jev-orchestrator
Node experiment where Jev picks the next step around a coding agent, with approved actions and Codex or Claude Code delegation.
Overview
Jev Orchestrator by petercr is a deliberately small Node and TypeScript experiment testing whether Jev can make useful next-step decisions around a coding agent. In the default decision-only mode it inspects a repository, sends a compact state object to Jev through a selected provider (TypeSafe, Vercel AI Gateway or OpenRouter), applies deterministic policy thresholds, prints the result and records a JSONL trace.
An explicit orchestrate mode adds a bounded, manually approved loop over safe actions: repository searches with ripgrep, bounded file reads, fixed read-only Git diagnostics, detected validation scripts, public GitHub issue reads, and delegation to Codex CLI or Claude Code. It cannot run arbitrary commands, and every executable candidate requires approval. A matched benchmark recorded twelve first-attempt observations on a bug-and-tests task, which the README says does not establish a general advantage. Support for Cloudflare Clef and OpenAI Decisions is planned.
What the decision model decides
| Type | Question | Answers |
|---|---|---|
| Choice | What is the safest useful next action for this repository?(inferred) | search · read file · git diagnostics · run validation · read issue · CALL_CODEX · CALL_CLAUDE |
How it works
- Inspect the repository and build a compact state
- Ask Jev for the next action
- Apply deterministic policy thresholds
- Print the decision, or in orchestrate mode request approval and execute
- Record every step in a JSONL trace
Strengths
- Decision-only default is safe to try
- Every executable action needs explicit approval
- Supports three Jev providers
Caveats
- Small benchmark sample with no general claim
- Requires Node 22+, pnpm and ripgrep
- Delegation requires authenticated Codex or Claude Code CLIs
Getting started
pnpm install, cp .env.example .env (add key), then pnpm exec node --env-file=.env --import tsx src/cli.ts -- . "Inspect this repo and choose the safest useful first action"At a glance
- Type
- Project
- Status
- prototype
- Language
- TypeScript
- Runtime
- Node CLI
- Has code
- yes
- Has tests
- yes
Use cases and integrations
How curated lists describe it
An mini node orchestrator that uses Jev to handle routing to different LLMs…
wh000wh000/awesome-jev-live
Frequently asked questions
What does Jev decide?
The next action around a coding agent, from a bounded set of safe actions or delegation.
Can it run arbitrary commands?
No. Only bounded actions, each requiring explicit approval.
Which providers are supported?
TypeSafe direct, Vercel AI Gateway and OpenRouter, selected explicitly.
Related projects
One skill and MCP server for Claude Code, Codex and Hermes where Jev gates large-model turns, routes models and guards Git steps.
MCP serverbetaPythonClaude Code function hook that sends the agent's this-or-that questions to Jev and optionally ranks installed skills per prompt.
ProjectbetaTypeScriptPython harness for open-ended goals: AI-authored tools and checks, Jev choosing the next tool and arguments from fresh state.
ProjectbetaPythonPython agent runtime where Jev Choice decisions pick tools and arguments, and low-confidence picks go to an LLM for review.
ServicebetaPythonDecision middleware for AI agents that uses local Laya to pick tools and judge results, falling back to an LLM when unsure.
ProjectbetaPythonSelf-improving multi-platform agent harness with an optional Jev companion for typed judgments and per-session model routing.
ServicebetaTypeScript