What problem does JevLang solve?

It takes decisions that usually live inside a prompt — routing, triage, approvals, escalation, guarding an agent's tools — and turns them into a declared, validated, inspectable policy file, so the logic is code you can read instead of part of the prompt.

Does the model choose the branch?

No. JevLang asks the model only the questions the policy needs, checks the answers and decides itself, and it can prove from a signed journal exactly why it decided that. A confidence gate above every question escalates to a human instead of guessing.

Install with bun add jevlang. There are no runtime dependencies and it needs Node 22+, so it drops into an existing project. Pure decisions, validation and replay work fully offline — no account and no network required.

Can I audit and replay decisions?

Yes. Each decision carries its readings, its confidence and the clause that fired, and recorded runs become fixtures that replay byte-for-byte, so you can diff a policy change against real history before it ships.

Does it work in both TypeScript and Python?

Declare the policy in either language. policy.toJSON() is a frozen artifact with a fingerprint and compile(json) loads it in the other language; the shared validator rejects the same mistake in both languages and in raw JSON before anything is deployed.

How are actions guarded?

Dispatching an action can require confirmation, respect cooldowns and budgets, pass per-target guards, and roll back through declared inverses, with an audit record for every dispatch so nothing silently goes nowhere.

Is this the same jevlang already in the directory?

No. The existing jevlang entry is sumanmichael/jevlang, a Python DSL where .jev files are rewritten to plain Python at import time. This project is timmikeladze/jevlang, a TypeScript and Python policy engine, which is why it carries the owner-prefixed id.

Jev directory

CommunityJev Repos & SDKs

JevLang — timmikeladze

A policy engine for LLM decisions: declare routes, gates and actions once in TypeScript or Python and every decision comes back validated, explainable, replayable and audited. The model answers only declared questions — application code owns the branch.

See highlights Open github.com

Stars: 1
Language: JavaScript
License: MIT
Reading time: 2 min
Published by: Community
Author: timmikeladze
Last verified: Oct 2026

timmikeladze/jevlang

User repository on GitHub

View on GitHub

A policy engine for LLM decisions: declare routes, gates and actions once in TypeScript or Python, and every decision comes validated, explainable, replayable and audited.

Forks: 0
Last pushed: Updated Oct 2026

Repo stats from the GitHub API, cached Oct 2026.Homepage

This is one of several Jev listings for the same project. The canonical entry is Jevlang — sumanmichael.

Tagscommunity catalog-import logicrw policy-engine typescript python

Works forJev for Routing Jev for Decision gating Jev for Tool calling

Used inJev for Software & DevTools Jev for Customer Support

Languages: TypeScript Python
Capabilities: Open sourceRuns locally

Highlights

Declare routes, thresholds, gates and fallbacks in plain TypeScript or Python; a typo like department.is('billling') is a build error, not a dead branch.
Three question kinds — noul, choice and score — with a confidence gate above every question that escalates to a human instead of guessing.
policy.toJSON() is a frozen artifact with a fingerprint; compile(json) loads it in the other language, verified identical in TypeScript and Python.
Actions are guarded, idempotent and journaled: confirmation, cooldowns, budgets, per-target guards and rollback through declared inverses.
No runtime dependencies and Node 22+; pure decisions, validation and replay work fully offline with no account and no network.

Quickstart

import { noul, definePolicy, rule, assign, hold } from 'jevlang';

const spam = noul('spam?', 'Is this message spam?');
const policy = definePolicy({
  name: 'hello',
  questions: [spam],
  route: {
    clauses: [rule(spam.yes(0.9), hold({ reason: 'almost certainly spam' }))],
    otherwise: assign('inbox'),
  },
});

Watch out

Copy reviewed against the repo README on 2026-10-15; the README's question-kind table is truncated in the digest and no benchmarks or pricing are published. An existing catalog entry with id jevlang is a different project (sumanmichael/jevlang, a Python DSL), so this import uses the owner-prefixed id timmikeladze-jevlang.

Questions about this entry

Short answers drawn from the linked project or page.

From the Jev guides

Original write-ups on what this entry is for.

The Open Reimplementations: Local Models That Speak System One

Sep 22, 2026

Closed weights have not stopped the ecosystem from rebuilding the shape in the open. API-compatible servers, retrained heads on small models, and inference-time tricks that skip generation entirely.

Jev Repos & SDKs#open-models#research#local

Read article

From the Jev AI model community

Posts from builders shipping with Jev right now.

Follow @typesafeai