J-Star Code Reviewer

Local-first, context-aware AI code reviewer powered by LlamaIndex + Groq.

Works with any language — TypeScript, Python, Rust, Go, you name it.

✨ Features

• Local Vector Index — Embeddings stored locally, no external DB
• Gemini Embeddings — Free tier friendly, no OpenAI key needed
• Chunked Reviews — Handles large diffs without rate limits
• Detective Engine — Deterministic checks for common issues
• Dashboard Output — Professional review reports with fix prompts
• Global CLI — Install once, use in any project

🚀 Quick Install

Option 1: Global CLI (Recommended)

# Install globally
npm install -g jstar-reviewer

# In any project directory:
jstar setup      # Create config files
jstar init       # Index the codebase
jstar review     # Review staged changes

Option 2: One-Curl (Adds to current project)

curl -fsSL https://raw.githubusercontent.com/JStaRFilms/jstar-code-review/v2.0.0/setup.js | node

After Install:

1. Check Config: The tool now auto-creates .env.example and .jstar/ when you run it.
2. Add Keys: Copy .env.example → .env.local and add your GEMINI_API_KEY and GROQ_API_KEY.
3. Index: Run jstar init (or pnpm run index:init) to build the brain.
4. Review: Stage changes (git add) and run jstar review (or pnpm run review).

For a detailed walkthrough, see ONBOARDING.md.

git diff --staged
       │
       ▼
┌──────────────────┐
│  Detective       │  ← Static analysis (secrets, console.log, "use client")
│  Engine          │
└────────┬─────────┘
         │
         ▼
┌──────────────────┐
│  Local Brain     │  ← Gemini embeddings via LlamaIndex
│  (Retrieval)     │
└────────┬─────────┘
         │
         ▼
┌──────────────────┐
│  Chunked Review  │  ← Splits diff by file, delays between calls
│  Queue           │
└────────┬─────────┘
         │
         ▼
┌──────────────────┐
│  Groq LLM        │  ← moonshotai/kimi-k2-instruct-0905
│  (The Judge)     │
└────────┬─────────┘
         │
         ▼
   📝 Review Report

🚀 Quick Start

1. Install Dependencies

pnpm install

2. Set Environment Variables

Create .env.local:

GEMINI_API_KEY=your_gemini_key
GROQ_API_KEY=your_groq_key

3. Index Your Codebase

pnpm run index:init

4. Review Staged Changes

git add <files>
pnpm run review

📁 Project Structure

scripts/
├── indexer.ts          # Scans codebase, builds vector index
├── reviewer.ts         # Orchestrates review pipeline
├── detective.ts        # Static analysis engine
├── gemini-embedding.ts # Google Gemini adapter
└── mock-llm.ts         # LlamaIndex compatibility stub

.jstar/
└── storage/            # Persisted embeddings (gitignored)

docs/features/
├── architecture-v2.md  # Full architecture docs
├── detective.md        # Static analysis rules
├── analyst.md          # LLM reviewer (The Judge)
└── ...

⚙️ Configuration

Edit scripts/reviewer.ts:

const MODEL_NAME = "moonshotai/kimi-k2-instruct-0905";
const MAX_TOKENS_PER_REQUEST = 8000;
const DELAY_BETWEEN_CHUNKS_MS = 2000;

📚 Documentation

• Architecture v2
• Detective Engine
• Token Budget
• Chunked Reviews

Built with ⚡ by J Star Studios