Skip to content
Metatext Decide — typed decisions for your agents, with a probability on every answerGet early access
Metatext
Sign Up

jev-claw

OpenClaw plugin adding a jev_route tool that classifies tasks with Jev and applies routing policy as code, with optional automatic routing.

ProjectbetaMIT
View on GitHub ↗

Overview

jev-claw adds one tool, jev_route, to OpenClaw agents to answer which model should do a task that just arrived. Instead of asking an LLM to think about the best model or using keyword rules, it classifies the task with Jev and applies the user's routing policy as ordinary code. The result includes task type, complexity, risk, the chosen route, whether a second opinion is needed and from which route, a confidence and reasons, for example escalating a stuck debugging task after two failed attempts to an independent critic.

The README argues routing is a classification problem needing a fixed answer set, probabilities and repeatable answers, and contrasts it with always using the best model, always using the cheap one, keyword rules, or asking an LLM. Version 1.1 adds opt-in automatic routing: a bounded local prefilter skips casual, writing-only and deterministic turns, and for meaningful software-engineering prompts a typed hook calls Jev once, adds a host-generated routing decision to the turn, and can enforce it when the agent delegates through sessions_spawn. It is MIT licensed.

What the decision model decides

TypeQuestionAnswers
ChoiceWhat type of task is this?—
ChoiceHow complex and risky is it?—
Yes / noDoes this need a second opinion?yes · no

How it works

  1. Task arrives (or prefilter passes a meaningful prompt)
  2. Jev classifies type, complexity and risk
  3. Routing policy in code selects a model route
  4. Optionally enforce when delegating via sessions_spawn

Strengths

  • Auditable, repeatable routing
  • Local prefilter avoids unnecessary calls
  • Second-opinion routing for stuck tasks

Caveats

  • Policy must be configured per setup
  • OpenClaw-specific

At a glance

Type
Project
Status
beta
Language
TypeScript
Runtime
OpenClaw plugin
Has code
yes
License
MIT

Use cases and integrations

How curated lists describe it

Frequently asked questions

What does jev_route return?

Task type, complexity, risk, a route, second-opinion advice, confidence and reasons.

Is routing automatic?

Optionally, since version 1.1, with a local prefilter and a hook.

Why not ask an LLM which model to use?

The README calls that expensive, slow, non-deterministic and unauditable.

Related projects

  • Helm

    Orchestrator that probes which coding agents are installed and logged in, then has Jev pick the best one for a task.

    Agent skillbetaPython
  • jev-agent-hooks

    Hooks for Claude Code, Codex and pi that use Jev to suggest a skill each turn, plus a rule gate for subagent model routing.

    HookbetaJavaScript
  • jev-pilot

    Claude Code plugin that asks Jev before each turn for reasoning effort, subagent model, strategy and skill, with a decision log.

    Agent skillbetaTypeScript
  • JevRouter

    Router that puts models, subagents, skills, MCP tools and CLIs in one candidate set and asks Jev one Choice about which handles the step.

    MCP serverbetaTypeScript
  • TriRouter

    Prompt router for Claude Code, Codex and Antigravity that picks model, effort, skill and language per request via Jev.

    ProjectbetaPython
  • WrongStack

    Open-source AI coding agent with memory and specialist agents; the list notes an optional Jev dispatch classifier.

    ProjectbetaTypeScript