Skip to content

jev-stack

One-command npx installer for a set of Jev-powered agent tools: context compaction, browser agent, prompt enhancer and skill router.

Projectbeta

Overview

jev-stack is an installer for the author's Jev-powered agent toolset. Running `npx jev-stack` opens an interactive menu that explains each component; flags install everything, list components or install a chosen subset. An alternative AGENT_PROMPT.md lets a coding agent walk you through and install the set.

The components are separate repositories: fast-jev-compaction (context compaction, a pre-push guardrail, a code-error finder and natural-language file search), jev-ultrafast (a browser agent at roughly 0.5 seconds per step), jev-prompt-enhancer (rewrites vague prompts), jev-skill-router (one decision per request to choose a skill, plus scanning skills for injection and dangerous commands) and a CDP browser launcher. By default tools call typesafe/jev-1.13 via OpenRouter's decisions endpoint at roughly $0.0001-0.0008 per decision, with direct provider keys also supported. A stated rule is that deterministic checks like tsc, builds and secret scans are never replaced by a model decision.

What the decision model decides

TypeQuestionAnswers
ChoiceWhich skill should the agent load for this request?(inferred)—
OtherPer-tool decisions such as next browser action or which context to keep(inferred)—

How it works

  1. Run npx jev-stack and choose components
  2. Each tool sends one small decision per step to Jev via OpenRouter
  3. Agent acts on the decision (load skill, click, compact, flag)
  4. Deterministic checks remain authoritative

Strengths

  • Single entry point to several related tools
  • README cites $0.0001-0.0008 per decision
  • Model decisions never replace deterministic safety checks

Caveats

  • Installer only; behaviour lives in separate repositories
  • Performance claims are the author's own and not shown here
  • Requires an OpenRouter or provider key

Getting started

npx jev-stack

At a glance

Type
Project
Status
beta
Language
JavaScript
Runtime
Node CLI
Has code
yes

Use cases and integrations

How curated lists describe it

Frequently asked questions

What does jev-stack install?

A menu of the author's tools: compaction and guardrails, a fast browser agent, a prompt enhancer, a skill router and a CDP launcher.

Which Jev endpoint does it use?

By default typesafe/jev-1.13 via OpenRouter with your OPENROUTER_API_KEY.

Does a model decide whether code is safe?

No. The README says deterministic checks are never replaced; Jev only decides where to look.

Related projects

  • Local Python MCP service that keeps the skill catalog out of the agent's context and lets Jev pick which SKILL.md files to load.

    MCP serverbetaPython
  • Claude Code, Codex and pi plugin that hands non-writing agent steps to Jev and escalates writing or unsure steps to the LLM.

    PluginbetaJavaScript
  • MCP gateway that trims noisy tool output and offers task classification and review advice, splitting work between local Laya and Jev.

    MCP serverprototypeTypeScript
  • Dependency-free CLI and Codex skill that send local state to Jev for typed judgments and return answers without echoing the state.

    Agent skillbetaPython
  • Agent-plugin browser tool where Jev picks one operation and one row from a numbered page table; a small LLM only writes typed text.

    ProjectbetaTypeScript
  • Chrome extension plus MCP server so Claude or Codex can drive your real Chrome, with Jev picking each click in about half a second.

    MCP serverbetaTypeScript