jev-planner
CLI where several coding agents draft implementation plans and Jev scores them, decides on cross-review and picks the finaliser.
Overview
jev-planner produces repository-aware implementation plans from several AIs that challenge each other, with TypeSafe's Jev making the final calls. Two or more agents draft plans independently: agent CLIs like Codex and Claude Code inspect the repository, while chat APIs such as DeepSeek, Kimi and GLM get a snapshot. Jev scores each plan for completeness, feasibility and risk coverage, and decides whether a cross-review would improve them.
If so, agents review each other's plans and revise their own. Jev then selects a finaliser to merge the strongest ideas, unless one reviewed plan already stands alone. A fast mode has Jev judge drafts as they arrive and accept the first good one. The README spells out agent-call counts per mode: with N agents, ultra makes 2N+1 calls plus N if Jev asks for another review, balanced as few as N+1, and fast N or N+1. Each Jev evaluation is a separate TypeSafe call.
What the decision model decides
| Type | Question | Answers |
|---|---|---|
| Score | How complete, feasible and risk-aware is this plan? | — |
| Yes / no | Would a cross-review improve these plans? | yes · no |
| Choice | Which agent should finalise the plan, or does one plan stand alone? | — |
How it works
- Agents draft plans independently from the repository
- Jev scores each draft
- Jev decides whether to run a cross-review
- Agents review and revise if requested
- Jev picks a finaliser, which merges the best ideas into one plan
Strengths
- Disagreement built in before implementation
- Supports CLI agents and chat APIs
- Modes trade cost for thoroughness
Caveats
- Needs credentials for at least two agents plus TypeSafe
- Plan quality still depends on the drafting agents
Getting started
npm install -g jev-planner, export TYPESAFE_API_KEY, jev-planner doctor, then jev-planner "Add per-user rate limiting to the public API"At a glance
- Type
- Project
- Status
- beta
- Language
- TypeScript
- Runtime
- Node CLI
- Has code
- yes
- License
- MIT
Use cases and integrations
How curated lists describe it
With \*\*N\*\* agents, \`--mode ultra\` makes \*\*2N + 1\*\* agent calls: drafts, reviews, and final synthesis, plus \*\*N\*\* if Jev requests another review. The default \`balanced\` makes as few as \*\*N + 1\*\* and never more than \`ultra\`; \`fast\` makes \*\*N\*\*, or \*\*N + 1\*\* when it merges. Each Jev evaluation is a separate TypeSafe call.
logicrw/awesome-jev-projects
Frequently asked questions
What does Jev decide?
Plan scores, whether a cross-review is worthwhile, and which agent finalises.
Which agents are supported?
Codex and Claude Code CLIs, plus DeepSeek, Kimi and GLM chat APIs.
How many agent calls does it make?
Depends on mode: from N in fast mode to 2N+1 (plus N if Jev requests another review) in ultra.
Related projects
Recursive language model layer for coding agents that keeps inputs in a local evaluator, with optional Jev judge_many for typed calls.
ProjectbetaRustAgent plugin that asks Jev whether the session is at a safe boundary to /compact and can trigger it on Pi and Claude Code.
PluginbetaTypeScriptCoding-agent hook where Jev decides whether a mid-turn message should steer the running task, be queued, or interrupt it.
PluginbetaPythonOne skill and MCP server for Claude Code, Codex and Hermes where Jev gates large-model turns, routes models and guards Git steps.
MCP serverbetaPythonNode experiment where Jev picks the next step around a coding agent, with approved actions and Codex or Claude Code delegation.
ProjectprototypeTypeScriptClaude Code skill that splits work into PBIs for Codex, Antigravity or Claude sub-agents, using Jev for model choice and triage.
ProjectprototypePython