshift

    Text box that sends each debounced input to a server route where one Jev call answers typed questions in parallel, such as which card fits and whether it is urgent, then deterministic code parses values and renders the card.

    Open-source tool

  61. no published number

    Games and entertainmentIn productionModel and agent routing

    Tutor

    Plain-English Magic: The Gathering card search in which Jev decides whether a query needs an added reasoning layer before results are ranked by fit; searches go to TypeSafe and Anthropic.

    Commercial product

  62. no published number

    AI infrastructureIn productionModel and agent routing

    UsageTap Model Lifecycle

    Classifies LLM model keys and recommends safe replacement models for deprecated ones with Jev and heuristics, via a live GET API and GitHub Action (UsageTap).

    Commercial product

02Message and ticket triage21

  1. ~580 conversations/secmeasured

    ProductivityBetaMessage and ticket triage

    One (Inbox Run)

    Open-source demo from One in which Jev answers four typed questions about every Gmail conversation, and One labels, archives, logs leads to HubSpot and Sheets and posts to Slack. Classified ~580 conversations/sec (company).

    12 stars · TypeScript · Open-source tool

  2. $0.00003measured

    Customer supportIn productionMessage and ticket triage

    Jev Orchestrator

    Jev reads each incoming Gmail, Slack, Discord, Typeform or X message through OpenRouter, picks one of six groups and rates FAQ coverage and risk; code rules then let only the matching agent draft a reply. About $0.00003 per message (author).

    9 stars · JavaScript · Open-source tool

  3. 70-500msmeasured

    IT operationsBetaMessage and ticket triage

    Jevops

    SRE log triage tool routes Elasticsearch log bursts through Jev to decide whether to page, digest or resolve; author cites 70-500ms latency (author).

    8 stars · Python · Open-source tool

  4. 300 msmeasured

    ProductivityIn productionMessage and ticket triage

    SipPulse Meet Capture

    Jev classifies each live Google Meet transcript line for intent, sentiment and action items in about 300 ms (company).

    5 stars · JavaScript · In-house system

  5. 0.87measured

    IT operationsAnnouncedMessage and ticket triage

    jev-oncall

    One Jev call per alert scores actionability, severity, owner and duplicates for incident triage; a demo shows a 0.87-matched duplicate routed to review instead of paging (jev-oncall).

    4 stars · HTML · Open-source tool

  6. 78%measured

    Developer toolsIn productionMessage and ticket triage

    hush (emreozyoruk)

    GitHub Action asks Jev four calibrated questions per new issue (label, spam, duplicate, needs info) and applies a label only above maintainer-set thresholds, otherwise stays silent. 78% precision on 800 closed issues at the default threshold (author).

    2 stars · JavaScript · Open-source tool

  7. 424 msvs a baseline

    ProductivityBetaMessage and ticket triage

    openpoke-meets-jev (0xShin0221)

    Fork of OpenPoke where Jev screens incoming email for importance and prompt injection, checks execution-agent tool calls, and filters search results before the LLM runs. Mean 424 ms per decision in an A/B against Claude Sonnet (author).

    1 stars · Python · Open-source tool

  8. 19/20measured

    E-commerce and retailIn productionMessage and ticket triage

    jev-inbox-triage

    Jev decision model sorts a store's incoming emails into routing categories inside an email inbox pipeline. 19/20 emails routed correctly (author).

    0 stars · HTML · Open-source tool

  9. 74%measured

    ProductivityAnnouncedMessage and ticket triage

    Surely – AI Triage (Devcore)

    Jev triages monday.com board items by status, priority and spam inside the Surely app; vendor demo claims 74% fully automatic sorting (Devcore).

    Commercial product

  10. 91%vs a baseline

    IT operationsIn productionMessage and ticket triage

    Swamp (Watson Labs)

    Jev answers one question per firing alert, given the alert and the last 20 signals: is this a real, actionable production problem? Below the threshold Swamp dismisses the alert; above it, Claude triages. 91% cheaper alert triage (author).

    Write-up

  11. no published number

    Customer supportAnnouncedMessage and ticket triage

    jev-call-router

    TypeScript service uses Jev to classify caller intent from speech and route PBX calls to predefined destinations, with low-confidence calls falling back.

    18 stars · HTML · Open-source tool

  12. no published number

    IT operationsBetaMessage and ticket triage

    jevernetes

    Rust CLI and terminal UI stream Kubernetes logs, apply offline rules, then send redacted batches to Jev, which sorts entries into Important, Routine and Needs Review tabs; collection is read-only and judgments are advisory.

    15 stars · Python · Open-source tool

  13. no published number

    Developer toolsIn productionMessage and ticket triage

    tia

    hey-tia[bot] GitHub App triages new issues live via Vercel Connect, using Jev for the classification decision behind each triage action.

    8 stars · TypeScript · Open-source tool

  14. no published number

    ProductivityIn productionMessage and ticket triage

    Career Journal

    Local job-search tracker whose daily run sends every message from the selected mailboxes for the previous day to Jev, with no keyword prefilter, to decide job-search relevance; the coding agent's LLM does the review when Jev is not configured.

    6 stars · JavaScript · Open-source tool

  15. no published number

    ProductivityIn productionMessage and ticket triage

    Notiq

    Scores whether each Android notification matches a user's plain-language filtering rules, using Jev or self-hosted FastJev (Notiq).

    4 stars · Kotlin · Open-source tool

  16. no published number

    IT operationsBetaMessage and ticket triage

    RuleRaven

    Read-only Kubernetes controller that runs deterministic rules over Pod, Event and workload state and calls Jev only for ambiguous signals; a deterministic composer then sets severity and action and sends a signed webhook.

    2 stars · Go · Open-source tool

  17. no published number

    Customer supportIn productionMessage and ticket triage

    Atisbo

    Atisbo turns support tickets, sales calls and meetings into a ranked list of problems; Jev classifies each piece of customer feedback as a problem, idea, question or noise inside that pipeline.

    Commercial product

  18. no published number

    Education and researchIn productionMessage and ticket triage

    IFNMG News Bot

    IFNMG's live Telegram bot uses Jev to classify RSS news posts into subscriber-chosen categories before routing them to chats.

    In-house system

  19. no published number

    Customer supportIn productionMessage and ticket triage

    Klaro Cards

    SMB operations hub routes incoming support tickets to the right department with TypeSafe's Jev, as a feature of its email routing module.

    Commercial product

  20. no published number

    ProductivityIn productionMessage and ticket triage

    Replaceme

    Jev answers narrow yes/no questions per message so a Mac menu bar app can compile one alert brief.

    Commercial product

  21. no published number

    Customer supportIn productionMessage and ticket triage

    YStone Support

    Jev powers badge reads on YStone's live central support portal for its games and apps.

    In-house system

03Agent tool and action selection61

  1. $0.0002vs a baseline

    ProductivityBetaAgent tool and action selection

    typesafe-computer-use

    Mac computer-use agent that reads the screen through OCR and accessibility data, then asks Jev a Choice over the numbered on-screen items for the next action, calling a writing model only for free text. $0.0002 per decision (author).

    1,220 stars · Python · Open-source tool

  2. ~5–10×vs a baseline

    Developer toolsIn productionAgent tool and action selection

    EZCollegeApp (via jev-browser-use)

    Codex skill where Jev picks the next click, toggle, scroll or navigation from the page's accessibility text inside Codex Computer Use, while Codex types, reads visuals and verifies the result. ~5–10× faster browser operations (author).

    1,068 stars · JavaScript · Open-source tool

  3. 25× cheapervs a baseline

    AI infrastructureIn productionAgent tool and action selection

    system1-agents

    Prebuilt openJiuwen agents in which Jev picks the next browser click, desktop action, game move or support queue from typed options while a chat model plans; installs as a Claude Code or Codex skill and MCP server. Up to 25× cheaper than a chat model (author).

    132 stars · Python · Open-source tool

  4. about 0.3–1.5 s per stepmeasured

    ProductivityBetaAgent tool and action selection

    jev-use (savka777)

    Voice-driven macOS computer use where Jev picks the next operation and target from the Accessibility tree, with no screenshots, and the app executes and re-reads the screen. About 0.3–1.5 s per step (author).

    122 stars · Swift · Open-source tool

  5. 41/42vs a baseline

    Developer toolsBetaAgent tool and action selection

    fastbrowse

    Browser agent that indexes each page into candidate controls and has Jev pick the next action, while an LLM plans and reads and code gates secrets and irreversible clicks. It passed 41/42 tasks against 39/42 for the Browser Use agent (author).

    114 stars · Python · Open-source tool

  6. 89.6-91.5%vs a baseline

    Developer toolsAnnouncedAgent tool and action selection

    Jev GUI Delegate

    Jev makes the step decisions in this Windows and Chrome GUI delegate for Codex, cutting main-model tokens 89.6 to 91.5% across the author's two benchmark runs (Jev GUI Delegate).

    104 stars · Python · Open-source tool

  7. 3.74svs a baseline

    Developer toolsIn productionAgent tool and action selection

    QuickE2E

    Picks each browser click for end-to-end tests; hosted Jev runs median 3.74s per step versus 25.29s for Claude Code plus Playwright MCP (author).

    92 stars · JavaScript · Open-source tool

  8. 458msmeasured

    Developer toolsBetaAgent tool and action selection

    jev-ui-test

    A browser-use fork drives UI test automation, scoring candidate elements with Jev; author reports 458ms median per action decision (author).

    55 stars · Python · Open-source tool

  9. 0.88 secondsmeasured

    ProductivityAnnouncedAgent tool and action selection

    partway

    Vendor claims this macOS voice-control app calls Jev on every spoken word, about 200ms, and opened Notes 0.88 seconds before the user finished speaking (author).

    31 stars · Swift · Open-source tool

  10. half a secondmeasured

    Developer toolsAnnouncedAgent tool and action selection

    Jev_steer_or_queue

    Claude Code hook lets Jev decide in about half a second whether a mid-turn message should steer, queue, or interrupt the running task (author).

    16 stars · Python · Open-source tool

  11. 85.9%measured

    Developer toolsIn productionAgent tool and action selection

    jev-for-all

    OpenCode plugin, with Claude Code and Hermes adapters, where Jev picks the skill to load, the tool subset for each step and each browser move via OpenRouter's Decisions API, failing open; skill routing hit rate 85.9% (author).

    15 stars · TypeScript · Open-source tool

  12. 15measured

    FinanceAnnouncedAgent tool and action selection

    trade-jev

    Backtests Jev as a BUY, SELL or HOLD trading decision maker against 15 days of real NQ order book data from Databento (justinhe16).

    12 stars · Python · Write-up

  13. 300msmeasured

    General-purposeIn productionAgent tool and action selection

    Home Assistant TypeSafe Conversation Agent

    Home Assistant voice agent asks Jev typed yes/no questions per command to resolve intent instead of using an LLM, resolving commands in about 300ms (the-sof).

    7 stars · Python · Open-source tool

  14. 35svs a baseline

    Developer toolsIn productionAgent tool and action selection

    jev-playwright (cooper667)

    Jev selects page elements and judges pass or fail checks for plain-English browser QA steps; vendor claims 24 steps run in 35 seconds versus 14 minutes manually (author).

    7 stars · TypeScript · Open-source tool

  15. 100% task completion vs 94% for Playwright alonevs a baseline

    Developer toolsBetaAgent tool and action selection

    jev-browser-wingman

    Jev picks each browser-automation step via CDP; claimed 100% task completion versus 94% for Playwright alone over 34 runs (author).

    5 stars · TypeScript · Open-source tool

  16. 100 applications submitted in under 10 minutes for about one dollarmeasured

    ProductivityIn productionAgent tool and action selection

    jev-job-search

    Jev decides and verifies job-application form answers during automated browser submissions. 100 applications submitted in under 10 minutes for about one dollar (author).

    5 stars · TypeScript · Open-source tool

  17. 3.96x to 8.5xvs a baseline

    Developer toolsAnnouncedAgent tool and action selection

    jev-ra

    Vendor claims Jev-driven single-round-trip browser navigation beat browser-use with gemini-flash by 3.96x to 8.5x on real tasks, e.g. 2.7s vs 23s on Wikipedia (author).

    4 stars · Python · Open-source tool

  18. 0.3smeasured

    Developer toolsIn productionAgent tool and action selection

    CloakBrowser Agent

    Jev picks the next browser automation step, click or fill action, with a probability score, in about 0.3s (CloakHQ).

    3 stars · Python · Open-source tool

  19. 98.8%vs a baseline

    Developer toolsBetaAgent tool and action selection

    toolJev

    Jev replaces LLM-based tool selection in MCP agents across 612 tools; author reports 98.8% accuracy on 400 tickets (author).

    3 stars · Python · Open-source tool

  20. 58%vs a baseline

    Developer toolsBetaAgent tool and action selection

    POK-Agent

    Jev serves as a fast System 1 layer that picks quick actions for a Windows computer-use agent before a slower planner runs; vendor claims 58% task success on replayed tasks (author).

    1 stars · Rust · Open-source tool

  21. 273msvs a baseline

    Developer toolsIn productionAgent tool and action selection

    Empryo

    Empryo's coding agent uses Jev, off by default, for skill suggestion, search re-ranking, retry triage on provider errors, an approval guard on UI clicks and next-action picks in its UI pilot. 273ms median on failure triage (company).

    Commercial product

  22. 3 MIT Integration Bee 2026 qualifying questionsmeasured

    Education and researchIn productionAgent tool and action selection

    Fieldnote (ClawScience)

    Web physics and math solver where Jev picks the topic, the next equation or move, and whether an assumption is safe from option lists, while ordinary code does and checks all algebra. Solved 3 MIT Integration Bee 2026 qualifying questions (author).

    Commercial product

  23. 27-198xvs a baseline

    Developer toolsBetaAgent tool and action selection

    Jev Browser Control

    Chrome extension and MCP server where Jev picks each click in about 0.5s; author measures the setup as 27-198x cheaper than Claude or Codex driving the browser directly (author).

    Commercial product

  24. zeromeasured

    Developer toolsAnnouncedAgent tool and action selection

    Jev Skill Typeahead (claude-code-templates)

    Jev predicts which installed Claude Code skill will be invoked before a prompt is sent; the author claims zero added Claude tokens (author).

    Open-source tool

  25. 3xvs a baseline

    Developer toolsAnnouncedAgent tool and action selection

    jev-browse (danielnc)

    PyPI package adds Jev-decided browser sub-tasks to browser-harness; author's benchmark shows a Google Flights search running about 3x faster and cheaper (author).

    Open-source tool

  26. 31 of 31measured

    ProductivityBetaAgent tool and action selection

    Jolt

    macOS launcher that sends a plain-language request and the current app's menu command names to Jev, which chooses one command for ordinary code to run; 31 of 31 short test queries found the correct command (company).

    Commercial product

  27. 64%measured

    Developer toolsIn productionAgent tool and action selection

    SuperQode

    Superagentic AI's coding-agent harness layer uses Jev to rank and activate tool schemas progressively and to score a closed set of tool choices once per turn, plus Choice, Score and Noul judgments; 64% average catalogue reduction (company).

    Open-source tool

  28. no published number

    Developer toolsAnnouncedAgent tool and action selection

    browser-use jev-ultrafast

    Jev picks the next browser action for this official browser-use agent, marketed as the fastest and cheapest option, with no numbers given (browser-use).

    22,603 stars · Python · Open-source tool

  29. no published number

    Developer toolsIn productionAgent tool and action selection

    MOMO CODE

    Jev powers a decision layer in MOMO CODE's v2.0 coding agent, shipped alongside a Pi framework integration and an RSI self-evolution loop.

    1,076 stars · TypeScript · Open-source tool

  30. no published number

    IT operationsIn productionAgent tool and action selection

    NetClaw

    NetClaw, a network AI agent, ships an optional Jev science officer decision role among its 233 skills to guide network investigations (author).

    678 stars · Python · Open-source tool

  31. no published number

    Developer toolsAnnouncedAgent tool and action selection

    Codify (Sidiora-Labs)

    Single-binary agent workflow engine uses Jev for its one remote decision call inside a fleet-scale multi-branch indexing and merge/PR pipeline.

    434 stars · C · Open-source tool

  32. no published number

    Developer toolsBetaAgent tool and action selection

    Jev DSH Decision (Devin-AXIS)

    Plugin for DeepSeek Harness and iPolloWork where Jev recommends a tool, skill or agent owner and scores output quality, returning choices with probabilities while the host agent executes.

    253 stars · JavaScript · Open-source tool

  33. no published number

    Media and socialAnnouncedAgent tool and action selection

    jev-social

    Jev picks each next social-research action, such as search, open post, or read comments, that this tool executes in a real browser, producing a source-linked report.

    163 stars · JavaScript · Open-source tool

  34. no published number

    FinanceBetaAgent tool and action selection

    Warren Duffer

    Intraday bot where Jev, via Vercel AI Gateway, classifies and ranks Nifty 50 stocks in two stages and decides take-profit exits, while code sizes positions, sets stops and places live MIS orders through Kotak Neo or Zerodha Kite.

    98 stars · TypeScript · Open-source tool

  35. no published number

    ProductivityBetaAgent tool and action selection

    Jev Desktop

    Codex Computer Use plugin where Jev selects an operation and target from an allowlisted accessibility snapshot in browser tabs and macOS apps; Computer Use executes and Codex verifies the result.

    76 stars · JavaScript · Open-source tool

  36. no published number

    IT operationsIn productionAgent tool and action selection

    Reflex (Datadog Labs)

    Recommends operational actions like circuit breaking or autoscaling on Datadog metrics using Jev, executed only if separate guards pass, in this Rust control-loop SDK (Datadog Labs).

    31 stars · Rust · Open-source tool

  37. no published number

    Developer toolsIn productionAgent tool and action selection

    Flick

    Runs an observe-decide-act loop with Jev to execute whole browser and macOS goals and returns verified outcomes, via a local stdio MCP server (Flick).

    26 stars · TypeScript · Open-source tool

  38. no published number

    Developer toolsBetaAgent tool and action selection

    Forma

    Forma website builder uses Jev to pick design settings, component variants and section order from a natural language description of the desired site.

    26 stars · JavaScript · Open-source tool

  39. no published number

    Developer toolsAnnouncedAgent tool and action selection

    jev-skill-router

    A Claude Code hook asks Jev which installed skill fits a prompt; author published it as a negative result, finding it unlikely to help a strong model.

    15 stars · Python · Open-source tool

  40. no published number

    Developer toolsIn productionAgent tool and action selection

    Optimus

    Rust coding harness with opt-in Jev feature gates that withdraw or filter tools, adjust reasoning effort, filter context, memory and code-search results, and judge loop control, retries and verification; the host validates every decision.

    14 stars · Rust · Open-source tool

  41. no published number

    Developer toolsIn productionAgent tool and action selection

    jev-agent-browser

    npm CLI and Node library in which Jev chooses the next typed browser action from a page snapshot and agent-browser executes it, handing ambiguous, repetitive or stuck states back to the parent agent.

    13 stars · JavaScript · Open-source tool

  42. no published number

    Developer toolsIn productionAgent tool and action selection

    Sedum

    Jev's Choice selects the target page element for plain-English Playwright steps, while a Noul model verifies test claims, cheap enough to run on every pull request.

    11 stars · TypeScript · Open-source tool

  43. no published number

    Developer toolsAnnouncedAgent tool and action selection

    jev-sim-use

    Navigates iOS Simulator and Android screens by sending on-screen labels to Jev each step, replacing a full LLM agent turn with one Jev call per tap.

    10 stars · Swift · Open-source tool

  44. no published number

    E-commerce and retailBetaAgent tool and action selection

    Hearth

    Browser agent built on Jev Ultrafast drives a local Chrome across Craigslist, Facebook Marketplace, Redfin and Zillow from one plain-language rental request, with Jev deciding each step, and returns one shortlist; it only reads pages.

    8 stars · JavaScript · Open-source tool

  45. no published number

    ProductivityBetaAgent tool and action selection

    agent0

    Desktop agent where optional Jev routing selects a relevant tool before each main-model step, with confidence-based fallback to the Kimi or OpenAI model; the Activity panel logs each decision, confidence and latency.

    5 stars · TypeScript · Open-source tool

  46. no published number

    Developer toolsAnnouncedAgent tool and action selection

    jev-router (Ex8-ca)

    Hermes plugin uses Jev to pick the best-matching skill each turn and pre-load top skills at session start.

    5 stars · Python · Open-source tool

  47. no published number

    FinanceBetaAgent tool and action selection

    Bob Trades

    Bob Trades, a mobile app, lets users connect their own Bybit account and has Jev decide buy or hold on spot trades, gated by separate risk checks (author).

    4 stars · TypeScript · Commercial product

  48. no published number

    Developer toolsAnnouncedAgent tool and action selection

    System One Browser Agent

    Browser agent combines sub-150ms Jev reflex decisions with Stagehand automation and an LLM fallback for harder steps; architecture described, no benchmark reported yet (author).

    4 stars · TypeScript · Open-source tool

  49. no published number

    FinanceBetaAgent tool and action selection

    jev-polymarket-trading

    Terminal trading agent asks Jev whether BTC finishes up or down each 5-minute Polymarket window, then abstains or buys once; no win rate published.

    2 stars · TypeScript · Open-source tool

  50. no published number

    ProductivityIn productionAgent tool and action selection

    Switchyard

    Jev decides which browser profile opens a link when no learned rule applies, in this macOS menu-bar router for Dia and Chrome.

    2 stars · Swift · Open-source tool

  51. no published number

    Data and analyticsIn productionAgent tool and action selection

    Analisa USA

    Jev routes natural-language queries to map views, flags low-competition federal contracts, and fact-checks generated answers in this public civic data tool.

    Commercial product

  52. no published number

    ProductivityAnnouncedAgent tool and action selection

    Caret

    Decides per frame whether to abstain, offer an inline edit, or propose an action in a Mac autocomplete app; hackathon first-place winner, no numbers published.

    Commercial product

  53. no published number

    Developer toolsIn productionAgent tool and action selection

    evoke

    CLI, package manager and TypeScript SDK where Jev reads a plain-language sentence, picks which installed reflex program runs and extracts its arguments, requiring higher confidence for write actions than for read-only ones.

    Open-source tool

  54. no published number

    Developer toolsIn productionAgent tool and action selection

    FastVibe

    Optional decision engine picks browser and computer-use control actions for this cross-platform desktop coding agent app, using Jev (FastVibe).

    Commercial product

  55. no published number

    Developer toolsBetaAgent tool and action selection

    harness-router

    Framework-agnostic System-1 tool-selection layer sends agent harness state to Jev via OpenRouter to pick the next tool call, falling back to the planner when uncertain.

    Open-source tool

  56. no published number

    ProductivityIn productionAgent tool and action selection

    Hermes Agent

    Hermes Agent runs continuously in Docker and uses Jev for reflex triage of job listings, deciding which to pass on to LLM extraction and form filling.

    0 stars · Python · Open-source tool

  57. no published number

    FinancePilotAgent tool and action selection

    jev-trader

    Jev decides buy or sell every ~300ms against live Kuru and Hyperliquid order books on this public trading dashboard, with fills simulated by default (author).

    Commercial product

  58. no published number

    Developer toolsPilotAgent tool and action selection

    jevopt

    Experiment that asks Jev whether to inline at each LLVM IR call site to shrink compiled binaries; the author reports it sometimes beats clang -Oz on binary size, with no figures in the post.

    Write-up

  59. no published number

    FinancePilotAgent tool and action selection

    Orus

    Approves, refuses, or holds proposed trades against strategy rules with Jev; live in paper trading for Premium subscribers (Orus).

    Commercial product

  60. no published number

    E-commerce and retailIn productionAgent tool and action selection

    Rummage TV

    A Jevathon hackathon runner-up shipped an eBay-browsing 90s shopping-channel app to production; the team says more Jev-driven logic runs under the hood (company).

    Commercial product

  61. no published number

    AI infrastructureIn productionAgent tool and action selection

    The One

    Multi-agent workshop app where a conventional-LLM agent can spawn Jev-powered subagents for computer use and stock-trading workflows.

    Commercial product

04Agent guardrails and approvals46

  1. 25 of 25vs a baseline

    Developer toolsBetaAgent guardrails and approvals

    Canny

    Hooks Claude Code and Codex CLI; Jev answers whether an edit breaks a CLAUDE.md or AGENTS.md rule and whether a message claims done, as notes only, while ledger checks block. A 25-pair Opus 5 A/B had 25 of 25 passes in both arms (author).

    122 stars · TypeScript · Open-source tool

  2. $0.00004measured

    Developer toolsIn productionAgent guardrails and approvals

    Jev Guard (leepokai)

    Risk-scores every coding-agent tool call as deny, ask, or allow with Jev across Claude Code, Codex, Copilot, Gemini and Cursor; vendor claims about $0.00004 per call (leepokai).

    63 stars · JavaScript · Open-source tool

  3. 0.0008 USDvs a baseline

    Developer toolsIn productionAgent guardrails and approvals

    jev-kit

    Claude Code PreToolUse hook where Jev judges tool calls its rules table leaves open, then allows, warns, rewrites or blocks; Jev also picks each browser-agent click, at 0.0008 USD of Claude spend a run vs 0.1868 USD with Sonnet (author).

    63 stars · Python · Open-source tool

  4. 2x faster decisions and zero unsafe allows across 93 labeled test casesvs a baseline

    Developer toolsIn productionAgent guardrails and approvals

    jev-permission-gate

    Jev replaces Claude Code's auto-mode tool-call classifier, judging whether each tool call is safe to run automatically. 2x faster decisions and zero unsafe allows across 93 labeled test cases (author).

    29 stars · TypeScript · Open-source tool

  5. $0.00002 per requestmeasured

    SecurityAnnouncedAgent guardrails and approvals

    dsh-jev-interceptor

    Vendor claims Jev classifies risk, irreversibility and injection-suspicion for every DeepSeek Harness tool call at about $0.00002 per request (author).

    27 stars · TypeScript · Open-source tool

  6. 25/25 fixture agreementmeasured

    Developer toolsBetaAgent guardrails and approvals

    dsh-jev

    DeepSeek Harness plugin where Jev narrows which tools are relevant, gates whether a tool call is safe to run and picks a model route; its answers can only narrow or hold. A live API check gave 25/25 fixture agreement (author).

    25 stars · TypeScript · Open-source tool · archived

  7. 77%measured

    SecurityBetaAgent guardrails and approvals

    toolgate

    Jev scores risk of each agent tool call before execution; an independent reviewer's test flagged a folder deletion at 77% destructive risk for confirmation (author).

    24 stars · TypeScript · Open-source tool

  8. $0.0002vs a baseline

    Developer toolsBetaAgent guardrails and approvals

    Software Factory

    A typesafe pipeline stage sends the diff to Jev with typed questions and routes on declared thresholds, parking the request for human review when a gate such as the secret check fails (company).

    11 stars · Python · Open-source tool

  9. 93.3%measured

    Developer toolsIn productionAgent guardrails and approvals

    jev-enforce

    Claude Code plugin whose Stop and PostToolUse hooks ask Jev one yes/no question per CLAUDE.md rule about each reply and file edit, then block with the broken rules quoted. 93.3% precision and recall at the default threshold on 59 labelled examples (author).

    10 stars · TypeScript · Open-source tool

  10. 13 of 13measured

    Developer toolsBetaAgent guardrails and approvals

    jev-engineering (eugeniughelbur)

    Jev routes code review and blocks risky coding-agent test commands; its review-router caught 13 of 13 CVE fixes (author).

    9 stars · Python · Open-source tool

  11. ~0.8–1.5svs a baseline

    Developer toolsBetaAgent guardrails and approvals

    ZCode Gatekeeper

    External approval hook for the ZCode coding agent in which Jev, as an optional reviewer backend, rates each gray-zone tool call against the current task as allow, uncertain or deny. Measured at ~0.8–1.5s per decision (author).

    9 stars · JavaScript · Open-source tool

  12. 9/9measured

    SecurityBetaAgent guardrails and approvals

    Jev Guard Harness

    Local Go gateway checks each coding-agent tool call against deterministic policy, asks Jev for risk and confidence on ambiguous calls, returns allow, ask or block, and fails closed when Jev is unreachable; policy fixtures pass 9/9 (author).

    6 stars · Go · Open-source tool

  13. 81.8%measured

    Developer toolsBetaAgent guardrails and approvals

    stingray

    Claude Code Stop hook that asks Jev whether a turn ended without acting, broke a declared promise, promised unwatched monitoring or used the wrong language, then blocks the exit with a nudge. 81.8% precision at full payload on labelled turns (author).

    6 stars · Shell · Open-source tool

  14. 39measured

    Developer toolsBetaAgent guardrails and approvals

    oh-my-plumb

    Jev checks every coding-agent edit against AGENTS.md rules; a replay of 93 real sessions flagged 39 edits (author).

    5 stars · TypeScript · Open-source tool

  15. sub-150msmeasured

    AI infrastructureBetaAgent guardrails and approvals

    JevGuard (SakshiKumari271)

    Jev arbitrates streaming RAG output against retrieved context in a Streamlit harness; author claims sub-150ms execution per check (author).

    2 stars · Python · Open-source tool

  16. 90.4%measured

    SecurityIn productionAgent guardrails and approvals

    DSH Jev Guard

    DeepSeek Harness plugin on tools/pre-execute asks Jev whether a shell command will irreversibly delete or overwrite real data, after offline hard rules, and maps the probability to allow, revise or block; 90.4% on a 114-case calibration set (author).

    0 stars · JavaScript · Open-source tool

  17. no published number

    Developer toolsIn productionAgent guardrails and approvals

    Foreman

    Foreman, an agent supervisor and software factory foreman, is built directly on Jev to route tasks and check agent work (author).

    737 stars · Python · Open-source tool

  18. no published number

    Developer toolsBetaAgent guardrails and approvals

    deepseek-harness-jev (luobosibing2)

    A DeepSeek Harness plugin uses Jev for agent and skill selection, supervision, corrections and approvals across 11 togglable features, tested on DSH 0.1.7-rc.2 (author).

    203 stars · JavaScript · Open-source tool

  19. no published number

    Developer toolsBetaAgent guardrails and approvals

    jevmem

    jevmem v0.5 adds a Jev-checked guard that screens memory lines not written locally for poisoning before Claude Code reads them (jevmem).

    121 stars · TypeScript · Open-source tool

  20. no published number

    Developer toolsIn productionAgent guardrails and approvals

    Nerve

    Hermes Agent plugin in which Jev judges unresolved completion criteria and forecasts a worker's trajectory at checkpoints, returning a replan or continue directive while the orchestrator keeps the final stop decision; also runs on Laya or OpenJev.

    38 stars · Python · Open-source tool

  21. no published number

    Developer toolsBetaAgent guardrails and approvals

    Pi Jev Guard (zszz3)

    Pi coding-agent plugin where Jev judges each shell, write and edit call for destructive action, data exfiltration, task drift and rule conflicts before it runs, alongside local rules for credentials and risky commands.

    21 stars · TypeScript · Open-source tool

  22. no published number

    Developer toolsIn productionAgent guardrails and approvals

    Jevvy

    OpenCode and Claude Code plugin that sends unresolved shell permission requests to Jev as four Noul questions (harmful, sensitive, untrusted, obscured) and auto-approves only when all pass calibrated thresholds; otherwise the normal prompt runs.

    20 stars · TypeScript · Open-source tool

  23. no published number

    Developer toolsIn productionAgent guardrails and approvals

    Command Code

    Official Command Code mod that asks Jev at every agent stop whether a nudge would help, whether the agent is waiting on the user and whether the last nudge made progress, then continues the run or lets it end.

    17 stars · TypeScript · Commercial product

  24. no published number

    Developer toolsIn productionAgent guardrails and approvals

    DSH VSC Integration

    Jev drives skill and file selection plus supervision and approvals for a Laya System One plugin bundled in a VS Code extension.

    17 stars · TypeScript · Open-source tool

  25. no published number

    SecurityIn productionAgent guardrails and approvals

    jes

    Open-source agent guardrails classify prompts, tool calls and responses using a decision model such as Jev instead of an LLM, to block prompt injection and data leaks (jes).

    14 stars · Python · Open-source tool

  26. no published number

    SecurityBetaAgent guardrails and approvals

    Jev LLM Guard

    Jev scores user, tool and model text against OWASP LLM Top 10 risks, returning allow, review or block, with a live Turkish and English demo.

    14 stars · JavaScript · Open-source tool

  27. no published number

    Developer toolsIn productionAgent guardrails and approvals

    jev-blindspot

    Flags blind spots in a submitted prompt before a coding agent acts, using a TypeSafe key inside a side panel for Claude Code and Codex CLI (jev-blindspot).

    12 stars · TypeScript · Open-source tool

  28. no published number

    Developer toolsIn productionAgent guardrails and approvals

    pi-heed

    Jev checks each side-effecting tool call in the pi coding agent against the user's stated constraints before it runs.

    12 stars · TypeScript · Open-source tool

  29. no published number

    SecurityAnnouncedAgent guardrails and approvals

    Ring Zero Security

    Enforces OS-kernel-level policy for AI coding agents and can call Jev as an optional pre-execution check alongside its syscall-level controls.

    11 stars · Rust · Open-source tool

  30. no published number

    Developer toolsIn productionAgent guardrails and approvals

    Yardsort

    Jev tags agent-written diffs as off-task, secret-leaking, test-weakening or CI-disabling inside this desktop multi-agent IDE, with tunable detection thresholds.

    10 stars · Rust · Open-source tool

  31. no published number

    Developer toolsIn productionAgent guardrails and approvals

    any-auto

    Auto-approval daemon for Antigravity and Pi coding agents; after local allowlist and blocklist rules, Jev answers Choice questions on risk and user authorization and a shared policy with a probability gate allows or denies the tool call.

    8 stars · Rust · Open-source tool

  32. no published number

    Developer toolsBetaAgent guardrails and approvals

    jev-guard (muratcakmak)

    Claude Code plugin that sends batched yes/no questions to Jev on each edit, shell command and finished answer, denying rule-breaking edits or unrequested deploys and nudging unsupported claims; it fails open when Jev is unreachable.

    7 stars · TypeScript · Open-source tool

  33. no published number

    Developer toolsBetaAgent guardrails and approvals

    pi-auto-reviewer

    A Pi extension tiers shell commands and routes ambiguous ones to Jev instead of an LLM for cheaper auto-review decisions.

    7 stars · TypeScript · Open-source tool

  34. no published number

    Developer toolsIn productionAgent guardrails and approvals

    pi-automode-classifier

    Classifies shell commands with Jev, or local Kev/Laya, to auto-run, confirm, or block tool calls before execution in the Pi coding agent.

    6 stars · TypeScript · Open-source tool

  35. no published number

    Customer supportBetaAgent guardrails and approvals

    DutyGate

    Jev answers a batch of yes/no questions on inbound support messages to flag legal and compliance obligations before a chatbot replies.

    4 stars · Python · Open-source tool

  36. no published number

    Developer toolsIn productionAgent guardrails and approvals

    pi-reply-guard

    Pi agent plugin uses Jev by default to score each reply against installed skills on a 0-1 scale and triggers a rewrite when the score falls short.

    4 stars · TypeScript · Open-source tool

  37. no published number

    Developer toolsBetaAgent guardrails and approvals

    pi-system-one-auto (lunaroute)

    Jev answers one question per safety rule to allow, block, or ask before each tool call the Pi coding agent executes, with safety behavior benchmarked using DSPy.

    4 stars · HTML · Open-source tool

  38. no published number

    Developer toolsAnnouncedAgent guardrails and approvals

    HearMemory

    Coding-agent memory plugin uses Jev to check agent claims such as 'tests pass' against recorded diffs and test runs before trusting them.

    3 stars · Python · Open-source tool

  39. no published number

    SecurityAnnouncedAgent guardrails and approvals

    grok-jev-guard

    Preflight layer for Grok Bot has Jev judge intent, risk and scope before approving, scoping down, or blocking a pending tool sequence.

    2 stars · Python · Open-source tool

  40. no published number

    Developer toolsAnnouncedAgent guardrails and approvals

    ForgeLoop

    Requires Jev as the bounded fast-judgment input for lifecycle gates and evidence-backed verification steps in this engineering protocol for AI coding agents (ForgeLoop).

    1 stars · JavaScript · Open-source tool

  41. no published number

    Developer toolsBetaAgent guardrails and approvals

    Bearing

    Linter that runs as a coding-agent hook: after each agent turn it sends the diff and AGENTS.md or skill rules written in words to Jev, and feeds the probability-scored results back to the agent as corrections.

    Open-source tool

  42. no published number

    Developer toolsIn productionAgent guardrails and approvals

    iCoderAgent

    macOS coding agent that, given a TypeSafe key, sends jev is, pick and rate questions to Jev and uses Jev for three agent decisions: which specialist role fits a subtask, whether a command will destroy something, and whether finished work works.

    Commercial product

  43. no published number

    Developer toolsIn productionAgent guardrails and approvals

    Mjolnir

    Coding-agent session control plane that by default sends recent prompt and reply text to Jev through its public proxy to judge whether a turn is waiting for the user, finished or still working, and to auto-continue requested work.

    Open-source tool

  44. no published number

    Developer toolsAnnouncedAgent guardrails and approvals

    osenv

    Jev judges every file write and command before a coding agent executes it, denying dangerous actions and routing corrections to matching models.

    Commercial product

  45. no published number

    Developer toolsIn productionAgent guardrails and approvals

    Revolve

    Open-source coding agent with an opt-in Jev block: Jev judges commands that would otherwise run silently and escalates low-confidence ones to a prompt, picks models for workers, and shapes or writes permission rules.

    Open-source tool

  46. no published number

    ProductivityIn productionAgent guardrails and approvals

    Slaide

    Slaide turns plain-text .slaide files into web, PDF and PowerPoint decks; Jev acts as the intent gate, scope enforcer and fabrication guard around its AI deck generation.

    Commercial product

05Agent context and memory36

  1. 4.70 percentage pointsvs a baseline

    Developer toolsIn productionAgent context and memory

    MemSearch (Zilliz)

    Zilliz's memory layer for coding agents optionally sends each query's ten candidate memories to Jev as independent yes/no questions and sorts by probability; Recall@5 rose 4.70 percentage points over the frozen order on 4,344 queries (company).

    2,736 stars · Python · Open-source tool

  2. 90%vs a baseline

    Developer toolsIn productionAgent context and memory

    Code Puppy

    Jev-powered smart grep replaces ripgrep for gathering code context in the coding agent, cutting LLM token usage by up to 90% per demo (author).

    852 stars · Python · Open-source tool

  3. ~100ms end to endmeasured

    Games and entertainmentIn productionAgent context and memory

    RikkaHub Plus 华灯版

    Android SillyTavern-compatible chat client in which Jev judges whether each stored memory is relevant to the conversation, replacing embedding retrieval, and answers the main model's questions through a judge tool. About 100ms end to end (author).

    100 stars · Kotlin · Open-source tool

  4. 350xvs a baseline

    ProductivityIn productionAgent context and memory

    Enzyme

    Local compiler over Markdown knowledge bases for agent retrieval; enzyme compile runs its selection step on Jev (typesafe/jev-1.13) through OpenRouter, with output in about 0.5s. 350x cheaper than frontier models (author).

    86 stars · TypeScript · Open-source tool

  5. 4.3kmeasured

    Developer toolsBetaAgent context and memory

    Hermes Tool Slimmer

    An experimental Jev mode cuts Hermes Agent tool-schema tokens to 4.3k per request, averaged over 40 prompts on a real install (author).

    39 stars · Python · Open-source tool

  6. 17/18vs a baseline

    AI infrastructureBetaAgent context and memory

    Jev Recall (samdotmak)

    Python library where Jev answers one calibrated yes/no per stored memory in a single request and keeps memories above a threshold instead of ranking by embedding similarity. Passed 17/18 benchmark requests (author).

    38 stars · TypeScript · Open-source tool

  7. 87.9%vs a baseline

    Developer toolsIn productionAgent context and memory

    pi-jev-skill-picker

    Pi extension that strips the skills catalog from the system prompt and has Jev score every skill against the task, loading the top matches on demand. On a 137-skill catalog it cut the per-request cost 87.9% on gpt-6-astra (author).

    36 stars · TypeScript · Open-source tool

  8. no published number

    AI infrastructureBetaAgent context and memory

    Astra (MatrixOrigin)

    Self-hosted agent runtime with Jev as an optional judgment layer for memory filtering, lesson dismissal, request classification and skill selection, evaluated against a no-Jev arm on synthetic memory cases.

    34 stars · Rust · Open-source tool

  9. 0.5s vs 35svs a baseline

    Developer toolsIn productionAgent context and memory

    jev-compaction-plus

    Claude Code compaction fork where Jev decides what context to keep verbatim, measured at 0.5s versus 35s compaction time on a 591k-token session (jev-compaction-plus).

    26 stars · TypeScript · Open-source tool

  10. 0.930vs a baseline

    Developer toolsBetaAgent context and memory

    AgentDir

    Coding-agent flight recorder can send a task and redacted local passages to Jev, which reranks them and drops low-relevance ones before the agent's briefing; on a synthetic benchmark nDCG@5 reached 0.930, above local selection (author).

    23 stars · Python · Open-source tool

  11. 91.67%→95.83%vs a baseline

    AI infrastructureBetaAgent context and memory

    aiduMEI

    Optional Jev classifies new memory types and verifies retrieval evidence in this memory system; author's self-test shows accuracy rising from 91.67% to 95.83%. (author)

    20 stars · Python · Open-source tool

  12. 0.9 smeasured

    Developer toolsIn productionAgent context and memory

    dsh-jev-tools

    DeepSeek Harness plugin where Jev rates segments of long tool output for task relevance and keeps the relevant ones, screens fetched pages for prompt injection and gates completion claims. Capped at 3 calls, the worst turn overhead was 0.9 s (author).

    14 stars · TypeScript · Open-source tool

  13. 9,553 charactersmeasured

    Developer toolsIn productionAgent context and memory

    pi-jev

    pi coding-agent extension in which Jev scores which tool calls and results to keep during compaction and rates each request's difficulty to pick a cheap or strong model per turn. One example run cut a span from 39,379 to 9,553 characters (author).

    10 stars · TypeScript · Open-source tool

  14. -31.3%vs a baseline

    Developer toolsIn productionAgent context and memory

    JIT-JEV Context OS

    Coding-agent context runtime where a Jev gate shadow-evaluates context facts, proposes evictions and prefetches memory, failing open to heuristics. On fixture tasks, Jev runs showed -31.3% turns per task vs a full-context baseline, with fewer solved (author).

    9 stars · Python · Open-source tool

  15. 79% smallervs a baseline

    Developer toolsIn productionAgent context and memory

    jev-skill-gate

    Scores every installed Claude Code skill against the current project in one parallel Jev pass and writes skillOverrides so only relevant skill descriptions load; the skill manifest came out 79% smaller on average (author).

    8 stars · JavaScript · Open-source tool

  16. 1,350 tokens per turnvs a baseline

    Developer toolsAnnouncedAgent context and memory

    jev-skills

    Vendor claims Jev decides which installed skills a Claude Code or Codex model sees each turn, cutting a measured 1,350-token-per-turn baseline to zero (author).

    6 stars · TypeScript · Open-source tool

  17. up to 85%measured

    Developer toolsIn productionAgent context and memory

    dsh-tokenslash

    Jev triages and prunes unused tool schemas in DeepSeek Harness, reducing prompt overhead by up to 85% (vendor).

    5 stars · JavaScript · Open-source tool

  18. 30 to 70 percentvs a baseline

    Developer toolsBetaAgent context and memory

    jev-codex-pilot

    jev-codex-pilot adds Jev-based routing and context optimization to Codex sessions, with the author claiming a 30 to 70 percent token reduction depending on session length (Charlyhno-eng).

    5 stars · TypeScript · Open-source tool

  19. 486vs a baseline

    AI infrastructureIn productionAgent context and memory

    jevtrim

    Benchmarks Jev as a context-compaction judge against retrieval and summarization on LoCoMo, across 486 questions over 10 conversations (jevtrim).

    5 stars · Jupyter Notebook · Open-source tool

  20. 14vs a baseline

    AI infrastructureBetaAgent context and memory

    Mnemon

    Mnemon long-term memory agent uses Jev for fast yes/no judgments on retrieved records, benchmarked for accuracy per token cost against 14 other systems (Grivn).

    4 stars · TypeScript · Open-source tool

  21. -38%vs a baseline

    Developer toolsIn productionAgent context and memory

    JevScout (yctimlin)

    Jev gates MCP tool output and codex commands in a coding agent pipeline, cutting agent time by 38% on Claude Code versus baseline. (yctimlin)

    3 stars · TypeScript · Open-source tool

  22. 19measured

    Developer toolsBetaAgent context and memory

    JevTrace

    Jev judges code relevance for an MCP server that trims coding-agent context to a token budget, based on the vendor's own 19-task retrieval benchmark (author).

    3 stars · TypeScript · Open-source tool

  23. 72.6%vs a baseline

    AI infrastructureBetaAgent context and memory

    PerfectRecall

    Hermes memory provider that asks Jev to judge every eligible memory against short criteria from the calling agent and returns the matching evidence, with no embeddings. On 120 LongMemEval-S questions final-answer errors fell 72.6% (author).

    3 stars · Python · Open-source tool

  24. 6/6 (100%)vs a baseline

    Developer toolsIn productionAgent context and memory

    ARMIN

    Rust sidecar for OpenCode and Claude Code; Jev selects verbatim decisions and open items from agent turns into a graph, injected as a brief each turn. A/B test: 6/6 (100%) cross-session consistency with memory vs 4/6 without (author).

    1 stars · Rust · Open-source tool

  25. 3.1% fewervs a baseline

    Developer toolsBetaAgent context and memory

    jev-usher (cvsgireesh)

    Claude Code launcher and hooks where Jev picks Haiku, Sonnet or Opus at session start and judges which tool output enters context, keeping originals for recovery. 3.1% fewer Claude tokens with filtering across 13 synthetic tasks (author).

    1 stars · TypeScript · Open-source tool

  26. R@1 0.41 to 0.62vs a baseline

    AI infrastructureIn productionAgent context and memory

    Hippo

    Local-first agent memory layer whose opt-in --reranker jev flag sends the query and candidate memories to Jev to rerank recall results. On a private 300-query developer store it moved R@1 0.41 to 0.62 against the local cross-encoder (author).

    Open-source tool

  27. as much as 75% or moremeasured

    Developer toolsBetaAgent context and memory

    Pi-Saver (amazingjoe)

    Pi Coder extension that groups conversation history into turns and asks Jev which ones the latest request needs, dropping the rest from the next call while saved history stays intact. As much as 75% or more context tokens saved (author).

    0 stars · TypeScript · Open-source tool

  28. 3.1 s to 1.9 svs a baseline

    Developer toolsIn productionAgent context and memory

    total-agent-memory

    MCP memory server for coding agents that can run the contradiction check inside memory_answer on Jev (MEMORY_CONTRADICTION_SCORER=jev) since v14.3.0. At the same accuracy the check dropped from 3.1 s to 1.9 s (author).

    Open-source tool

  29. no published number

    Developer toolsBetaAgent context and memory

    OpenAgents Coder

    Terminal coding agent whose Coder One loop sends one Jev request per step to pick which candidate files and output chunks reach the generator and to label the last command's outcome; Jev also shapes the brief handed to Claude Code or Codex.

    455 stars · Rust · Open-source tool

  30. no published number

    Developer toolsBetaAgent context and memory

    Cheshi

    macOS Codex workspace, in preview, can send a search question and candidate passages from past conversations to Jev to recall prior sessions with their sources, falling back to OpenAI Luna low; recall is off by default.

    18 stars · C · Open-source tool

  31. no published number

    Developer toolsBetaAgent context and memory

    fast-jev-compaction-opencode

    OpenCode V2 plugin asks Jev which completed tool calls and results are still needed, compacting agent context without a model-generated summary.

    13 stars · TypeScript · Open-source tool

  32. no published number

    Developer toolsIn productionAgent context and memory

    jevmory

    Local memory tool for Claude Code and MCP agents in which Jev grades each MEMORY.md line against stored session evidence for support and contradiction, labelling it stale, wrong, unsupported or keep, with each judgment stored as a receipt.

    11 stars · Python · Open-source tool

  33. no published number

    Developer toolsBetaAgent context and memory

    pi-pair-programmer

    A background multi-reviewer code review extension for Pi/OMP uses a Jev-based filter to strip duplicate and inherited findings from agent context.

    9 stars · TypeScript · Open-source tool

  34. no published number

    Developer toolsBetaAgent context and memory

    OpenCode Jev Compactor

    OpenCode plugin asks Jev narrow Noul, Choice and Score questions to decide keep, drop or truncate content during session checkpoint compaction.

    7 stars · TypeScript · Open-source tool

  35. no published number

    Developer toolsIn productionAgent context and memory

    optim-jev

    Claude Code plugin uses Jev for context compaction, model routing, and skill selection, reporting per-session dollar savings.

    6 stars · Python · Open-source tool

  36. no published number

    Developer toolsIn productionAgent context and memory

    LiteLLM

    Jev makes the fast context-compaction decisions inside LiteLLM's own gateway, used internally in production.

    Open-source tool

06Code and security review28

  1. 1-2measured

    SecurityIn productionCode and security review

    fingers (chainreactors)

    Jev verifies fingerprint rule matches in a web fingerprinting engine without naming products; vendor claims about 1-2 extra requests per hit page, cached (chainreactors).

    271 stars · Go · Open-source tool

  2. 92%vs a baseline

    Developer toolsIn productionCode and security review

    jgrep

    CLI that sends code chunks to Jev as yes/no questions and turns the probabilities into file:line hits for behavior search, English-rule diff linting and test selection; test selection caught 92% of touched tests, above name matching (author).

    59 stars · TypeScript · Open-source tool

  3. $0.050measured

    Developer toolsIn productionCode and security review

    JevGate

    Rust CLI code-review gate that parses a repository into functions, files and tests, asks Jev typed questions about each, and combines the answers in code into located findings; a run on a production TypeScript backend cost $0.050 (author).

    9 stars · Rust · Open-source tool

  4. 56/60measured

    SecurityBetaCode and security review

    Jevscan

    Jev scores a fact sheet for every on-chain transaction to flag hacks; vendor benchmark claims 56 of 60 known hacks caught (company).

    6 stars · Python · Commercial product

  5. ~100 msmeasured

    Developer toolsIn productionCode and security review

    eslint-plugin-jev (ShahriarBijoy)

    ESLint plugin whose rules are plain-English questions; each function goes to Jev, and a probability above the configured threshold (name contradicts body, unhelpful error message) becomes a lint warning. Judged in ~100 ms (author).

    3 stars · TypeScript · Open-source tool

  6. 34 of 34measured

    Developer toolsIn productionCode and security review

    five-lines

    Rust CLI that reviews a PR diff against the ten Five Lines of Code refactoring rules: tree-sitter counts the mechanical rules and Jev answers yes/no questions per changed method for the rest. Its second eval run scored 34 of 34 labelled judgments (author).

    2 stars · Rust · Open-source tool

  7. about one cent per repo scannedmeasured

    SecurityBetaCode and security review

    jev-attack-surface-analysis

    Jev ranks codebase files and lines by vulnerability likelihood under a fixed budget. Live demo claims about one cent per repo scanned (author).

    2 stars · Python · Open-source tool

  8. 19-20 of 20vs a baseline

    Developer toolsAnnouncedCode and security review

    JeVerifier

    Codebase tool uses Jev to flag rule violations and doc inconsistencies; author reports catching 19-20 of 20 planted violations with zero false flags (author).

    2 stars · Python · Open-source tool

  9. 460 of 617measured

    Developer toolsAnnouncedCode and security review

    prompt-fossils

    CLI audits CLAUDE.md and AGENTS.md files; Jev's judgment narrowed 617 pattern hits to 460 real findings across 101 of GitHub's 1,000 most-starred repos (author).

    2 stars · HTML · Open-source tool

  10. about $0.02measured

    Developer toolsIn productionCode and security review

    Clean Code Review

    Web page and MCP server that sends each file of a public GitHub pull request to Jev with up to 34 Clean Code questions in one call; a separate LLM writes the review from the probabilities. A 24-file review measured about $0.02 (author).

    Open-source tool

  11. 80ms p50measured

    SecurityBetaCode and security review

    jev-git

    Git pre-commit and pre-push hook uses Jev to flag secrets, prompt injection and destructive commands before a commit or push completes; author measures roughly 80ms p50 latency (author).

    Open-source tool

  12. about a cent per megabyte of sourcemeasured

    Developer toolsIn productionCode and security review

    Supercov (Supercorp)

    CLI where Jev answers yes/no property checks on each source file for code quality and security and Supercov computes the score, inside a coding-agent test and coverage loop. About a cent per megabyte of source scored (company).

    Open-source tool

  13. no published number

    Developer toolsBetaCode and security review

    Abide

    Codifies AGENTS.md and CLAUDE.md rules into a JSON rubric that Jev checks against, catching rule violations static linters miss in coding agents.

    578 stars · TypeScript · Open-source tool

  14. no published number

    Developer toolsAnnouncedCode and security review

    jev-code-reviewer

    Jev classifies each agent-generated diff into P0, P1 or P2 priority so reviewers see high-priority changes first; runs locally and posts nothing to GitHub.

    140 stars · JavaScript · Open-source tool

  15. no published number

    Developer toolsBetaCode and security review

    oxlint-plugin-jev

    Oxlint plugin that sends each matched function, call, JSX element or file to Jev with a plain-English yes/no rule, one request per file, and reports an error when the yes-probability meets the rule's cutoff.

    86 stars · TypeScript · Open-source tool

  16. no published number

    SecurityIn productionCode and security review

    is-malicious (luantak)

    CLI that sends a codebase's source, config, build and CI files to Jev and reports files and lines with probable data theft, hidden network activity or backdoors, exiting nonzero on high-severity findings.

    37 stars · TypeScript · Open-source tool

  17. no published number

    Developer toolsBetaCode and security review

    jev-spec

    An npm CI tool asks Jev one typed question per Markdown spec requirement, failing the build when code and probability drift below threshold.

    11 stars · TypeScript · Open-source tool

  18. no published number

    Developer toolsBetaCode and security review

    JevPR

    GitHub App sends normalized pull-request context from each webhook to Jev, which returns LOW, NORMAL or SPECIALIST; a YAML policy maps that decision to auto-approval, a review request or a check-run update.

    10 stars · Python · Open-source tool

  19. no published number

    Developer toolsIn productionCode and security review

    Jev Auto Approve (MetalBear)

    GitHub Action that asks Jev several yes/no questions about a pull request in one call and approves it only when every answer clears a confidence threshold, commenting the numbers otherwise; mirrord runs it on demand.

    9 stars · JavaScript · Open-source tool

  20. no published number

    Developer toolsAnnouncedCode and security review

    nouls

    Semantic linter sends Jev a batched yes/no question per parsed function to flag intent bugs, such as docstring drift or missing auth, that static tools miss.

    8 stars · Python · Open-source tool

  21. no published number

    Developer toolsBetaCode and security review

    Greenwash

    Jev answers yes/no questions on every PR diff hunk in a GitHub App to catch coding agents weakening tests to fake a green build.

    6 stars · Python · Open-source tool

  22. no published number

    Developer toolsBetaCode and security review

    JEV OAS Sentinel

    CLI and GitHub Action that diffs two OpenAPI documents, runs deterministic structural checks, then sends changed prose (retries, pagination, auth, errors) to Jev as bounded questions; code maps the probabilities to pass, review or block.

    5 stars · Python · Open-source tool

  23. no published number

    Developer toolsIn productionCode and security review

    Tranche

    Jev judges pull-request duplicates and groups fixes into reviewable batches for the Omarchy project's 2,200-PR backlog triage, with humans making the final merge call (author).

    3 stars · Rust · Open-source tool

  24. no published number

    Developer toolsBetaCode and security review

    nudgement

    Jev answers a fixed set of judgment questions on each commit message, code comment or doc line, then thresholds a nudge severity score.

    2 stars · TypeScript · Open-source tool

  25. no published number

    Developer toolsBetaCode and security review

    cargo-crap

    cargo-crap uses Jev to decide whether duplicate Rust functions share logic or shape and whether merging is worthwhile, caching answers and running opt-in.

    Open-source tool

  26. no published number

    Developer toolsBetaCode and security review

    Hapsland

    Jev reviews code edits in realtime for recurring errors agents miss, shipped as an early-feedback coding review tool.

    Commercial product

  27. no published number

    SecurityBetaCode and security review

    jev-scout

    CLI uses Jev System One scoring to vet open-source repo and crate recommendations against live registry metadata, to catch hallucinated packages.

    Open-source tool

  28. no published number

    Developer toolsBetaCode and security review

    JevLint

    npm CLI that sends selected source files and plain-English plugin rules, such as magic-strings and descriptive-names, to Jev and reports each file's probability of violating a rule, flagging it at a default 0.8 threshold.

    Open-source tool

07Evaluation and testing29

  1. p50 244msvs a baseline

    AI infrastructureIn productionEvaluation and testing

    jevals (Openlayer)

    Python library that sends all of a trace's agent evals and guardrails (tool choice, groundedness, scope, indirect injection, PHI) to Jev as typed questions in one request, offline or inside the agent loop. Openlayer measured p50 244ms per request (company).

    104 stars · Python · Open-source tool

  2. 51% fastervs a baseline

    Developer toolsIn productionEvaluation and testing

    grounded-ai (CascadeEvaluator)

    Routes agent-turn judgments to Jev before a larger model runs inside an eval harness, benchmarked 51% faster on 200 real turns (vendor).

    71 stars · Python · Open-source tool

  3. ~30x cheapervs a baseline

    Developer toolsIn productionEvaluation and testing

    Pinecone Cultivar

    Skill-testing CLI with an optional Jev grader (--backend typesafe) that judges each agent run against plain-English task criteria and returns a calibrated pass probability and failure kind. ~30x cheaper than Claude grading (company).

    42 stars · Python · Open-source tool

  4. $0.00098measured

    Developer toolsIn productionEvaluation and testing

    jev-test-filter

    CLI sends a git diff to Jev once as shared state with one question per test, then prints filter arguments for vitest, jest, Playwright, cargo test or go test so only affected tests run; one 110-test run cost $0.00098 (author).

    35 stars · TypeScript · Open-source tool

  5. 7,671vs a baseline

    AI infrastructureBetaEvaluation and testing

    Jev Arena

    A local dashboard benchmarks 13 decision-model profiles including Jev across 7,671 cases each, recording latency, accuracy and workflow-replay comparisons (author).

    33 stars · Python · Open-source tool

  6. 29 secondsvs a baseline

    Developer toolsBetaEvaluation and testing

    touchpress

    Playwright mobile e2e library whose act() loop asks Jev, via the AI SDK's experimental_evaluate, which on-screen move comes next or whether the step passes; a sign-in flow took 29 seconds on Jev, faster than Claude Haiku 4.5 (author).

    28 stars · TypeScript · Open-source tool · archived

  7. 78.43%vs a baseline

    FinanceBetaEvaluation and testing

    SmartMoney-Cub

    Read-only trading journal harness where Jev answers typed noul, choice and score judgments on a 240-case finance benchmark while Python handles arithmetic and dates. 78.43% overall accuracy, below the rule baseline (author).

    27 stars · Python · Open-source tool

  8. 86.7 %vs a baseline

    AI infrastructureBetaEvaluation and testing

    MetaCog

    Inference-time loop in which Jev scores each reasoning path a larger model writes and returns the highest-scoring answer, branching more paths when confidence is low. Lifted accuracy from 76.7 % to 86.7 % on GPQA Diamond and AIME rows (author).

    25 stars · Python · Open-source tool

  9. 79.8%vs a baseline

    Developer toolsPilotEvaluation and testing

    jevotron

    Scores agent trace steps with Jev to detect anomalies and harmful steps; a 24-trace pilot matched 130 of 163 published step-quality labels, 79.8% agreement (author).

    12 stars · Python · Open-source tool

  10. 80.5%measured

    Developer toolsIn productionEvaluation and testing

    Decision Model Benchmark (DMB)

    An independent reproducible benchmark measured Jev at 80.5% accuracy on banking intents (nibzard).

    11 stars · HTML · Open-source tool

  11. about 35 seconds and one centmeasured

    Developer toolsIn productionEvaluation and testing

    c9r browser-check

    Jev judges plain-English browser QA checklist steps as pass/fail inside the c9r browser-check plugin. A 24-step flow takes about 35 seconds and one cent (author).

    7 stars · TypeScript · Open-source tool

  12. a centmeasured

    Developer toolsAnnouncedEvaluation and testing

    jev-browse (cooper667)

    Claude Code plugin uses Cloudflare-hosted Jev to judge browser QA steps. A run costs about a cent (author).

    7 stars · TypeScript · Open-source tool

  13. 110× cheapervs a baseline

    Developer toolsIn productionEvaluation and testing

    pytest-jev

    A pytest fixture whose jev.expect sends every holds or lacks claim about an LLM app's reply to Jev in one request and fails the test on the returned probabilities. On 12 example tests it matched Claude Sonnet 5's verdicts and ran 110× cheaper (author).

    7 stars · Python · Open-source tool

  14. 81%measured

    Education and researchAnnouncedEvaluation and testing

    jevbench (Gxutxm)

    An independent preregistered study on 1,500 ChaosNLI pairs finds Jev calibrated on consensus items but overconfident on contested ones, at 81% agreement (author).

    1 stars · Python · Write-up

  15. 7,500vs a baseline

    Education and researchAnnouncedEvaluation and testing

    DoubtBench

    Benchmarks Jev and rival models on 7,500 questions scored against human annotator disagreement from NVIDIA HelpSteer2, rather than one correct answer (trust123500).

    Write-up

  16. 130 millisecond responses on averagevs a baseline

    SecurityPilotEvaluation and testing

    Dreadnode ScopeJudge

    Dreadnode ran Jev as the pre-execution judge on its ScopeJudge benchmark, deciding whether each pentest-agent tool call violates the user's scope. Jev was competitive with leading LLM judges at 130 millisecond responses on average (company).

    Write-up

  17. 272×vs a baseline

    Developer toolsBetaEvaluation and testing

    jeval.dev

    Hosted evaluation API that answers judgment-based autoevals, RAGAS, UiPath and promptfoo evaluators as Jev questions in one request per item, returning a score, distribution and pass/fail each. 272× cheaper than an LLM judge on its bundled benchmark (company).

    Commercial product

  18. $0.00035/callvs a baseline

    AI infrastructureIn productionEvaluation and testing

    LangChain

    LangChain tested Jev as an agent-eval judge against GPT-5.6 and Claude Sonnet 4.6 judges on identical traces, then added Jev-as-a-judge evaluators to LangSmith Evals. Jev cost $0.00035/call (company).

    Write-up

  19. 97%vs a baseline

    AI infrastructureIn productionEvaluation and testing

    LangWatch Instant Evals

    An eval() function in LangWatchQL, the CLI and the REST API asks Jev a question over every stored conversation, trace or LLM call and returns a calibrated verdict per row. 97% agreement with human labels on a support-chat benchmark (company).

    Commercial product

  20. $0.08measured

    Developer toolsPilotEvaluation and testing

    Vercel eve agent triage

    Jev classified full chat sessions from a custom Vercel eve agent by failure reason, reading each whole session log and returning one typed failure label. The batch cost $0.08 (author).

    In-house system

  21. no published number

    Education and researchIn productionEvaluation and testing

    Math-To-Manim

    Reviews every checkpoint of GPT-6-authored math animation scripts via a separate Jev API call, advisory by default with an optional gated mode (Math-To-Manim).

    2,759 stars · Python · Open-source tool

  22. no published number

    Games and entertainmentAnnouncedEvaluation and testing

    IntelliConnect

    Runs structured evaluation decisions inside an IoT agent platform for AI companion hardware, paired with local multilingual emotion recognition.

    147 stars · Java · Open-source tool

  23. no published number

    Developer toolsIn productionEvaluation and testing

    Typed Evals

    Python toolkit that sends RAG, response and agent-trace metrics to Jev in one judge request and checks proposed tool calls against policy before they run, with optional threshold calibration against human labels.

    16 stars · Python · Open-source tool

  24. no published number

    Developer toolsAnnouncedEvaluation and testing

    Jev Benchmark (ywchiu)

    Benchmarks Jev against open-source decision models Clef, Cygnet and Gemma in a simulated enterprise RAG agent routing scenario (ywchiu).

    6 stars · Python · Write-up

  25. no published number

    Developer toolsBetaEvaluation and testing

    playwright-jev (AdriaanVE)

    Jev powers failure triage, automatic retries, locator healing and semantic assertions as Playwright reporter and test plugins.

    2 stars · TypeScript · Open-source tool

  26. no published number

    AI infrastructureIn productionEvaluation and testing

    Arize AX

    Runs as a remote LLM output evaluator inside the Arize AX observability platform, with a runnable evaluation cookbook using Jev (Arize).

    Commercial product

  27. no published number

    Developer toolsBetaEvaluation and testing

    Convoy

    E2E test runner that dumps the iOS, Android or web accessibility tree and asks Jev whether the control a plain-language step names is on screen and which one matches, then a gate taps, types or reports it not found.

    Open-source tool

  28. no published number

    Developer toolsIn productionEvaluation and testing

    LangSmith Evals

    Jev serves as judge in LangSmith evals, scoring trajectory difficulty and correctness as separate typed answers in a single pass per trace.

    Commercial product

  29. no published number

    Developer toolsIn productionEvaluation and testing

    TraceRoot

    Judges detector outputs inside this open-source agent tracing platform using Jev, with users powering detection through their own TypeSafe key (TraceRoot).

    Open-source tool

08Search, ranking and matching50

  1. $0.01–0.03measured

    Developer toolsIn productionSearch, ranking and matching

    jegrep

    Semantic grep CLI where Jev returns a calibrated yes/no probability per file for a natural-language query, then reports matching files and line ranges with no index. $0.01–0.03 per search over a few thousand files (author).

    115 stars · Rust · Open-source tool

  2. ~2k free uses/monthmeasured

    Developer toolsIn productionSearch, ranking and matching

    pi-web-search

    Jev classifier ranks and filters web and code search results and screens for prompt injection inside a production Pi harness extension, with about 2,000 free uses per month (pi-web-search).

    20 stars · TypeScript · Open-source tool

  3. 81.8%vs a baseline

    ProductivityIn productionSearch, ranking and matching

    Jev Graph Search

    Jev reranks shortlisted passages in Obsidian/Logseq knowledge graphs; top-five recall rose from 63.9% to 81.8% on an 8,851-node graph (vendor).

    12 stars · JavaScript · Open-source tool

  4. 0.444vs a baseline

    AI infrastructureBetaSearch, ranking and matching

    Jev RAG

    Vector-free BM25 RAG pipeline where Jev reranks the top candidates; author's BEIR NFCorpus benchmark shows nDCG@10 rising to 0.444 in the hybrid setup (author).

    10 stars · Python · Open-source tool

  5. 13–44% fewer tokensvs a baseline

    Developer toolsBetaSearch, ranking and matching

    oko (bartlomein)

    Local code search served over MCP to Codex, Claude Code and OpenCode; Jev ranks the selected source snippets the agent reads. Across 162 paired sessions, agents used 13–44% fewer tokens with it (author).

    9 stars · Rust · Open-source tool

  6. 83%vs a baseline

    Developer toolsIn productionSearch, ranking and matching

    jevsearch

    shadcn command-palette block that streams keyword hits, then sends the top 20 to Jev in one request and reranks them by intent, keeping keyword order if Jev is slow. Hit@1 was 83% vs 41% for keyword ranking alone on 41 labelled queries (author).

    8 stars · TypeScript · Open-source tool

  7. 1500measured

    Developer toolsIn productionSearch, ranking and matching

    GStars

    Reranks GitHub Stars keyword search results by relevance with Jev, syncing about 1500 stars in 35 seconds (GStars).

    6 stars · TypeScript · Open-source tool

  8. $0.0049vs a baseline

    Developer toolsIn productionSearch, ranking and matching

    askgrep

    Rust CLI that sends each code chunk to Jev with one plain-language question, one chunk per request, and returns the functions ranked by probability. A sweep of the 819 functions in its own repo costs $0.0049 (author).

    4 stars · Rust · Open-source tool

  9. 7%vs a baseline

    Developer toolsIn productionSearch, ranking and matching

    genigrep

    Jev ranks candidate files and passages for code search, lowering cost by 7% at equal quality across 26 questions (company).

    3 stars · TypeScript · Open-source tool

  10. 59.3%vs a baseline

    ProductivityBetaSearch, ranking and matching

    Jev 求职助手 (jev-job-helper)

    Jev semantically filters and ranks BOSS Zhipin job listings for relevance; author's controlled 300-job test cut processing time 59.3%, from 42:58 to 17:28. (author)

    2 stars · TypeScript · Open-source tool

  11. 0.7smeasured

    Media and socialIn productionSearch, ranking and matching

    AATF AI News Daily

    In a multi-agent daily AI-news pipeline, Jev answers a relevance Choice for each gathered article under a frozen policy, and that choice alone decides which articles stay for the writing model. 69 articles sorted in 0.7s (author).

    0 stars · Python · In-house system

  12. 380 entries across 311 reposmeasured

    Developer toolsIn productionSearch, ranking and matching

    Awesome Jev (valentynkit)

    Directory site generated from a README, where Jev reorders project shortlists and each card shows build-time Jev probabilities; nothing on the site is gated on a Jev answer. Lists 380 entries across 311 repos, refreshed daily (author).

    Open-source tool

  13. 100measured

    Media and socialIn productionSearch, ranking and matching

    DigestFrog

    Scores each scraped news item 0 to 10 with Jev across 100 sources checked every 2 hours, before an LLM summarizes only top-ranked items (DigestFrog).

    Commercial product

  14. about 420msmeasured

    E-commerce and retailIn productionSearch, ranking and matching

    Everyframe Try-on

    In a virtual try-on app, Jev picks the pieces that complete an outfit, and the app renders the result on the user's photo, at about 420ms per styling decision (author).

    Commercial product

  15. 67%vs a baseline

    Developer toolsAnnouncedSearch, ranking and matching

    jev-digest

    Jev extracts the verbatim passages that answer a query for this MCP tool, cutting input tokens by 67% and tool rounds in half (jev-digest).

    0 stars · Python · Open-source tool

  16. 3,717measured

    Developer toolsIn productionSearch, ranking and matching

    matchcn

    MCP server tags shadcn-registry components on six dimensions through classifier.dev, a service running Jev, then ranks them against a plain-language brief and has Jev resolve close calls; the index covers 3,717 components (author).

    Open-source tool

  17. 10xvs a baseline

    General-purposeIn productionSearch, ranking and matching

    Metaview

    Every agent in a candidate sourcing product uses Jev to judge search matches; vendor reports same accuracy at about 10x the speed (Metaview).

    Commercial product

  18. 1,500measured

    Developer toolsIn productionSearch, ranking and matching

    Risetive

    Matches natural-language search queries against listings in a directory of about 1,500 Jev projects, using Jev for its own project search (Risetive).

    Commercial product

  19. 20xvs a baseline

    ProductivityAnnouncedSearch, ranking and matching

    Rox

    Rox uses Jev to retrieve sales context, with the vendor claiming retrieval 20x faster than GPT-5 Mini (Rox).

    Commercial product

  20. AUC 0.76 vs 0.65vs a baseline

    Media and socialIn productionSearch, ranking and matching

    Taste Buds

    Jev scores films and shows against user taste profiles with Choice and Noul to power recommendations; author reports AUC 0.76 versus a 0.65 baseline across 1.29M calls. (author)

    Commercial product

  21. no published number

    ProductivityIn productionSearch, ranking and matching

    Jevbox (Extend)

    Jev searches a permission-aware document hierarchy to ground chat answers in cited source documents.

    705 stars · TypeScript · Open-source tool

  22. no published number

    Education and researchBetaSearch, ranking and matching

    DeepDoc

    Local deep-research tool where Jev reranks Qdrant chunks and checks evidence coverage per report section before an LLM writes it; project has 305 GitHub stars (author).

    312 stars · Python · Open-source tool

  23. no published number

    SecurityIn productionSearch, ranking and matching

    SiftRank

    Semantic ranking CLI with a Jev provider: Jev compares pairs of items within randomized batches, and SiftRank combines the probabilities into a converged ranking of prose, code or tickets.

    225 stars · Go · Open-source tool

  24. no published number

    Developer toolsBetaSearch, ranking and matching

    sys1grep

    A CLI replaces regex grep with Jev-scored semantic line matching, combining meanings with AND/OR/NOT across languages; 145 GitHub stars (author).

    149 stars · JavaScript · Open-source tool

  25. no published number

    E-commerce and retailIn productionSearch, ranking and matching

    RentalCore

    Event-rental management system has Jev, called through OpenRouter, pick the matching catalog product, package or service for each OCR-extracted invoice line; below a confidence threshold or on timeout it falls back to the legacy matcher.

    69 stars · Go · Open-source tool

  26. no published number

    AI infrastructureIn productionSearch, ranking and matching

    Milvus Model (Zilliz)

    pymilvus's milvus-model library ships JevRerankFunction, which sends every candidate document to TypeSafe's API as a noul relevance question in one request and returns the top results sorted by probability.

    61 stars · Python · Open-source tool

  27. no published number

    Developer toolsAnnouncedSearch, ranking and matching

    Truffler

    This Rails search library asks Jev typed yes/no, choice and score questions at index and query time to rank and rerank records (Truffler).

    51 stars · Ruby · Open-source tool

  28. no published number

    AI infrastructureIn productionSearch, ranking and matching

    jev-reranker

    Python library scores each retrieved document with Jev for its usefulness as evidence, sorts them and drops those below a threshold before RAG context assembly, splitting long candidate lists and handling concurrency and retries.

    43 stars · Python · Open-source tool

  29. no published number

    SecurityAnnouncedSearch, ranking and matching

    Jevline

    Jev scores how related processes, accounts and hosts are to one confirmed-malicious indicator, expanding it into a full incident timeline for analysts.

    37 stars · TypeScript · Open-source tool

  30. no published number

    ProductivityBetaSearch, ranking and matching

    Jevfill

    Chrome extension that sends a page's form-field context and plain-text notes to Jev, which matches each field to the best note line above a confidence threshold before filling it; password and payment fields are never sent.

    20 stars · TypeScript · Open-source tool

  31. no published number

    SecurityIn productionSearch, ranking and matching

    NucleiSniper

    Scores Nuclei templates 0-4 by relevance to a fingerprinted target so high-value checks run first, across 13,764+ templates.

    18 stars · Python · Open-source tool

  32. no published number

    Media and socialIn productionSearch, ranking and matching

    Jev Highlight Cutter

    AI agent skill where Jev scores transcript evidence to select non-overlapping highlight clips from long interviews for automated video cutting.

    14 stars · Python · Open-source tool

  33. no published number

    ProductivityIn productionSearch, ranking and matching

    jev-apply

    Selects which saved fact fills each Greenhouse, Ashby or Lever job-application form field using Jev, while an LLM writes only genuinely new text (jev-apply).

    13 stars · JavaScript · Open-source tool

  34. no published number

    ProductivityIn productionSearch, ranking and matching

    BOSS Auto Job Helper

    BOSS Zhipin userscript where Jev scores each job posting for fit and research depth against the user's profile and vetoes labor-dispatch, paid-training and disguised-sales listings; top-tier matches get an automatic greeting, unclear ones go to review.

    10 stars · JavaScript · Open-source tool

  35. no published number

    AI infrastructureAnnouncedSearch, ranking and matching

    llama-index-jev

    Drop-in LlamaIndex reranker and router scores or chooses via Jev's typed Score and Choice outputs, positioned as cheaper than LLM-as-judge rerankers.

    8 stars · Python · Open-source tool

  36. no published number

    Developer toolsIn productionSearch, ranking and matching

    jev-search-mcp

    Jev decides which sources, time window and keywords to search, then scores every result for relevance, inside this MCP search plugin for Claude Code and Codex (jev-search-mcp).

    7 stars · JavaScript · Open-source tool

  37. no published number

    Developer toolsBetaSearch, ranking and matching

    semble-jev

    A CLI retrieves code locally via Semble, then Jev scores each snippet's relevance before returning selected source to a coding agent.

    6 stars · Python · Open-source tool

  38. no published number

    Developer toolsAnnouncedSearch, ranking and matching

    Hunch

    Sends each in-scope code chunk to Jev to answer plain-English review questions and returns scored source locations for a coding agent to investigate.

    4 stars · TypeScript · Open-source tool

  39. no published number

    Developer toolsIn productionSearch, ranking and matching

    jev-assist

    Coding-agent skill and CLI that takes piped candidates such as grep hits or file lists and asks Jev a noul, choice or score question per line, returning them sorted with a cut line so the agent reads only the rows above it.

    4 stars · JavaScript · Open-source tool

  40. no published number

    ProductivityBetaSearch, ranking and matching

    Semantic Find userscript

    Semantic Find userscript uses Jev for natural-language Ctrl+F on webpages, locating the original matching text without generating a summary; the author calls it an unverified trial version (dingdinglz).

    2 stars · TypeScript · Open-source tool

  41. no published number

    ProductivityIn productionSearch, ranking and matching

    AI Search: GuardianBrowser

    WunderCorp's AI search browser for iPhone, iPad and Mac, with ad and tracker blocking, Agent Assist and AI Text Signals. Neither the App Store listing nor guardianbrowser.sh names Jev, so what Jev decides in the app is not stated.

    Commercial product

  42. no published number

    Media and socialIn productionSearch, ranking and matching

    av (Pixel ML)

    Video-intelligence CLI whose av ask command uses Jev to filter retrieved sources for relevance, rank merged scenes by relevance probability times retrieval score, and check with a Noul whether the answer is supported.

    Open-source tool

  43. no published number

    ProductivityBetaSearch, ranking and matching

    CriteriaCue

    Chrome extension where Jev's Choice question rates LinkedIn evidence against recruiter-defined role criteria; no live accuracy demo shown.

    Open-source tool

  44. no published number

    E-commerce and retailIn productionSearch, ranking and matching

    Crumb

    Chrome extension that sends the search query and each listing's text to TypeSafe for relevance classification on Facebook Marketplace and Amazon, then hides unrelated listings and sponsored tiles.

    Commercial product

  45. no published number

    E-commerce and retailBetaSearch, ranking and matching

    Fit Receipt (NUDE)

    Lingerie fitting room where Jev judges style and set trade-offs against the shopper's confirmed needs and suggests which to compare first, while code checks product facts and prices; the page labels the results prototype judgments.

    Commercial product

  46. no published number

    Developer toolsIn productionSearch, ranking and matching

    Jev-bot (venu)

    Web tool that turns a plain-language idea into a Jev integration design: Jev evaluates a library of published Jev implementations and research for fitting patterns, and GLM 5.3 drafts the workflow and JSON.

    Commercial product

  47. no published number

    ProductivityIn productionSearch, ranking and matching

    Sez Where

    Sez Where uses Jev to point to the relevant passages in a document that answer a stated question, without generating an AI summary.

    Commercial product

  48. no published number

    Developer toolsIn productionSearch, ranking and matching

    Starguide

    Serverpod docs chatbot asks Jev to pick answering pages from the documentation table of contents as choice options, gate whether a vector search is still needed, and judge whether each answer resolved the question.

    Open-source tool

  49. no published number

    E-commerce and retailAnnouncedSearch, ranking and matching

    What to Order

    Matches restaurant menu dishes to a diner's stated mood using Jev; launched on Product Hunt with no metrics published.

    Commercial product

  50. no published number

    ProductivityBetaSearch, ranking and matching

    Yank

    Offers Jev-based semantic re-ranking as an opt-in, bring-your-own-key upgrade over the default on-device search in a clipboard manager.

    Commercial product

09Moderation and content filtering30

  1. 6 of 21measured

    SecurityIn productionModeration and content filtering

    JevPromptShield

    Scores incoming text with Jev to flag prompt injection before an agent acts; a test on 21 published Gandalf jailbreak prompts blocked 6 of 21 (author).

    202 stars · TypeScript · Open-source tool

  2. from 37% to 5%vs a baseline

    SecurityBetaModeration and content filtering

    jev-edge

    OpenResty, APISIX and edge-worker module in which Jev scores each incoming natural-language request for prompt injection or abuse and blocks it above a threshold, with caching and fail-open. A deployment context cut misses from 37% to 5% (author).

    41 stars · Lua · Open-source tool

  3. 19measured

    Media and socialIn productionModeration and content filtering

    slop-filter

    Scores every X, LinkedIn and Reddit post with 19 Jev questions to hide likely AI-generated content, with weights tunable to user labels (slop-filter).

    23 stars · JavaScript · Open-source tool

  4. About $0.00003 per postmeasured

    Media and socialIn productionModeration and content filtering

    Sift

    Chrome extension that labels every X post and reply via live Jev calls as Substance, Humor, Chit-chat, Promo or Junk, plus AI-written and Off-topic flags, and hides the labels the user picks. About $0.00003 per post (author).

    12 stars · JavaScript · Open-source tool

  5. 180msmeasured

    SecurityPilotModeration and content filtering

    ElevenLabs

    ElevenLabs pairs its call transcription with Jev risk scoring for realtime scam-call detection, taking about 180ms per request at $0.00043 for 11 calls total (company).

    Commercial product

  6. about a secondmeasured

    E-commerce and retailIn productionModeration and content filtering

    FakeCatch

    Chrome extension that sends each shopping review's text through its server to Jev for a trust score and badges reviews on Amazon, Flipkart, Trustpilot and other stores; scores are cached by text hash. Answers in about a second (company).

    Commercial product

  7. 0.987measured

    SecurityBetaModeration and content filtering

    kassad

    .NET guardrail middleware batches every policy for a stage into one Jev request and turns the typed answers into allow, flag, review or block verdicts on prompts, completions and tool calls; prompt-injection ROC AUC 0.987 on deepset (author).

    0 stars · C# · Open-source tool

  8. 96 per centvs a baseline

    Media and socialPilotModeration and content filtering

    Near Here

    Jev decides from each local-event listing's title and description whether to reject it as unsuitable after deterministic checks, in a published test against Mistral Small 4 and Gemini 3.5 Flash-Lite. 96 per cent accuracy (author).

    Write-up

  9. no published number

    Media and socialIn productionModeration and content filtering

    ClubDoorman

    Telegram anti-spam bot uses Jev to screen text messages across chats in the background without blocking delivery; project has 69 GitHub stars (author).

    69 stars · C# · Open-source tool

  10. no published number

    Media and socialIn productionModeration and content filtering

    Sharp

    Chrome and Firefox extension that scores each X post against user-written filter rules with a decision model by default (Jev via TypeSafe, OpenRouter or Vercel AI Gateway) and hides posts or holds close calls behind a banner before display.

    30 stars · TypeScript · Open-source tool

  11. no published number

    Developer toolsIn productionModeration and content filtering

    Tomarigi

    Watches Claude Code and Codex sessions and sends each message to Jev to flag abusive language toward the AI with a live anger mark.

    17 stars · TypeScript · Open-source tool

  12. no published number

    Media and socialBetaModeration and content filtering

    Jev Anti-Spam Bot

    Telegram bot that sends each group message to Jev with twenty-one spam questions and deletes it, with linked recent messages from the same campaign, when any signal reaches 0.90; messages stay when Jev is unavailable.

    16 stars · TypeScript · Open-source tool

  13. no published number

    SecurityAnnouncedModeration and content filtering

    JEV Safety Gateway

    Go reverse-proxy between nginx and LLM backends sends every user request to Jev to block harmful content and pass normal traffic through transparently.

    16 stars · Go · Open-source tool

  14. no published number

    Media and socialIn productionModeration and content filtering

    lkclean

    Chrome extension sends each LinkedIn post and the user's interests to Jev in one request of typed questions; local thresholds turn the probabilities into collapsing engagement bait, self-promotion and blocked topics, with scores shown on the post.

    14 stars · TypeScript · Open-source tool

  15. no published number

    Media and socialIn productionModeration and content filtering

    AnyFilter

    Chrome extension that classifies each X post with Jev against preset categories and rules the user writes in a sentence, hides matches, and lists hidden posts by reason in a side panel with a put-back control.

    11 stars · TypeScript · Open-source tool

  16. no published number

    E-commerce and retailAnnouncedModeration and content filtering

    jev-sap-commerce

    SAP Commerce extension uses Jev to moderate reviews and suggest categories and attributes, gated by configurable thresholds with a full audit record per decision.

    10 stars · Java · Open-source tool

  17. no published number

    Media and socialIn productionModeration and content filtering

    Twitter Jev Guard

    Chrome and Edge extension whose service worker sends visible X posts to Jev for a low-quality probability and a content type, then stamps a STOP watermark on posts above the threshold and an AD mark on promoted posts.

    9 stars · JavaScript · Open-source tool

  18. no published number

    Media and socialIn productionModeration and content filtering

    Bilibili Ad Skipper

    Jev classifies sponsored ad segments in a Chrome extension when no community sponsor tags exist for a video.

    6 stars · JavaScript · Open-source tool

  19. no published number

    SecurityIn productionModeration and content filtering

    Agents Mail

    Optional Jev labels inbound email as spam, phishing or marketing and flags likely prompt injection in this self-hosted agent mailbox.

    5 stars · TypeScript · Open-source tool

  20. no published number

    Media and socialAnnouncedModeration and content filtering

    Slop Scope

    Browser-run AI image detector sends measured image facts to a Jev model, the sole classifier labeling images likely_ai, likely_real or uncertain.

    3 stars · JavaScript · Open-source tool

  21. no published number

    Media and socialIn productionModeration and content filtering

    BiliBili Jev Filter

    BiliBili Jev Filter is a Quantumult X rewrite that uses Jev to filter Bilibili recommendation feed items by title, description, category and tags before they display.

    2 stars · JavaScript · Open-source tool

  22. no published number

    SecurityIn productionModeration and content filtering

    iMessage Spam Blocker

    Jev performs semantic classification of iMessage spam alongside Apple's native sender block list in a notarized macOS app.

    2 stars · Swift · Open-source tool

  23. no published number

    General-purposeBetaModeration and content filtering

    Frenzy

    Jev scores whether referral sign-ups are real humans or bots and ranks their influence, live in Frenzy's hosted beta.

    Commercial product

  24. no published number

    Media and socialIn productionModeration and content filtering

    musecases.app

    Prompt-sharing site that sends each newly published prompt's text to TypeSafe AI (Jev), which suggests a category tag and flags possibly harmful prompts for a moderator; the flags never hide a prompt on their own.

    In-house system

  25. no published number

    Media and socialBetaModeration and content filtering

    Slop Mop

    Chrome extension that asks Jev eleven questions about each LinkedIn post (nine slop signals, two counter-signals for human voice and usefulness), combines them into one slop score, and hides or outlines the post before it is read.

    Open-source tool

  26. no published number

    Media and socialBetaModeration and content filtering

    Soter

    Discord bot asks Jev, via OpenRouter, whether each message is hate speech or spam, deletes clear cases, flags borderline ones to moderators and escalates repeat violations from warnings to doubling timeouts.

    Open-source tool

  27. no published number

    Media and socialIn productionModeration and content filtering

    SVPN News

    Jev ranks news item quality and drives automatic moderation of reader discussions on this live news site.

    In-house system

  28. no published number

    Media and socialIn productionModeration and content filtering

    The Thing

    Social network where Jev assesses what each submitted thought adds to its conversation under that community's plain-language rules; only qualifying posts enter the queue for the next feed slot, and reactions do not reorder the feed.

    Commercial product

  29. no published number

    Media and socialIn productionModeration and content filtering

    txt.421.news

    Jev pre-approves only clearly clean posts on a live text-only forum before Claude or a human reviews the rest.

    In-house system

  30. no published number

    Media and socialIn productionModeration and content filtering

    Vettly

    Content-moderation API that scores text, image and video against programmable policies with parallel probabilistic signals, with Jev as one provider backend alongside OpenAI and Hive, and enforces the verdicts deterministically.

    Commercial product

10Document and record classification39

  1. 99.47%vs a baseline

    Education and researchIn productionDocument and record classification

    dailypaper-skills

    Jev scores and filters daily arXiv papers for relevance, with a claimed 99.47% cost savings and about 20x faster scoring than GPT-6 Astra. (huangkiki)

    1,286 stars · Python · Open-source tool

  2. 34× cheapervs a baseline

    FinanceIn productionDocument and record classification

    Kyotofin tax-doc-classifier

    Classifier where Jev reads each tax PDF page's text and returns a probability over 261 IRS forms and 7 page kinds, used across the company's tax document corpus. 34× cheaper per page than the prior Sonnet classifier (company).

    503 stars · TypeScript · Open-source tool

  3. 0.2smeasured

    Education and researchIn productionDocument and record classification

    LatteReview

    LatteReview v1.3.0 adds Jev as a reviewer that answers typed screening questions with probabilities to classify articles, taking about 0.2s per article (author).

    122 stars · Jupyter Notebook · Open-source tool

  4. 96.9%measured

    Education and researchIn productionDocument and record classification

    JEV Paper Radar

    Daily GitHub Action asks Jev one Noul per reader interest and exclusion plus a paper-type Choice for each new arXiv paper, and code builds must-read and maybe lists; screening mode recalled 96.9% of Cochrane reviewers' included studies (author).

    30 stars · Python · Open-source tool

  5. $0.0022vs a baseline

    Data and analyticsBetaDocument and record classification

    PDF Race

    Benchmark app in which Jev labels Docling-parsed arXiv papers by category and picks each title from candidate lines, raced against two Gemini 3.8 Flash lanes on the same papers. The Jev lane cost $0.0022 in total (author).

    13 stars · JavaScript · Open-source tool

  6. about 1 cent per daymeasured

    ProductivityIn productionDocument and record classification

    Laser (Omarchy)

    Judges each screen against a stated task with Jev to flag distraction in a Linux focus-mode app on the Omarchy marketplace; author claims about 1 cent per day (author).

    7 stars · Python · Open-source tool

  7. 73.7vs a baseline

    Education and researchAnnouncedDocument and record classification

    JevSpan

    Jev classifies entity spans in a zero-shot NER model that averages 73.7 strict F1 across 12 benchmarks (company).

    6 stars · Python · Open-source tool

  8. 5,857measured

    Education and researchBetaDocument and record classification

    Astrolabe

    Reads every accepted conference abstract and has Jev score topic relevance; author reports judging NeurIPS 2025's 5,857 papers (author).

    2 stars · TypeScript · Open-source tool

  9. 3,518 applications checked in 4:54measured

    ProductivityIn productionDocument and record classification

    Alan AI

    In-house review console at Alan AI in which Jev screens internship applications for evidence of real work and surfaces a shortlist for human review. Checked 3,518 applications in 4:54 (company).

    In-house system

  10. 6,000+ documents converted in the last 30 daysmeasured

    ProductivityIn productionDocument and record classification

    file2markdown.ai

    Opt-in AI check on PDF-to-Markdown output: excerpts go to Jev as yes/no questions (garbled, split sentences, broken table) and code thresholds the probabilities into issue codes. The site reports 6,000+ documents converted in the last 30 days (company).

    Commercial product

  11. 0.70measured

    Games and entertainmentIn productionDocument and record classification

    Football Agent

    Classifies fantasy-football news into injury, role_change, transaction, speculation, team_context or noise with Jev; only noise above 0.70 confidence is dropped (Football Agent).

    Commercial product

  12. 573 projectsmeasured

    Developer toolsIn productionDocument and record classification

    jevbooks

    Gallery site in which Jev judges whether each submitted repo is built on Jev and which design patterns it uses, with the probabilities published on each project page. Catalogs 573 projects (author).

    In-house system

  13. 99%vs a baseline

    Data and analyticsIn productionDocument and record classification

    Resurface

    Jev answers resume-screening questions per quoted resume line, reaching 99% agreement with Claude Sonnet 5.5 (author).

    0 stars · TypeScript · Open-source tool

  14. 4/4 on the arXiv fixturesmeasured

    ProductivityIn productionDocument and record classification

    watfile

    PyPI CLI that sends each file's extracted text to a Jev Choice question, batching up to ~100 files per call, then symlinks, copies or moves the file into the chosen category folder. Jev scored 4/4 on the arXiv fixtures (author).

    Open-source tool

  15. no published number

    Education and researchBetaDocument and record classification

    Jadense in Zotero

    A Zotero research plugin routes its literature-radar paper classification through Jev when cloud mode is enabled; 172 GitHub stars (author).

    175 stars · TypeScript · Open-source tool

  16. no published number

    Education and researchIn productionDocument and record classification

    Thesys

    Local research workspace whose Compare mode sends each retrieved passage pair from two papers to Jev, which labels it corroboratory, contradictory or neutral for a claim, with confidence listed in a side panel.

    168 stars · TypeScript · Open-source tool

  17. no published number

    Media and socialIn productionDocument and record classification

    Daily News

    Self-hosted AI news dashboard optionally uses Jev to classify intent, AI relevance, quality, importance and prompt-injection risk for aggregated news articles (Daily News).

    111 stars · Python · Open-source tool

  18. no published number

    FinanceIn productionDocument and record classification

    Formanator

    Formanator's --category-provider jev option sends a Forma claim's merchant, description and valid benefit and category pairs to Jev to pick the category, falling back to the configured LLM below 0.5 confidence. Receipt reading stays on a vision LLM.

    102 stars · Rust · Open-source tool

  19. no published number

    Media and socialAnnouncedDocument and record classification

    ReelQL

    Claude skill turns a video link into one typed JSON of scenes, brands and transcript, then has Jev judge and rank the extracted fields (ReelQL).

    30 stars · HTML · Open-source tool

  20. no published number

    ProductivityAnnouncedDocument and record classification

    MultiTool Office Next

    Jev offers optional assisted judgment to help OCR field extraction on contract documents inside a Windows office app.

    20 stars · TypeScript · Commercial product

  21. no published number

    E-commerce and retailIn productionDocument and record classification

    JEV 双休购

    Chrome extension in which Jev guesses from product, shop and brand text whether a Taobao or JD seller gives staff a two-day weekend, and stamps PASS on listings it judges do not.

    8 stars · JavaScript · Open-source tool

  22. no published number

    Media and socialIn productionDocument and record classification

    Aletheia

    Self-hosted AIGC-detection API and Chrome extension in which Jev labels an article page as original, repost or ad, returning per-category probabilities in a floating badge; Jev can also serve as the AI-text detector.

    6 stars · JavaScript · Open-source tool

  23. no published number

    Developer toolsIn productionDocument and record classification

    Cachix Jev Action

    Runs Jev judgments in CI via a GitHub Action, including choosing a pull request label from a repository's own label descriptions (Cachix).

    6 stars · TypeScript · Open-source tool

  24. no published number

    ProductivityBetaDocument and record classification

    Docket

    Docket is a Go terminal tool that asks Jev four typed questions per document to classify category, sensitivity, urgency and required action.

    6 stars · Go · Open-source tool

  25. no published number

    Data and analyticsIn productionDocument and record classification

    GTM Engineer Starter

    Scores prospect-account evidence packets against ICP rules with Jev as the calibrated judge in an open-source GTM starter kit.

    5 stars · Python · Open-source tool

  26. no published number

    Media and socialIn productionDocument and record classification

    CaminoASI

    Jev classifies and categorizes AI news articles before they are published on this Spanish-language AI tracking site, sorting content for display.

    In-house system

  27. no published number

    FinanceIn productionDocument and record classification

    Clearbook

    Live bank-statement tool uses Jev to categorize Nigerian transactions, identify beneficiaries, and reconcile VAT and bank fees.

    Commercial product

  28. no published number

    SecurityIn productionDocument and record classification

    Complee

    Jev classifies company descriptions as provider or deployer and assigns risk tier plus tool and data nouns in Complee's live EU AI Act compliance checker.

    Commercial product

  29. no published number

    FinanceIn productionDocument and record classification

    Expensent

    Receipt-forwarding service connects to business email inboxes, uses Jev to detect receipt and invoice emails in place of a cheaper non-reasoning LLM, and forwards them to an accountant or accounting software.

    Commercial product

  30. no published number

    ProductivityIn productionDocument and record classification

    grind.fyi

    Classifies live job postings scraped from Greenhouse, Lever, Workday and Ashby, flags listings it cannot label, and uses Jev to help auto-fill applications via a browser extension (grind.fyi).

    Commercial product

  31. no published number

    Education and researchIn productionDocument and record classification

    Jev Reviewer

    Browser tool for systematic-review extraction: Jev screens every stretch of each uploaded study for all RoB 2, ROBINS-I, QUADAS-2 or TIDieR questions, then checks each promising line, marking answers Reported, Unclear or Not found.

    Commercial product

  32. no published number

    ProductivityIn productionDocument and record classification

    JFocus

    JFocus, a paid macOS menu-bar app, sends the active app name and window title to Jev, which judges whether the current activity is on-task work, a brief detour or a rabbit hole, and alerts the user on sustained distraction.

    Commercial product

  33. no published number

    ProductivityIn productionDocument and record classification

    kaku

    Local-first Markdown editor offers optional Jev tagging to automatically organize AI-edited notes (kaku).

    Commercial product

  34. no published number

    ProductivityIn productionDocument and record classification

    Mochi

    Mochi, a free Mac file organizer, wires Jev as a direct model adapter alongside OpenAI, Gemini and Ollama to suggest file destinations.

    Commercial product

  35. no published number

    Education and researchIn productionDocument and record classification

    Paper Trellis Citation Verifier

    Jev scores whether each cited paper supports its citing sentence in a live peer-review tool, with a human making the final call.

    Commercial product

  36. no published number

    Developer toolsIn productionDocument and record classification

    pastewise

    Classifies unrecognized pasted text as stack trace, code, or prose to find the root cause, using Jev (pastewise).

    Commercial product

  37. no published number

    ProductivityBetaDocument and record classification

    Qualm

    Mac screen-time app asks Jev, or a local Kev model, to classify each screen as learning, work or endless-scroll entertainment before deciding whether to interrupt.

    Commercial product

  38. no published number

    Media and socialIn productionDocument and record classification

    Tattle

    Jev classifies live spoken claims as worth fact-checking and flags duplicate claims in this open-source Mac app.

    Open-source tool

  39. no published number

    ProductivityBetaDocument and record classification

    thusfar

    Jev links free-text entries in an invite-only goal journal to multiple user goals and suggests values, reviewed before saving.

    Commercial product

11SQL and data pipelines24

  1. $0.389vs a baseline

    Data and analyticsIn productionSQL and data pipelines

    JevSDSQL

    PostgreSQL-backed database with 41 semantic SQL operators built on Jev's Choice, Score and Noul for classification, extraction, ranking, matching and checks. Jev query planning cost $0.389 per 100 BIRD attempts vs $3.262 for an LLM, at lower accuracy (author).

    238 stars · Python · Open-source tool

  2. sub-cent cost per few thousand linesmeasured

    Developer toolsAnnouncedSQL and data pipelines

    grev

    Unix-style CLI filters pipe text through Jev for typed yes/no and classification decisions, at a claimed sub-cent cost per few thousand lines (author).

    56 stars · Go · Open-source tool

  3. 7.7xvs a baseline

    SecurityBetaSQL and data pipelines

    Jev IDS

    Jev classifies network flow records as attack or benign on the NSL-KDD dataset; vendor claims 7.7x faster than an LLM at similar precision (author).

    18 stars · Python · Open-source tool

  4. 1.3 secondsvs a baseline

    Data and analyticsAnnouncedSQL and data pipelines

    jev-harness (AntonioCoppe)

    Vendor claims Jev cut a row-filter job from 48.9 seconds on Claude Code CLI to 1.3 seconds using this confidence-gated harness with shadow mode (author).

    15 stars · TypeScript · Open-source tool

  5. 41% less moneyvs a baseline

    Data and analyticsIn productionSQL and data pipelines

    jev-ultralightspeed

    Python client that packs up to 32 items into each Jev request to answer one question per row across a dataset. Over 30,000 judgements against human labels it ran for 41% less money than one request per item, with no detectable accuracy difference (author).

    12 stars · Python · Open-source tool

  6. $2.95measured

    Education and researchAnnouncedSQL and data pipelines

    jlink

    Vendor claims Jev-based record linkage matched 171,354 firm-name pairs for economists at $2.95 total, about 250 pairs per second, from Python, Stata and R (author).

    6 stars · Python · Open-source tool

  7. 130/144 exact outcome decisionsmeasured

    Data and analyticsBetaSQL and data pipelines

    signal-weave

    MCP and webhook service in which Jev judges each watch item and question on an insight card against Superset BI evidence and picks the outcome, such as notify or investigate. Made 130/144 exact outcome decisions in a synthetic trial (author).

    4 stars · Python · Open-source tool

  8. mean F1 of 0.90vs a baseline

    IT operationsIn productionSQL and data pipelines

    jevgrep

    jevgrep asks Jev one yes/no question per piped log or text line and prints the lines whose yes probability clears a threshold, fast enough for tail -f. On its hand-labelled benchmark it scored a mean F1 of 0.90 against 0.87 for Claude Haiku 4.5 (author).

    2 stars · Python · Open-source tool

  9. $0.50vs a baseline

    Data and analyticsIn productionSQL and data pipelines

    MotherDuck

    MotherDuck's prompt_jev() SQL function runs Jev as a typed classifier over table rows, matching frontier-LLM accuracy while classifying 100k rows for $0.50, versus 32 minutes and $37 for an LLM (company).

    Commercial product

  10. 1,000 rows in about a secondmeasured

    Data and analyticsIn productionSQL and data pipelines

    Vibefilter

    Scores each admin-table row with Jev's calibrated probability against a plain-English filter, processing 1,000 rows in about a second (Vibefilter).

    Commercial product

  11. no published number

    Data and analyticsIn productionSQL and data pipelines

    pg-jev

    Postgres extension exposing Jev as SQL functions (jev, jev_prob, jev_choice, jev_score) for semantic filtering, scoring and ranking of rows without an embedding index.

    1,112 stars · Shell · Open-source tool

  12. no published number

    Data and analyticsAnnouncedSQL and data pipelines

    jev4pg

    Adds 41 Jev operators for filtering, extraction, ranking and verification to a semantic layer over PostgreSQL text-to-SQL queries (Sheltercosmo).

    238 stars · Python · Open-source tool

  13. no published number

    Data and analyticsBetaSQL and data pipelines

    duckdb-jev

    DuckDB extension whose jev_choice, jev_score, jev_noul and jev_ask SQL functions send each row's text to Jev with constant criteria and return a typed column: an ENUM, a DOUBLE score or probability, or a STRUCT.

    29 stars · C++ · Open-source tool

  14. no published number

    Data and analyticsIn productionSQL and data pipelines

    jevframe

    Python library adds a .jev accessor to pandas and Polars DataFrames that sends each row as state with typed questions to Jev and returns labels, scores and full probability distributions as columns.

    21 stars · Python · Open-source tool

  15. no published number

    Data and analyticsBetaSQL and data pipelines

    Jevaro

    Jevaro batches Jev requests concurrently and returns typed answers as Apache Arrow IPC streams for pandas, Polars and DuckDB pipelines.

    16 stars · Python · Open-source tool

  16. no published number

    Data and analyticsIn productionSQL and data pipelines

    jev-datafusion

    Registers Jev's noul, choice, score and ask as async SQL functions on an Apache DataFusion SessionContext, so a query answers typed questions for every row through TypeSafe, OpenRouter or a compatible endpoint.

    12 stars · Rust · Open-source tool

  17. no published number

    Developer toolsBetaSQL and data pipelines

    jeq

    CLI pipes ndjson records through Jev typed questions, then gates or routes them in shell scripts by the returned probabilities, exiting 0, 10 or 11 for pass, reject or uncertain.

    10 stars · Go · Open-source tool

  18. no published number

    SecurityIn productionSQL and data pipelines

    sensored

    An optional Jev-based semantic confirmation step reduces false positives in this streaming PII redaction library by confirming whether flagged spans are truly sensitive.

    10 stars · TypeScript · Open-source tool

  19. no published number

    Data and analyticsAnnouncedSQL and data pipelines

    datafusion-jev

    Rust crate adds a prompt_jev() SQL function to Apache DataFusion 55 so queries can pull typed Jev choices or scores directly from row text.

    7 stars · Rust · Open-source tool

  20. no published number

    Developer toolsAnnouncedSQL and data pipelines

    HekaJev

    CLI runs Jev over Git commit history to filter and classify each commit, saving the evidence, decision and cost behind every run (HekaJev).

    6 stars · Python · Open-source tool

  21. no published number

    Data and analyticsBetaSQL and data pipelines

    decision-query

    SQLite and PostgreSQL extension adding noul, choice and score SQL functions that run a local Laya checkpoint by default or call TypeSafe's Jev endpoint over HTTP, so one query can filter, classify or rank rows.

    2 stars · C++ · Open-source tool

  22. no published number

    E-commerce and retailIn productionSQL and data pipelines

    Competitor Price Monitor

    First published Apify actor using Jev, monitoring Shopify competitor prices and stock and surfacing changes through Apify datasets and schedules (Competitor Price Monitor).

    Commercial product

  23. no published number

    Data and analyticsAnnouncedSQL and data pipelines

    Databricks ai_decide()

    Jev runs as ai_decide(), a native SQL function answering one typed decision question per row over governed lakehouse data in Databricks.

    Commercial product

  24. no published number

    FinanceIn productionSQL and data pipelines

    Model vs Market

    Live site has Jev read news and price Polymarket and Kalshi questions alongside OpenAI Decisions and Cloudflare Clef, compared against real market prices.

    Write-up

12Writing and content scoring20

  1. ~$0.0008measured

    FinanceBetaWriting and content scoring

    Jev X Sentiment Analysis

    Crypto terminal where Jev returns a buy, sell or hold call, a sentiment score and squeeze risk from live price data and pre-processed tweet statistics; trades are placed manually. About $0.0008 Jev cost per search (author).

    216 stars · Python · Open-source tool

  2. $0.050measured

    Education and researchBetaWriting and content scoring

    Passage (Working Memory Jev)

    Experimental localhost teaching aid has Jev make bounded choices about familiarity, grouping and relationships at each reading step to estimate working-memory demand in instructional text; analyzing the sample passage cost $0.050 (author).

    76 stars · Python · Open-source tool

  3. about a tenth of a centmeasured

    ProductivityIn productionWriting and content scoring

    Osso

    A browser extension asks Jev per sentence whether text is what the reader came for; vendor claims a full article costs about a tenth of a cent (Osso).

    16 stars · TypeScript · Open-source tool

  4. $0.0004measured

    ProductivityBetaWriting and content scoring

    riff

    riff prose linter uses Jev for semantic rule checks, returning calibrated probabilities per finding; an example run costs about $0.0004 (Scale Venture Partners).

    7 stars · Python · Open-source tool

  5. 77%measured

    Media and socialIn productionWriting and content scoring

    Community Signals (Joshua Poddoku)

    Live page tracking Bluesky and Hacker News posts about AI-written code; Jev reads each post's tone and depth, with DeepSeek as fallback. On 238 hand-labelled posts Jev matched the author's tone reading 77% of the time (author).

    In-house system

  6. about half a centmeasured

    Media and socialBetaWriting and content scoring

    dotmap.ai

    Hosted SEO and answer-engine audit that crawls a site and has Jev score each page on seven rubrics and five yes/no checks in one call, beside code-run technical checks. A 50-page site costs about half a cent in inference (company).

    Commercial product

  7. $0.00028measured

    Media and socialPilotWriting and content scoring

    Maxfusion

    Maxfusion pulls an advertiser's live Meta video ads and breaks each one down shot by shot; Jev then judges every shot and rates the whole ad out of 100, scoring each shot in about 1.5 seconds (author).

    Write-up

  8. about four thousandths of a centmeasured

    Media and socialIn productionWriting and content scoring

    Postneedle

    Postneedle sends each X post or draft to Jev with five yes/no questions in one request (AI-written, humblebrag, engagement bait, platitude, buried point) and turns the probabilities into needle scores. A reading costs about four thousandths of a cent (author).

    Commercial product

  9. 50xvs a baseline

    ProductivityIn productionWriting and content scoring

    Proseward

    Proseward uses Jev to identify the speaker of unattributed dialogue tags in fiction, with the author reporting it runs 50x faster and cheaper than Sonnet at similar accuracy (Proseward).

    Commercial product

  10. no published number

    ProductivityIn productionWriting and content scoring

    slop-grader

    npm CLI asks Jev every grading rule against every line of a text file in parallel, caching unchanged lines, then hands the violations to an AI agent that separates false positives and drafts fixes.

    33 stars · TypeScript · Open-source tool

  11. no published number

    Media and socialBetaWriting and content scoring

    HookMeter

    Chrome extension that scores X/Twitter drafts as they are typed: Jev rates curiosity gap and emotional arousal, picks the viral-hook pattern and flags clickbait or throttling risk, through a FastAPI or Cloudflare Worker proxy.

    22 stars · HTML · Open-source tool

  12. no published number

    Media and socialBetaWriting and content scoring

    Transcript Lens

    Turkish-language Next.js app where Jev classifies YouTube transcript passages into chapters, claims, opinions and topic matches without rewriting text or checking factual accuracy.

    18 stars · TypeScript · Open-source tool

  13. no published number

    ProductivityIn productionWriting and content scoring

    Sales Copilot (moritzkremb)

    Live sales-call copilot asks Jev about 37 typed questions per utterance to update closing probability and coach next moves in real time (author).

    9 stars · Python · Open-source tool

  14. no published number

    ProductivityIn productionWriting and content scoring

    DoubaoTypeless

    Phone-to-PC text and image input tool, now named Pocket Composer, with an optional off-by-default Jev check that marks likely transcription errors, four reference dimensions and tone in the current draft before the user sends it.

    7 stars · Python · Open-source tool

  15. no published number

    ProductivityIn productionWriting and content scoring

    PageGrade

    Grades page sections on ten rubric scores with Jev via Vercel AI Gateway, producing an A-E grade for each page in a Chrome extension.

    7 stars · TypeScript · Open-source tool

  16. no published number

    ProductivityIn productionWriting and content scoring

    FreeResume.site

    Jev makes bounded yes/no judgments on resume text in the live app, staying silent when confidence is weak or signals conflict.

    Commercial product

  17. no published number

    Media and socialIn productionWriting and content scoring

    HN Judge

    Sends each comment on the most-discussed Hacker News threads to Jev for a stance, a substance score, a quotability value and an open-question flag, then assembles a per-thread verdict from those numbers in Neon Postgres.

    Open-source tool

  18. no published number

    Developer toolsIn productionWriting and content scoring

    jev-seo

    Rust SEO CLI and MCP server where Jev classifies keyword intent and scores pages for answer-engine citation, gated by confidence and a per-run spend cap.

    Open-source tool

  19. no published number

    Media and socialIn productionWriting and content scoring

    kite

    Live social-growth copilot uses Jev to pick which post ideas and drafts to publish, showing a head-to-head win score on each draft.

    Commercial product

  20. no published number

    ProductivityIn productionWriting and content scoring

    undertone

    Web text box that sends each draft message to Jev as eight typed tone questions in one call, covering passive-aggression, sarcasm, over-apologizing, urgency and manager-safety; code turns the answers into a verdict, color and phrase tips.

    Open-source tool

13Chat bots and games16

  1. 50-70msmeasured

    Media and socialIn productionChat bots and games

    astrbot_plugin_jev

    Jev runs a System One recheck before an AstrBot chat bot interjects into a group conversation, gating whether it responds, with per-group rate limiting. 50-70ms decision latency (author).

    3 stars · Python · Open-source tool

  2. 0.8smeasured

    Media and socialIn productionChat bots and games

    hikomi

    Jev fixes misheard names, decides whether a screenshot is needed within 0.8s, and recalls past chat in a live AI companion app. (hikomi)

    Commercial product

  3. 约 1 秒measured

    ProductivityIn productionChat bots and games

    Jev Chat Jarvis

    Android and desktop assistant reads WeChat, QQ and X chats through the accessibility service, has Jev judge intent and risk, then ranks three generated reply drafts by probability for the user to send by hand; about 1 second per result (author).

    Open-source tool

  4. up from 77.5% to 97.9%vs a baseline

    FinanceIn productionChat bots and games

    LendEasy.ai

    LendEasy.ai replaced LLM calls with Jev to classify user intent and conversation pivots and select pre-approved speech responses in its lending assistant. Accuracy rose from 77.5% to 97.9% (company).

    In-house system

  5. 3-23x fastervs a baseline

    Games and entertainmentIn productionChat bots and games

    Sausage Battlefield

    Jev picks the AI's move live in-game, compared against 9 LLMs, claimed 3-23x faster with comparable agreement (author).

    Commercial product

  6. no published number

    Media and socialBetaChat bots and games

    狗头军师 Chat

    Optional Jev strategy layer picks a reply strategy for the current WeChat chat before a DeepSeek or OpenAI-compatible model drafts candidate replies in this Mac sidebar assistant; the user copies or fills a draft and sends it.

    380 stars · Python · Open-source tool

  7. no published number

    Games and entertainmentBetaChat bots and games

    MineAI

    Minecraft Forge server mod whose NPC agent hands ambiguous next-action and task choices, plus blocked, progress and goal checks, to an optional TypeSafe Jev judgment layer beside its DeepSeek LLM, tracing each call on the HUD.

    69 stars · Java · Open-source tool

  8. no published number

    Media and socialAnnouncedChat bots and games

    Orbs

    In multi-bot chat rooms, Jev decides which bot should reply when no bot is explicitly mentioned by name.

    36 stars · TypeScript · Open-source tool

  9. no published number

    Games and entertainmentIn productionChat bots and games

    DesktopAssistant

    Desktop AI toolkit's 'game brain' feature defaults to OpenRouter plus Jev (typesafe/jev-1.13) as the decision model answering gameplay questions (DesktopAssistant).

    17 stars · Python · Open-source tool

  10. no published number

    Media and socialBetaChat bots and games

    Xiaoke (小可)

    Self-hosted QQ group chatbot in which Jev judges when to join a conversation and whether to reply by voice and in which style, alongside other semantic checks logged in a web dashboard.

    11 stars · Python · Open-source tool

  11. no published number

    Games and entertainmentIn productionChat bots and games

    Soupbase

    Bilingual turtle-soup puzzle site where Jev acts as host, judging each player question as yes, no or irrelevant and scoring submitted solutions; low-confidence answers show as unconfirmed. Runs through TypeSafe, OpenRouter or Vercel AI Gateway.

    9 stars · TypeScript · Open-source tool

  12. no published number

    Media and socialIn productionChat bots and games

    astrbot_plugin_oven_multi

    AstrBot group-chat plugin where Jev decides whether the bot should proactively reply after a run of messages and reads addressee, intent, emotion, attitude and risk when the bot is mentioned, injecting that reading into the LLM request.

    5 stars · Python · Open-source tool

  13. no published number

    Media and socialAnnouncedChat bots and games

    Jev Active Reply (AstrBot plugin)

    AstrBot plugin uses Jev to decide in real time whether a group chat message is worth an active reply before the main model generates one.

    2 stars · Python · Open-source tool

  14. no published number

    Games and entertainmentIn productionChat bots and games

    Abstractle

    Jev judges daily word-game answers via two parallel calls, a fixed blind lineup and a rubric score, live and free to play.

    Commercial product

  15. no published number

    Developer toolsIn productionChat bots and games

    Jauvex

    Jev powers the fast decisions behind Jauvex's voice channel for Claude and Codex coding sessions, shipped in the macOS v1.0 desktop app.

    Commercial product

  16. no published number

    Games and entertainmentIn productionChat bots and games

    THAT THAT GAME

    Jev classifies and rates clever player responses in a live daily word game.

    Commercial product

14SDKs and workflow integrations70

  1. 434 ms median warmmeasured

    AI infrastructureIn productionSDKs and workflow integrations

    JarvisCore (Prescott Data)

    Python multi-agent framework that injects a Jev client as self.decisions for Choice, Score and Noul judgments, with opt-in Jev subagent routing, model-tier selection and RAG passage filtering. Three questions in one call: 434 ms median warm (company).

    192 stars · Python · Open-source tool

  2. 40–150msmeasured

    Developer toolsIn productionSDKs and workflow integrations

    Jev for Apple Foundation Models

    Swift 6 bridge maps Apple Foundation Models @Generable Bool, enum and range types to Jev's noul, choice and score primitives and evaluates them against app state through the Jev API in 40–150ms (author).

    99 stars · Swift · Open-source tool

  3. 70–500msmeasured

    Developer toolsIn productionSDKs and workflow integrations

    Laravel Jev

    Laravel package exposes Jev::is, choose and score on Laravel's HTTP client, batching several questions per request, for webhook triage, submission filtering, inquiry routing and form validation; typical responses 70–500ms (author).

    8 stars · PHP · Open-source tool

  4. 11.1×vs a baseline

    Developer toolsIn productionSDKs and workflow integrations

    JevForce

    Dependency-free Apex SDK exposing Jev's Choice, Score and Noul as typed Salesforce methods, with case-triage and lead-qualification demos. One live run of seven judgments was 11.1× faster than Agentforce Models API with Opus 4.8 (author).

    7 stars · Apex · Open-source tool

  5. 35 to 500 millisecondsmeasured

    Developer toolsAnnouncedSDKs and workflow integrations

    openpave-jev

    PAVE/Claude skill wraps Jev's typed Choice, Score and Noul calls for CLI classification and routing at a claimed 35 to 500 millisecond latency (author).

    2 stars · JavaScript · Open-source tool

  6. 90measured

    Developer toolsAnnouncedSDKs and workflow integrations

    Semantic Validator

    Open-source SDKs for Python, TS, Go and Java use Jev for BYOK semantic validation of text fields; author published a 90-case evaluation of results (author).

    2 stars · Go · Open-source tool

  7. 87.7%measured

    AI infrastructureIn productionSDKs and workflow integrations

    classifier.dev

    Public zero-shot classification API: the fast tier returns Jev's calibrated label probabilities in one round trip, and the smart tier re-asks a reasoning model when Jev's confidence falls below 0.7. 87.7% on four-way news topics (company).

    Commercial product

  8. ~$0.00002/decisionmeasured

    AI infrastructureIn productionSDKs and workflow integrations

    Juspay NeuroLink

    TypeScript SDK adds decide, a Jev-backed inference type returning boolean, choice or score judgments, and calls it internally for model routing, context compaction, MCP tool routing and RAG planning. ~$0.00002 per decision (company).

    Open-source tool

  9. 70–500msmeasured

    AI infrastructureIn productionSDKs and workflow integrations

    Yao Agents

    Self-hosted multi-device agent platform that exposes Jev through its Tao gateway as a Structured Decisions model setting, so agents send typed choice, score and noul questions such as batch ticket classification. Responses in 70–500ms (company).

    Commercial product

  10. no published number

    AI infrastructureIn productionSDKs and workflow integrations

    AgentScope

    Agent framework ships JevClassifierModel, which maps AgentScope binary, choice and score questions to Jev's Noul, Choice and Score primitives through the TypeSafe SDK and returns typed answers with probabilities.

    33,076 stars · Python · Open-source tool

  11. no published number

    Developer toolsIn productionSDKs and workflow integrations

    LLPhant

    LLPhant's new JevClassifier adds typed Noul, Choice and Score classification to this PHP GenAI framework, which has 1700+ GitHub stars (LLPhant).

    1,717 stars · PHP · Open-source tool

  12. no published number

    AI infrastructureIn productionSDKs and workflow integrations

    Heym

    Heym's Decision node, added in v0.0.112, sends workflow state and typed noul, choice or score questions to a decision model such as Jev and returns probabilities the self-hosted workflow branches on for routing, triage or gating.

    1,167 stars · Python · Open-source tool

  13. no published number

    ProductivityIn productionSDKs and workflow integrations

    OpenClip

    Mac-wide clipboard utility sends selected text to Jev for classification against user-defined prompts, working across any application (OpenClip).

    658 stars · Swift · Open-source tool

  14. no published number

    AI infrastructureBetaSDKs and workflow integrations

    jev-align (Sutro)

    Sutro CLI that runs a Jev classifier over a dataset, surfaces uncertain rows for human labeling, and uses GEPA to rewrite the classifier definition from those labels.

    307 stars · Python · Open-source tool

  15. no published number

    Developer toolsAnnouncedSDKs and workflow integrations

    Pi Desktop

    Desktop GUI for the Pi and oh-my-pi coding agents ships an installable Jev skill for agent decisions once a TypeSafe API key is saved (Pi Desktop).

    300 stars · TypeScript · Open-source tool

  16. no published number

    Developer toolsIn productionSDKs and workflow integrations

    jevcore

    TypeScript packages that give DeepSeek Harness and any MCP host three Jev tools for typed yes/no, choice and ranking decisions, plus opt-in safety and context gates; the provider stays an offline mock until a live key is set.

    106 stars · TypeScript · Open-source tool

  17. no published number

    Developer toolsIn productionSDKs and workflow integrations

    sttp-ai

    Scala HTTP client toolkit ships a Jev module alongside its OpenAI, Claude and Gemini clients; project has 101 GitHub stars (author).

    103 stars · Scala · Open-source tool

  18. no published number

    Developer toolsBetaSDKs and workflow integrations

    AgentRun

    Workflow DSL whose judge nodes send focused per-step questions to Jev and branch on answer and confidence, calling a full agent only when investigation is needed and escalating for review otherwise; ships as npm packages and a Pi extension.

    94 stars · TypeScript · Open-source tool

  19. no published number

    AI infrastructureIn productionSDKs and workflow integrations

    jevcache

    Local cache that memoizes Jev decisions keyed on model, schema and redacted state, so repeat calls cost nothing and CI can replay fixtures to catch model drift; runs as a CLI or HTTP sidecar.

    75 stars · Open-source tool

  20. no published number

    Developer toolsBetaSDKs and workflow integrations

    jev-mcp

    Stdio MCP server that gives Cursor, Codex and other MCP clients nine Jev tools for step routing, tool-call selection, patch review, claim verification, screening and ranking, each returning auto, review or escalate.

    62 stars · TypeScript · Open-source tool

  21. no published number

    Developer toolsBetaSDKs and workflow integrations

    Jeview

    Local gateway between application code and the Jev API forwards each request with the user's key, stores every call in SQLite and draws calls on a live map, linking follow-up requests to the answers that triggered them.

    61 stars · JavaScript · Open-source tool

  22. no published number

    Developer toolsIn productionSDKs and workflow integrations

    Spring AI TypeSafe SDK

    Community Java client for Jev built on Spring RestClient, plus Spring AI integrations that use Jev as a model judge, guardrail advisor, RAG post-processor and tool index; published to Maven Central.

    52 stars · Java · Open-source tool

  23. no published number

    Developer toolsIn productionSDKs and workflow integrations

    vibecheck

    Wraps Jev decision calls as plain Python functions (check, classify, label, score) returning calibrated probabilities with no prompt templating (vibecheck).

    42 stars · Python · Open-source tool

  24. no published number

    Education and researchIn productionSDKs and workflow integrations

    US Legal MCP

    An optional Jev filter drops low-confidence matches in a live MCP server that searches Congress, Federal Register and court opinions.

    38 stars · TypeScript · Open-source tool

  25. no published number

    Developer toolsAnnouncedSDKs and workflow integrations

    genai (maruel)

    Go AI package genai adds Jev as a typed decision backend via Provider.SystemOne(), alongside Cloudflare CLEF and local llama.cpp models. (maruel)

    32 stars · Go · Open-source tool

  26. no published number

    Developer toolsBetaSDKs and workflow integrations

    llm-typesafe

    Plugin for Simon Willison's LLM CLI and Python library that adds a jev model returning typed yes/no probabilities, category choices and rubric scores as JSON instead of text, with templates and stdin support.

    27 stars · Python · Open-source tool

  27. no published number

    Developer toolsBetaSDKs and workflow integrations

    jev-code (rhighs)

    TypeScript SDK and coding-agent CLI where Jev picks one option from a typed router or decision tree and application code runs the effects.

    20 stars · TypeScript · Open-source tool

  28. no published number

    Developer toolsIn productionSDKs and workflow integrations

    jev-feels

    Ruby gem adds feels?, decide and score methods and a validates_feeling Rails validation that send a model field to Jev and return booleans, choices or scores for questions declared on the class.

    20 stars · Ruby · Open-source tool

  29. no published number

    Developer toolsBetaSDKs and workflow integrations

    AnyDecisionModel

    Swift package with typed yes/no, choice and score sessions, in which the JevDecisionModel backend calls TypeSafe's System One API alongside a local MLX backend on Apple silicon.

    15 stars · Swift · Open-source tool

  30. no published number

    Developer toolsAnnouncedSDKs and workflow integrations

    typesafe-ai-php

    PHP 8.2+ client SDK sends a state and typed questions to Jev and returns typed answers per question (typesafe-ai-php).

    15 stars · PHP · Open-source tool

  31. no published number

    Developer toolsBetaSDKs and workflow integrations

    Jev Workflows for Codex

    Codex plugin with 8 MCP tools uses Jev to consult on tool and model choice, diagnose failed commands, and check completion claims against evidence.

    12 stars · JavaScript · Open-source tool

  32. no published number

    Developer toolsAnnouncedSDKs and workflow integrations

    jev4k

    Kotlin DSL and client lets developers declare typed noul, choice and score questions to Jev and read back calibrated answers (jev4k).

    12 stars · Kotlin · Open-source tool

  33. no published number

    Developer toolsBetaSDKs and workflow integrations

    jevelry

    CLI, TUI and library that send typed questions from ready-made jevels (tickets, alerts, pull requests, failing tests) to Jev, map confidence to act, mark or fall back, and log every decision for review and threshold tuning; installs as an agent skill.

    11 stars · TypeScript · Open-source tool

  34. no published number

    Developer toolsBetaSDKs and workflow integrations

    jevflow

    TypeScript library that declares Jev noul, score and choice judgments as data and composes them into workflows whose explicit thresholds return an ordered, explainable action list, such as sending a payment to human review.

    11 stars · TypeScript · Open-source tool

  35. no published number

    Developer toolsIn productionSDKs and workflow integrations

    TypeSafe SDK for Java

    Community Java SDK with a Spring Boot starter that sends Choice, Score and Noul questions over application state to Jev and returns typed answers with calibrated probabilities in one round trip; published on Maven Central.

    11 stars · Java · Open-source tool

  36. no published number

    Data and analyticsAnnouncedSDKs and workflow integrations

    llmClassificR

    R package adds native Jev Choice, Noul and Score functions for zero-shot text classification, alongside Groq LLM and embedding baselines.

    10 stars · R · Open-source tool

  37. no published number

    Developer toolsIn productionSDKs and workflow integrations

    dfinke/Jev

    A PowerShell SDK wraps Jev's typed yes/no, choice and score questions so scripts can branch on structured decisions, such as paging on-call (author).

    8 stars · PowerShell · Open-source tool

  38. no published number

    Developer toolsIn productionSDKs and workflow integrations

    typesafe-sdk (joshmn)

    Ruby gem that sends Choice, Score and Noul questions to Jev's System One API and returns typed answers, with examples for support-ticket routing, frustration scoring and urgency checks; defaults to jev-latest.

    7 stars · Ruby · Open-source tool

  39. no published number

    Developer toolsIn productionSDKs and workflow integrations

    Vybez

    TypeScript library whose is, pick and rate tagged templates embed semantic rules in ordinary conditionals, answered by Jev by default as probabilities, choices or ratings; questions on the same state in one tick share one request.

    7 stars · TypeScript · Open-source tool

  40. no published number

    Developer toolsAnnouncedSDKs and workflow integrations

    jev-java (Olti1947)

    Idiomatic Java 17+ SDK for TypeSafe's Jev API with typed Choice and Noul primitives, structured instructions and probability responses.

    6 stars · Java · Open-source tool

  41. no published number

    Developer toolsIn productionSDKs and workflow integrations

    AI Provider for Jev

    WordPress plugin wires Jev's Choice, Score, and Noul primitives into WordPress as a structured-decision AI provider.

    5 stars · PHP · Open-source tool

  42. no published number

    Developer toolsIn productionSDKs and workflow integrations

    decisions-judge-mcp

    MCP server exposing a judge tool that sends a JSON state and noul, choice or score questions to Jev through the TypeSafe SDK or Cloudflare Workers AI, returning typed answers and a fallback envelope when the model is unreachable.

    5 stars · JavaScript · Open-source tool

  43. no published number

    Data and analyticsBetaSDKs and workflow integrations

    jevr

    jevr is a native R client that sends typed questions to Jev and returns structured choice, score and probability answers in R.

    5 stars · R · Open-source tool

  44. no published number

    Developer toolsIn productionSDKs and workflow integrations

    questions (nitoba)

    TypeScript library that defines typed decisions with Zod 4 schemas, answered through Jev on TypeSafe or Vercel, with ordered policies, confidence gates, deadlines, retries and replay; generative models plug in through the AI SDK adapter.

    5 stars · TypeScript · Open-source tool

  45. no published number

    Developer toolsAnnouncedSDKs and workflow integrations

    TypeSafeSharp

    Unofficial .NET SDK wraps TypeSafe's System One API for typed Choice, Noul and Score calls, matching the behavior of the official JavaScript and Python SDKs.

    5 stars · C# · Open-source tool

  46. no published number

    Developer toolsAnnouncedSDKs and workflow integrations

    zio-typesafe-ai

    Scala 3/ZIO client for Jev's System One API with fully typed requests and responses and compile-time-validated Choice and Score criteria.

    5 stars · Scala · Open-source tool

  47. no published number

    Developer toolsIn productionSDKs and workflow integrations

    SystemOneClient

    Elixir client library gives a single interface to Jev and other System One decision models, returning full probability distributions instead of a single pick.

    4 stars · Elixir · Open-source tool

  48. no published number

    AI infrastructureAnnouncedSDKs and workflow integrations

    langgraph-jev

    Jev runs as a LangGraph node that routes tasks deterministically and flags low-confidence cases for human review.

    3 stars · Python · Open-source tool

  49. no published number

    Developer toolsAnnouncedSDKs and workflow integrations

    taurus-jev-sdk-go

    Dependency-free Go client for TypeSafe's System One/Jev API with response validation, masked credentials and bounded retries.

    3 stars · Go · Open-source tool

  50. no published number

    AI infrastructureIn productionSDKs and workflow integrations

    dopp

    Jev-compatible decision calls pass through a proxy that logs every request and trains a small owned model to replace Jev.

    2 stars · TypeScript · Open-source tool

  51. no published number

    AI infrastructureBetaSDKs and workflow integrations

    JDE (Jev Decision Engine)

    Central service where an agent's Jev-backed judgments are requested, thresholded and recorded for downstream systems; no metrics published.

    2 stars · TypeScript · Open-source tool

  52. no published number

    Developer toolsBetaSDKs and workflow integrations

    use-jev

    use-jev is a Claude Code and Codex skill that hands off bounded semantic decisions to Jev via OpenRouter, returning a typed choice or score instead of prose.

    2 stars · JavaScript · Open-source tool

  53. no published number

    Developer toolsIn productionSDKs and workflow integrations

    AgentJ

    Java agent framework treats Jev as a first-class primitive alongside LLM generation and Java code for typed routing, classification, and scoring decisions.

    Open-source tool

  54. no published number

    Developer toolsIn productionSDKs and workflow integrations

    cargo-ai

    Rust CLI agent builder that added TypeSafe Jev as a provider in 0.4.2, so JSON-declared agents ask Jev Choice and Score questions, such as a support department and an urgency score; release qualification runs a live Jev journey.

    Open-source tool

  55. no published number

    AI infrastructureIn productionSDKs and workflow integrations

    Dify

    Dify's Question Classifier workflow node can run Jev as its model, sending each input to Jev to pick which workflow branch handles it, per Dify's announcement.

    Commercial product

  56. no published number

    Developer toolsIn productionSDKs and workflow integrations

    DSPy

    DSPy 3.4.0 adds native Jev support with typed Noul, Choice and Score signatures and a ReAnchor optimizer built for System One models (DSPy).

    Open-source tool

  57. no published number

    Developer toolsBetaSDKs and workflow integrations

    gut (gutpy.dev)

    A Python library resolves one-line YES/NO/UNSURE judgment calls with Jev first, falling back to other small models when needed.

    Open-source tool

  58. no published number

    Developer toolsIn productionSDKs and workflow integrations

    Jev CLI

    npm CLI (jevctl) exposing Jev as shell verbs (compact, screen, verify, extract, route, classify, match, rerank) that return a probability and an exit code for pipelines; a Claude Code plugin runs jev compact as a compaction hook.

    Open-source tool

  59. no published number

    AI infrastructureAnnouncedSDKs and workflow integrations

    jevchain

    Open-source harness composes typed Jev decisions into agent-chain graphs, logging every question, probability and branch not taken.

    Open-source tool

  60. no published number

    Developer toolsAnnouncedSDKs and workflow integrations

    libtypesafe

    Unofficial C++17 client giving C++ apps direct access to TypeSafe's Jev decision endpoint (libtypesafe).

    0 stars · C++ · Open-source tool

  61. no published number

    Developer toolsIn productionSDKs and workflow integrations

    LiteFlow

    LiteFlow's liteflow-agent-jev module adds JevSwitchComponent, a SWITCH node that sends business state and candidate chain IDs to Jev's Choice API and routes to the chosen chain, falling back to DEFAULT below the confidence threshold.

    Open-source tool

  62. no published number

    Developer toolsBetaSDKs and workflow integrations

    LUCID

    LUCID's agentic IDE routes its typed judgments (classifications, yes/no checks, the agent's judge() calls, thinking-effort selection and the browser_run click policy) to Jev when a TypeSafe key is set, in the v2.3.0-beta.1 prerelease.

    Commercial product

  63. no published number

    Developer toolsIn productionSDKs and workflow integrations

    NeuroSquad (JEV Oracle)

    NeuroSquad wires Jev into its 19-agent CLI marketplace as an on-demand JEV Oracle card for decision calls inside the app (company).

    Commercial product

  64. no published number

    Developer toolsIn productionSDKs and workflow integrations

    Nola

    TypeScript superset (.tsi) for typed LLM inference; its typesafe() provider serves Choice, Scale and Prob decision types from Jev as calibrated distributions, alongside OpenAI, Anthropic and Google providers.

    Open-source tool

  65. no published number

    Developer toolsIn productionSDKs and workflow integrations

    Pydantic AI / Logfire

    Jev powers yes/no and pick-from-list decisions in Pydantic AI agents, with Logfire showing each answer's probability. Vendor claims it runs faster and cheaper than frontier models (company).

    Open-source tool

  66. no published number

    Developer toolsIn productionSDKs and workflow integrations

    Replit

    Replit AI Integrations ships native Jev support so apps built on Replit can classify content, route requests and score leads without managing API keys.

    Commercial product

  67. no published number

    AI infrastructureBetaSDKs and workflow integrations

    Venice (AskVenice)

    Venice added Jev to its inference API in beta: an app sends its state and typed questions and gets back a probability, a chosen option or a rubric score for its code to branch on.

    Commercial product

  68. no published number

    AI infrastructureIn productionSDKs and workflow integrations

    Vercel AI Gateway

    Vercel's AI Gateway serves Jev through the TypeSafe client, a POST /v1/evaluate HTTP API and the AI SDK's experimental_evaluate with model typesafe-ai/jev, billing all three paths through the gateway.

    Commercial product

  69. no published number

    Developer toolsAnnouncedSDKs and workflow integrations

    Vercel AI SDK for Python

    Jev is wired into the Vercel AI SDK for Python, demoed deciding live language detection and step by step Python code generation tasks.

    Open-source tool

  70. no published number

    ProductivityIn productionSDKs and workflow integrations

    Watchflows

    Watchflows, a paid macOS automation app, ships three Jev nodes (Yes or No, Choose, Score) that each ask one typed question about the flow's payload and route the flow on the answer or its confidence, with no separate Condition step.

    Commercial product

Before you rely on it

  • Literal. Jev 1.13 answers the question as worded; double negatives and indirection are answered less reliably. TypeSafe
  • No arithmetic. It does not count reliably, reads dates as text rather than ordered values, and cannot reliably tell whether two numbers are close. TypeSafe
  • Noise hurts. Accuracy falls as the state fills with content unrelated to the decision. TypeSafe
  • Steerable. Content written to steer the model can move the answer; one planted false fact flipped every answer in a community test. TypeSafe, radar findings
  • Not repeatable to the point. On a 94-paragraph benchmark, recall ranged from 80.8% to 92.3% across eight shuffled runs of the same items. co-dm study
  • No prose out. It is not trained to generate text; input is text only, as a string, a JSON object or an array of text. TypeSafe

The API, by TypeSafe's numbers

CallPOST https://api.typesafe.ai/v1/systemone with state, model and a map of questions docs
Modeljev-latest, which points at jev-1.13.0 as of 2026-09-24 docs
Question typesChoice (one of up to 255 options), Score (a place on a scale), Noul (a yes/no probability), all against one state docs
Price$0.042 per million input tokens; output tokens free docs
Rate limits250,000 tokens per second and 1,200 requests per minute docs
Request size64k tokens, of which 32k for the state plus the longest question docs
When busy429 for rate limits, 529 for overload; TypeSafe asks for exponential backoff docs
Other routesOpenRouter and Vercel AI Gateway docs

Reading TypeSafe docs · Quick start · Question types · Writing the state · jev-1.13 weaknesses · co-dm benchmark · Skill-selection benchmark

Past highlights