陣Jin

LOCAL LLM × MULTI-AGENT CLI

Your next move,
backed by a formation.

Specialized agents modeled after shogi pieces interpret requirements written in natural language. Jin is a fully local CLI for learning and experimenting with the mechanics of agentic development found in tools such as Claude Code.

Get started → View source ↗
  • No API key
  • No cloud uploads
  • MIT License
jin — zsh
$ jin

     ██╗██╗███╗   ██╗
     ██║██║████╗  ██║
     ██║██║██╔██╗ ██║
██   ██║██║██║╚██╗██║
╚█████╔╝██║██║ ╚████║
 ╚════╝ ╚═╝╚═╝  ╚═══╝

Your next move, backed by a formation

● Local LLM detected
● Kin is organizing the request

  Kin ─ command ─┬── Gin UI / UX
                    ├── Hi  Engineering
                    └── Ka  Quality

◆ Formation ready

WHY JIN

Learn the process,
not just the answer.

Jin does not leave everything to one model. Shogi-inspired agents contribute perspectives on requirements, experience, engineering, and quality, while Kin orchestrates the whole formation.

It independently reconstructs concepts seen in Claude Code as of July 2026—including agents, skills, hooks, and project context—for hands-on learning and experimentation.

01

Fully local

Runs with Ollama. Try it in your own environment without sending requirements or code to an external API.

02

Separated by role

Seven pieces cover specialized domains, exposing a workflow built around multiple perspectives and review.

03

Extensible mechanics

Define agents, skills, and hooks as files, then evolve Jin around your own development workflow.

THE FORMATION

Seven perspectives
refine one vision.

Kin chooses the pieces required by each request and guides them through analysis, review, and implementation.

金
KIN · ORCHESTRATOR

Kin

Requirements, formation selection, and synthesis

銀
GIN · EXPERIENCE

Gin

UI / UX and user experience design

飛
HISHA · ENGINEERING

Hisha

Technical implementation and architecture

角
KAKU · QUALITY

Kaku

Quality, risk, and test design

桂
KEIMA · DATA

Keima

Data models, APIs, and metrics

香
KYOSHA · SECURITY

Kyosha

Security, authorization, and permissions

歩
FU · DOCUMENTATION

Fu

Documentation and task organization

HOW IT WORKS

From vision to implementation.

  1. 一
    REQUEST

    Describe the request

    Enter even a rough idea using natural language.

  2. 二
    FORMATION

    Build the formation

    Kin reads the request and selects the specialists it needs.

  3. 三
    REVIEW

    Refine from every angle

    Each piece analyzes the idea, and Kin reviews and combines their proposals.

  4. 四
    IMPLEMENT

    Review and implement

    Promoted pieces generate code and documents for review before changes are applied.

LEARNING TOOL

Not a replacement
for Claude Code.

Jin is neither an official Anthropic or Claude Code project nor a fully compatible implementation. It is an experimental environment for observing agentic development concepts locally and learning by building.

See the supported scope in the README →

GET STARTED

Make your
first move.

All you need is Node.js 24 or later and Ollama.

npm install -g @tanishi-z/jin
jin

To try it without installing, run npx @tanishi-z/jin