Find a Jev implementation for agent routing, browser actions, coding tools or context compaction. See which decision Jev handles, how it is integrated and whether the source evidence has been reviewed.
Community directory snapshot 2026-09-28 Review method 2026 ecosystem report
The catalog includes 1,334 repository or integration records, 61 article or discussion records and 33 other resources. Repository links resolve to 1,332 unique GitHub repositories; a public link alone does not establish an open-source license.
Jev projects apply Choice, Score, or Noul to bounded judgments such as routing, safety gates, browser actions, context compaction, and robot simulation. Start with the popular projects, then filter by use; use the model directory for local alternatives.
For deeper examples by task, inspect jev-git for a pinned Git-hook source review, Oh My Pi for an Agent integration, or jev-trader for dry-run versus live-trading boundaries. Stars signal attention, not correctness, reproducibility, or safety.
These comparisons explain implementation differences at reviewed revisions. Follow each analysis to the primary evidence; star counts do not establish outcomes.
The browser project builds operation and target tables from the DOM, using a text model for TYPE_TEXT. The macOS project merges OCR and accessibility elements and handles off-screen controls and post-input checks.
Choose by observation interface first: a web DOM is not interchangeable with native desktop controls. Neither establishes unattended reliability across arbitrary sites or apps.
The joelhooks version filters model context while retaining the JSONL transcript and rebuilding branch state from a decision ledger. The joslynSmall version puts tool evidence into a compaction summary, whose total budget still limits retention.
For auditability, inspect transcript storage and branch replay. For evidence inside a summary, inspect budget-exhaustion behavior. “Never loses original text” is not a property shared by all compaction approaches.
memsearch uses per-candidate Noul relevance judgments, retains candidate text and propagates failures. Kody uses a 0–3 Score in batches of eight and returns the hybrid order after low mean confidence or failure.
Decide which layer owns failures and which quantity drives ordering. Thresholds in source are author policies, not accuracy calibrated on your retrieval data.
Ten widely followed projects spanning browser automation, context compaction, trading, official skills, computer use, review, orchestration, and data.
Search by project, author, use, or tag, then filter by category or activity. Open project details to see how it uses Jev and review its source links.
langchain-ai
An optional Jev classifier integration for Python LangChain workflows.
Submits binary, categorical and ordered-score questions and returns typed answers with probabilities.
Adds a classification step that can participate in existing LangChain pipelines.
An optional Jev classifier integration for Python LangChain workflows.
Submits binary, categorical and ordered-score questions and returns typed answers with probabilities.
Adds a classification step that can participate in existing LangChain pipelines.
This entry covers the TypeSafe partner module, not a claim that LangChain uses Jev by default.
virattt
An educational AI hedge-fund prototype with an optional Jev adapter for structured judgments in fund decision workflows.
Converts strategy questions to System One requests and normalizes native answers to the project’s result format.
Allows Jev or other model backends to be used in the same research workflow.
An educational AI hedge-fund prototype with an optional Jev adapter for structured judgments in fund decision workflows.
Converts strategy questions to System One requests and normalizes native answers to the project’s result format.
Allows Jev or other model backends to be used in the same research workflow.
The author describes an educational research prototype that currently does not execute trades. It is not an official Jev adapter, a zero-hallucination system, or evidence of profitability. This site has not backtested it.
BerriAI
LiteLLM can use Jev to classify requests for its complexity-based model router.
Maps requests to configured complexity classes that drive backend routing.
Makes the complexity judgment behind routing inspectable.
LiteLLM can use Jev to classify requests for its complexity-based model router.
Maps requests to configured complexity classes that drive backend routing.
Makes the complexity judgment behind routing inspectable.
README and integration source reviewed at a fixed commit; not independently run or benchmarked by this site.
can1357
Oh My Pi includes an optional TypeSafe judgment provider for bounded decisions in coding-agent workflows.
Sends agent state and typed questions to Jev and parses structured answers.
Adds a replaceable judgment provider to existing agent workflows.
Oh My Pi includes an optional TypeSafe judgment provider for bounded decisions in coding-agent workflows.
Sends agent state and typed questions to Jev and parses structured answers.
Adds a replaceable judgment provider to existing agent workflows.
README and integration source reviewed at a fixed commit; not independently run or benchmarked by this site.
davila7
A community Claude Code Templates mod that uses Jev to suggest subagent models and reasoning levels.
Evaluates task tier, reasoning needs and production risk; local policy maps results to invocation settings.
Makes routing rules and per-request choices explicit and configurable.
A community Claude Code Templates mod that uses Jev to suggest subagent models and reasoning levels.
Evaluates task tier, reasoning needs and production risk; local policy maps results to invocation settings.
Makes routing rules and per-request choices explicit and configurable.
A davila7 community project, not an official Anthropic repository or mod. Aggregate savings and safety effects were not verified here.
ComposioHQ
An optional TypeSafe provider for Composio that uses Jev to choose among tools and bounded argument options.
Uses a routing Choice, Noul action gates and finite argument questions, then returns abstain, partial or call. Missing required arguments remain explicit; destructive tool calls retain a confirmation requirement.
Separates selecting a tool from binding its arguments and executing it, allowing callers to handle no-action, uncertainty and partial inputs explicitly.
An optional TypeSafe provider for Composio that uses Jev to choose among tools and bounded argument options.
Uses a routing Choice, Noul action gates and finite argument questions, then returns abstain, partial or call. Missing required arguments remain explicit; destructive tool calls retain a confirmation requirement.
Separates selecting a tool from binding its arguments and executing it, allowing callers to handle no-action, uncertainty and partial inputs explicitly.
Pinned provider decision code reviewed, not a live Composio run. The adapter's thresholds and required-confirmation flag are application rules; they do not prove safe execution unless the caller enforces them.
vercel
The TypeSafe provider in AI SDK lets TypeScript applications call Jev through the shared evaluate interface.
The TypeSafe evaluation provider maps AI SDK boolean questions to native Noul and converts the responses back. Native Choice/Score confidence is retained in providerMetadata.typesafe.confidence rather than the generic answer object.
Allows a TypeScript AI SDK application to use a common evaluation interface while keeping text generation and finite decisions separate.
The TypeSafe provider in AI SDK lets TypeScript applications call Jev through the shared evaluate interface.
The TypeSafe evaluation provider maps AI SDK boolean questions to native Noul and converts the responses back. Native Choice/Score confidence is retained in providerMetadata.typesafe.confidence rather than the generic answer object.
Allows a TypeScript AI SDK application to use a common evaluation interface while keeping text generation and finite decisions separate.
Pinned adapter source reviewed. No live inference was run; native and AI SDK response fields differ, so this integration is not evidence that a generic evaluation probability equals TypeSafe confidence.
trycua
Cua’s preview jev-use example pairs Driver observation and execution with bounded Jev browser-action choices.
Reads DOM or supported visual-region descriptions and returns a supplied candidate action ID.
Provides Python and TypeScript loops with separate offline and live checks.
Cua’s preview jev-use example pairs Driver observation and execution with bounded Jev browser-action choices.
Reads DOM or supported visual-region descriptions and returns a supplied candidate action ID.
Provides Python and TypeScript loops with separate offline and live checks.
The Jev integration is a public-preview example, not the entire Cua framework or CUA-S1. Offline fixture checks do not establish live model performance. README and integration source reviewed at a fixed commit; not independently run or benchmarked by this site.
pydantic
An optional TypeSafe provider and Jev model integration for Pydantic AI.
Converts supported structured output fields into typed Jev questions and maps answers back to the output model.
Lets supported classification outputs use the existing Pydantic AI interface.
An optional TypeSafe provider and Jev model integration for Pydantic AI.
Converts supported structured output fields into typed Jev questions and maps answers back to the output model.
Lets supported classification outputs use the existing Pydantic AI interface.
An optional provider rather than a general chat-model replacement; supported output types depend on this implementation. Not run here.
elizaOS
An optional TypeSafe HTTP adapter in Eliza’s source, not registered with the Agent runtime by default.
Only an explicit systemOne call sends state and questions, returning validated typed answers.
Provides a reusable server-side module; its presence is not evidence of deployed Jev adoption.
An optional TypeSafe HTTP adapter in Eliza’s source, not registered with the Agent runtime by default.
Only an explicit systemOne call sends state and questions, returning validated typed answers.
Provides a reusable server-side module; its presence is not evidence of deployed Jev adoption.
The adapter is not registered with the runtime by default; source presence does not establish real adoption or deployment.
browser-use
A browser Agent that uses Jev to choose actions and page controls, calling a text model only when input is needed.
Chooses an action and its matching DOM element in one request; a text model generates input text.
Separates UI choices, text generation, and browser execution so each step can be inspected.
A browser Agent that uses Jev to choose actions and page controls, calling a text model only when input is needed.
Chooses an action and its matching DOM element in one request; a text model generates input text.
Separates UI choices, text generation, and browser execution so each step can be inspected.
The author reports a median of about 7.1 seconds on one Google Flights task, with three runs per setup. Initial navigation is excluded; this is not a general Agent benchmark and was not independently retested.
langchain-ai
An optional LangChain.js TypeSafeClassifier integration for sending state and predefined questions to Jev.
Uses invoke to call TypeSafe and parse choice, noul, score and probability fields.
Adds typed decisions to LangChain workflows without treating Jev as a chat generator.
An optional LangChain.js TypeSafeClassifier integration for sending state and predefined questions to Jev.
Uses invoke to call TypeSafe and parse choice, noul, score and probability fields.
Adds typed decisions to LangChain workflows without treating Jev as a chat generator.
Reviewed the fixed-version README and call path.
vercel-labs
A Jev UI-composition experiment inside the json-render website that selects from predefined component and property options.
Evaluates component configurations through Vercel AI Gateway, then composeSpec assembles the UI specification.
Provides an inspectable composition path alongside token-by-token JSON generation.
A Jev UI-composition experiment inside the json-render website that selects from predefined component and property options.
Evaluates component configurations through Vercel AI Gateway, then composeSpec assembles the UI specification.
Provides an inspectable composition path alongside token-by-token JSON generation.
This covers the website’s Jev experiment, not the whole framework. Claims such as 3.21 s versus 880 ms were not independently verified.
openchamber
OpenChamber’s optional automatic model router uses Jev to classify a message before selecting a configured model and reasoning level.
Jev selects a task category; local category mappings determine the model configuration.
Keeps routing categories and their model assignments explicit.
OpenChamber’s optional automatic model router uses Jev to classify a message before selecting a configured model and reasoning level.
Jev selects a task category; local category mappings determine the model configuration.
Keeps routing categories and their model assignments explicit.
Only the optional routing module was inspected; cost or quality improvements for model combinations were not verified.
0xPlaygrounds
An experimental TypeSafe crate in Rig for expressing Jev questions and answers with Rust types.
Sends application state and questions to Jev and parses Choice, Score or Noul answers.
Reuses field layouts between questions and answers and validates answers against questions.
An experimental TypeSafe crate in Rig for expressing Jev questions and answers with Rust types.
Sends application state and questions to Jev and parses Choice, Score or Noul answers.
Reuses field layouts between questions and answers and validates answers against questions.
README and integration source reviewed at a fixed commit; not independently run or benchmarked by this site.
zilliztech
Deep Searcher includes a frozen-query evaluation of Jev as a search-stopping judge; this experiment is separate from the full research agent.
Jev judges enough_information from the main query and accumulated intermediate context; code compares stopping decisions under a bounded round protocol.
Makes stopping behavior and call traces inspectable; the protocol explicitly does not score final-answer quality.
Deep Searcher includes a frozen-query evaluation of Jev as a search-stopping judge; this experiment is separate from the full research agent.
Jev judges enough_information from the main query and accumulated intermediate context; code compares stopping decisions under a bounded round protocol.
Makes stopping behavior and call traces inspectable; the protocol explicitly does not score final-answer quality.
Fixed-version documentation and implementation material were inspected on 2026-10-24. No 100-query comparison or live API was run. The evaluation directory does not prove the main application enables this path by default.
zilliztech
Semantic cache with a Jev evaluator that uses Noul judgments to check whether a cached response can serve an incoming request.
Five Noul dimensions check cached-answer compatibility; the evaluator returns their minimum score, and a request failure returns zero.
Prevents a strong task match from averaging away weak format or freshness compatibility, without claiming to verify the answer’s truth.
Semantic cache with a Jev evaluator that uses Noul judgments to check whether a cached response can serve an incoming request.
Five Noul dimensions check cached-answer compatibility; the evaluator returns their minimum score, and a request failure returns zero.
Prevents a strong task match from averaging away weak format or freshness compatibility, without claiming to verify the answer’s truth.
Fixed-version documentation and implementation material were inspected on 2026-10-24. No cache service or live request was run. The evaluator’s float conversion is not an independent calibration or correctness guarantee.
kunchenguid
Firstmate can optionally use Jev to match task briefs to dispatch rules before local policy chooses an Agent configuration.
Sends the task brief and candidate rules to Jev, then resolves execution profiles with confidence and local conditions.
Separates semantic matching from the final dispatch policy.
Firstmate can optionally use Jev to match task briefs to dispatch rules before local policy chooses an Agent configuration.
Sends the task brief and candidate rules to Jev, then resolves execution profiles with confidence and local conditions.
Separates semantic matching from the final dispatch policy.
This is opt-in dispatch assistance, not a claim that the whole Firstmate workflow uses Jev by default. Task briefs are sent to TypeSafe.
tamaratran
Prunes old tool calls and outputs for Claude Code while retaining surviving text verbatim.
Separately judges whether a tool call and its full output are still needed; code keeps, truncates or drops them.
Avoids rewriting paths, commands and errors into a new summary.
Prunes old tool calls and outputs for Claude Code while retaining surviving text verbatim.
Separately judges whether a tool call and its full output are still needed; code keeps, truncates or drops them.
Avoids rewriting paths, commands and errors into a new summary.
Requests are split to fit budgets, rather than always using one call. Pruning can remove useful context; the animation is not a live API benchmark. README and integration source reviewed at a fixed commit; not independently run or benchmarked by this site.
jaredpalmer
Tiny Jev-like decision head built atop Qwen2.5-0.5B that can be trained and run locally on Apple Silicon MacBooks.
Attaches a parallel decision head to an open 0.5B model to answer discrete questions directly from token activations.
Provides a small local decision-model architecture and probes for packed-versus-separate questions, option order and calibration. Hosted teacher calls for data collection remain a separate network and cost dependency.
Tiny Jev-like decision head built atop Qwen2.5-0.5B that can be trained and run locally on Apple Silicon MacBooks.
Attaches a parallel decision head to an open 0.5B model to answer discrete questions directly from token activations.
Provides a small local decision-model architecture and probes for packed-versus-separate questions, option order and calibration. Hosted teacher calls for data collection remain a separate network and cost dependency.
The pinned README and hosted-teacher adapter were read. This site did not train the student or reproduce its calibration figures. Local inference capability does not mean every data-generation step is offline or free.
jev-chat
An on-device conversation copilot for WeChat, QQ, X, and Feishu that reads the visible chat, suggests replies, and fills the input field only when asked. It reads the screen without hooking or modifying the host app; the user still decides whether to send.
This listing includes an overview and project link; a reviewed Jev decision point is not yet available.
Open the project link to review its source material and implementation.
An on-device conversation copilot for WeChat, QQ, X, and Feishu that reads the visible chat, suggests replies, and fills the input field only when asked. It reads the screen without hooking or modifying the host app; the user still decides whether to send.
This listing includes an overview and project link; a reviewed Jev decision point is not yet available.
vercel
Vercel's open agent framework, which ships Jev as the default evaluation model in its experimental evaluate path.
This listing includes an overview and project link; a reviewed Jev decision point is not yet available.
Open the project link to review its source material and implementation.
Vercel's open agent framework, which ships Jev as the default evaluation model in its experimental evaluate path.
This listing includes an overview and project link; a reviewed Jev decision point is not yet available.
agentgateway
A Jev guardrail example in Agentgateway using a webhook to inspect model requests and responses.
Jev scores jailbreaks, harmful content and secret disclosure; thresholds or evaluation errors reject requests.
Demonstrates gateway integration, not guaranteed prevention of all attacks.
A Jev guardrail example in Agentgateway using a webhook to inspect model requests and responses.
Jev scores jailbreaks, harmful content and secret disclosure; thresholds or evaluation errors reject requests.
Demonstrates gateway integration, not guaranteed prevention of all attacks.
Reviewed the fixed-version README and call path.
latitude-dev
Latitude includes an optional Jev preclassifier for conversation checks and their selection records.
Judges which checks apply and can add checks when thresholds and rate limits permit.
Records models, thresholds, latency and selection reasons alongside the baseline.
Latitude includes an optional Jev preclassifier for conversation checks and their selection records.
Judges which checks apply and can add checks when thresholds and rate limits permit.
Records models, thresholds, latency and selection reasons alongside the baseline.
README and integration source reviewed at a fixed commit; not independently run or benchmarked by this site.