Skip to content

Jev Zig CLI

jevx is a Zig terminal agent that keeps Codex for code generation and confines Jev, reached through OpenRouter, to bounded choice, noul and score decisions for routing, hazard checks, impact and verification. The Zig side, not the model, holds the permissions and the final control flow.

Stars
1
Language
Zig
License
MIT
Published by
Community
Author
SupratimSircar05
Last verified
Oct 2026

About this project

An auditable terminal agent that combines Codex coding with Jev typed decisions and deterministic policy enforcement. Built in Zig. jevx uses the authenticated OpenAI Codex CLI for generative coding work. TypeSafe's Jev, accessed through OpenRouter, is restricted to bounded choice, noul, and score decisions for routing, hazard detection, impact scoring, and verification. Ordinary Zig code—not Jev—owns permissions and final control flow.

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

Highlights

  • One validated preflight Jev request batches action routing, five hazard checks and impact scoring before Codex ever receives the prompt.
  • Permissions are deterministic Zig: aggressive allows automatic action only at route confidence 0.70+, impact under 1.5 and every hazard noul under 0.20.
  • Audit records are redacted, encrypted, authenticated, sequenced and hash-chained, and a failed durable pre-action write blocks the mutation.
  • Codex owns its ChatGPT OAuth credentials and jevx never reads them; the OpenRouter key lives in Keychain, Credential Manager or a session variable.

Watch out

Every coding turn needs both providers, an authenticated Codex login and an OpenRouter key for Jev. That is deliberate: if Jev is unreachable the turn stops before Codex receives the prompt rather than falling back to read-only, and only doctor, audit, version, policy explain and standalone decide keep working. Release archives bundle a pinned Codex runtime, and the app-server backend is preview-only, since with that pin jevx rejects it and falls back to exec because the release cannot verifiably ignore inherited user integrations. Windows x86_64 is manual-install only, the app-server JSONL path is contract-tested rather than exercised in production, and the project states it claims no better model intelligence and no desktop-app parity. README ends mid-table on the JSONL output. 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.

More Jev resources like this

9GitHub stars
Research-memory skill that keeps experiment claims traceable with ten operations, provenance gates, and optional batched Jev audits over memory files.
Jev Practices & Patterns#community#awesomejev#catalog-import
11GitHub stars
Local memory layer that stores verbatim agent quotes in per-project SQLite and grades them with Jev into stale, wrong, unsupported, or keep verdicts.
Jev Practices & Patterns#community#awesomejev#catalog-import
Minecraft setup where Hermes plans at milestones, Jev picks one bounded harness action per step, and Mineflayer code owns validity and execution with no screenshots.
Jev Practices & Patterns#community#awesomejev#catalog-import
7GitHub stars
Local snake-arena demo where you steer a golden snake against four NanoJev AI snakes, with each AI snake showing its live direction-probability bars from a single forward pass.
Jev Practices & Patterns#community#awesomejev#catalog-import
20GitHub stars
Local harness that runs the open Laya decision model on your own machine and wires it into browser agents through a TypeSafe-compatible /v1/systemone dialect.
Jev Practices & Patterns#community#awesomejev#catalog-import
16GitHub stars
Free hands-on course of 12 notebooks that pairs Jev decisions with LLM writing across 13 small tools, needing one OpenRouter key and printing cost and self-checks per lesson.
Jev Practices & Patterns#community#awesomejev#catalog-import

From the Jev guides

Original write-ups on what this entry is for.

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
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
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