Does installing this restrict which native tools my agent can run?

No. The root tool allow-list is ["*"], so Jev imposes no extra restriction on native tools and the host still decides execution permissions. What the gate can change is delegation and which tools the root session may run while a job is admitted.

Is model routing actually calibrated, or just a confidence number?

The README is explicit that the 0.6 and 0.8 confidence criteria are policy inputs, not measured success rates. Product-role downgrades in both hosts use the 0.6 confidence, control and risk criteria while upgrades and unknown relationships keep the 0.8 default, and results vary by task.

What exactly leaves my machine when a key is configured?

Gate, the Router and Evidence semantic search can send a request or some source text to TypeSafe. The repository as a whole is not uploaded. Compact and Output never call Jev, and exact-symbol lookups plus Evidence read-backs are not sent. Without a key, Gate, Lean and Router preserve native behavior and Evidence searches locally.

Why do I have to restart the host after installing?

Claude Code reads some execution settings before it loads plugins, so the first start shows a notice asking for one restart after the plugin prepares hooks, worker launch settings and trace directories. An already-running host also retains its launch environment and provider. These are host lifecycle requirements, not extra configuration. In Codex, jev_agent action: "cancel" releases ownership only after an observed terminal event.

What model targets are eligible by default?

Simple lookup and mechanical worker tasks prefer current Haiku in Claude or current account-listed Luna in Codex, and new automatic targets exclude previous generations. Automatic Claude Fable and Codex Astra targets are off by default; enable both hosts with frontierEnabled set to true in the jev-gate config file, and an explicit false overrides the legacy allow options.

How does background delegation interact with the main conversation?

Owned workers, planners, and Lean executors dispatch in the background by default, and new messages steer the main conversation without replacing the original job or its contracts. A launch receipt is separate from completion and contract acceptance, and interrupting the main turn does not cancel an already running background worker.

How are model facts kept current?

Every pull request runs a models CI job against the official Claude model and effort documentation, the Claude Code changelog, and the OpenAI Codex catalog. Stale Claude facts fail CI and produce a model-update patch artifact, and the models:check and models:update commands exercise the same checks locally.

CommunityJev Practices & Patterns

Jev Gate — MongLong0214

Plugin for Claude Code and native Codex that layers Jev-assisted task routing, evidence search, context compaction and readable tool output on top of the host, where typed judgments inform policy but the host keeps ownership of permissions and execution.

See highlights

Open github.com

Stars: 3
Language: TypeScript
Published by: Community
Author: MongLong0214
Last verified: Oct 2026

MongLong0214/jev-gate

User repository on GitHub

View on GitHub

Not every coding task needs your best model. Experimental Jev-powered model routing for Claude Code — V3 prototype runs today, V4 routes at the task boundary.

Forks: 0
Last pushed: Updated Oct 2026

Repo stats from the GitHub API, cached Oct 2026.

About this project

Jev-assisted model routing, delegation, evidence search, and context compaction for Claude Code and native Codex. Jev judges the task; code applies the policy; your coding host executes it. One local dashboard shows both hosts, their decisions, and observed results. Install the plugin and enter your Jev API key. No project config, shell exports, trace-directory setup, dashboard terminal, or claude --debug flag is required. Node.js 22+ and your host's native login are required.

Quoted from the project's README. Read the full README.

Tagscommunity jevbest catalog-import awesomejev

Works forJev for Classification Jev for Routing

Used inJev for Software & DevTools

Languages: TypeScript

Highlights

Installed in Claude Code with two slash commands, then a Jev API key in the local screen that opens; no checkout, npm step or config file.
Key stored under $XDG_CONFIG_HOME/jev-gate/auth/credentials.json in a 0700 directory and 0600 file, never in a project, URL, argument or log.
Owned planners and workers run in the background, so the main conversation returns after dispatch and a new message does not cancel the job.
Node 22.15+; the automatic Codex connection was tested with Codex CLI 0.160.0.

Watch out

A TypeSafe key is not a Claude login, and TypeSafe billing and limits are separate from a Claude subscription. The root tool allow-list is ["*"]: Jev adds no restriction, the host decides execution permissions. Everything defaults to on, including main model routing, manual Compact and remote Evidence, with up to 16 isolated workers and plans of up to 64 tasks. Router preparation is bounded to 800 ms by default and requests that cannot change the final settings are skipped. With a key, Gate, the Router and Evidence can send a request or some source text to TypeSafe; Compact and Output do not call Jev. An already-open host keeps its original model provider, so a fresh host is needed after install, and Codex may require hook trust approval. A start notification is not an accepted result, and no end-to-end speed or token saving is guaranteed. Copy reviewed against the repository README on 2026-10-04, not from running it.

Questions about this entry

Short answers drawn from the linked project or page.

Other listings of this project

Forks and copies of this project in the directory. This entry is the canonical one.

jev-gate

A tool-call gate for Claude Code, Codex, Cursor, and other harnesses backed by Jev: deterministic rules and a read-only allowlist run first, then one request answers allow, ask, or deny in about 400 ms.

Jev Tools & CLIs#community#python#guardrails

Community

eugeniughelbur/jev-engineeringPython

More Jev resources like this

jev-lm

Word-level language model experiment that uses Jev as its output layer, with explicit tokenizer, sampler, repetition mask, stop rules, and chunk verification.

Jev Practices & Patterns#community#jevbest#catalog-import

Community

y0usaf/jev-lmTypeScript

jevgpt

Chat client over jev-latest, which cannot generate text: the transcript becomes the state and each step is a choice over a 20,000-word dictionary, so replies are built one word at a time.

Jev Practices & Patterns#community#jevbest#catalog-import

Community

Bewinxed/jevgptTypeScript

jev-starter

A TypeScript starter for the layer around the model: decision contracts, confidence thresholds that route to auto, fallback or human review, deterministic mock providers, and an offline evaluation harness.

Jev Practices & Patterns#community#awesomejev#catalog-import

Community

hamakyo/jev-starterTypeScript

jev-routing-experiment

Runs Jev as an LLM router over two public benchmarks, with a no-Jev ablation and RouterArena's own scorer so the numbers sit next to the public leaderboard.

Jev Practices & Patterns#community#jevbest#catalog-import

Community

TokenTrim/jev-routing-experimentPython

jev-headline-bench

Upworthy's randomized headline A/B tests, filtered to variants that shared an image: Jev picks the click winner 64.5% of the time on a held-out confirmatory split, and 74.7% when the real gap was decisive.

Jev Practices & Patterns#community#jevbest#catalog-import

Community

Gaurav-Gosain/jev-headline-benchGo

jev-lab

A small Node.js TypeScript CLI for trying Jev on synthetic support text: one urgency question, three questions in a single request, or a latency comparison. No web server and no UI.

Jev Practices & Patterns#community#jevbest#catalog-import

Community

Menny1337/jev-labTypeScript

From the Jev guides

Original write-ups on what this entry is for.

Designing the State: What to Include and What to Drop

Oct 12, 2026

Most Jev accuracy problems are state problems. What belongs in a state, what is a pre-made judgment to leave out, and why naming the slice you ask about matters more than the wording.

Jev Practices & Patterns#state#design#primitives

Read article

Noul, End to End: One Probability and a Dozen Thresholds

Oct 12, 2026

A Noul returns one number and no separate confidence. How to read the value, pin its boundary with criteria, split compound questions, and set a per-action threshold that matches the cost of being wrong.

Jev Practices & Patterns#noul#primitives#confidence

Read article

Scoring Without Measuring: Bands, Rubrics, and Composite Triage

Oct 12, 2026

A Score is a weighted position, not a ruler. Why the number can land between levels, why levels are judged separately, and how to combine several atomic scores in code instead of one crowded rubric.

Jev Practices & Patterns#score#primitives#scoring

Read article

Back to all resources

From the Jev community

Posts from builders shipping with Jev right now.

Follow @typesafeai