⚡ easy-rag-cli

Zero-config RAG for any codebase or document folder.

Install it, index your project, and start asking questions in plain English — via CLI, browser, or code.

Install

npm install easy-rag-cli
# or globally
npm install -g easy-rag-cli

Quick Start

# 1. Create a config file (optional — works with defaults too)
npx easy-rag-cli init

# 2. Set your API key
export OPENAI_API_KEY=sk-...

# 3. Index your project
npx easy-rag-cli index

# 4. Ask a question
npx easy-rag-cli ask "How does authentication work in this project?"

# 5. Or open the browser UI
npx easy-rag-cli serve

Commands

| Command | Description | |---|---| | easy-rag-cli init | Create easy-rag-cli.config.json | | easy-rag-cli config | Interactive provider/model setup wizard | | easy-rag-cli config --show | Print current configuration | | easy-rag-cli config --set key=value | Set a single config value | | easy-rag-cli index | Scan & embed files into local vector store | | easy-rag-cli index --full | Force full re-index (ignore cache) | | easy-rag-cli ask "question" | Ask a question via CLI | | easy-rag-cli ask -i | Interactive Q&A session | | easy-rag-cli serve | Start web UI at localhost:3141 | | easy-rag-cli serve --port 8080 | Start web UI on custom port | | easy-rag-cli status | Show index stats |

Programmatic API

import { index, ask, askStream, searchChunks } from 'easy-rag-cli';

// Index the codebase
await index({
  onProgress: ({ stage, done, total }) => console.log(`${stage}: ${done}/${total}`)
});

// Ask a question (returns full answer)
const { answer, sources } = await ask('What does the main function do?');
console.log(answer);
console.log('Sources:', sources);

// Stream the answer
for await (const delta of askStream('Explain the folder structure')) {
  process.stdout.write(delta);
}

// Just search for relevant chunks
const chunks = await searchChunks('database connection', { topK: 3 });
chunks.forEach(c => console.log(c.filePath, c.score));

Configuration

easy-rag-cli.config.json:

{
  "provider": "openai",
  "openai": {
    "apiKey": "sk-...",
    "embeddingModel": "text-embedding-3-small",
    "chatModel": "gpt-4o-mini"
  },
  "ollama": {
    "baseUrl": "http://localhost:11434",
    "embeddingModel": "nomic-embed-text",
    "chatModel": "llama3"
  },
  "index": {
    "include": ["**/*.js", "**/*.ts", "**/*.md", "**/*.pdf"],
    "exclude": ["**/node_modules/**", "**/.git/**"],
    "chunkSize": 500,
    "chunkOverlap": 50,
    "maxFileSize": 500000
  },
  "serve": {
    "port": 3141,
    "openBrowser": true
  }
}

Quick config via CLI

# Switch to Ollama
npx easy-rag-cli config --set provider=ollama

# Set OpenAI key
npx easy-rag-cli config --set openai.key=sk-abc123

# Set Ollama URL
npx easy-rag-cli config --set ollama.url=http://localhost:11434

# Set chat model
npx easy-rag-cli config --set ollama.chat=mistral

# View everything
npx easy-rag-cli config --show

Using Ollama (local, free)

{
  "provider": "ollama",
  "ollama": {
    "baseUrl": "http://localhost:11434",
    "embeddingModel": "nomic-embed-text",
    "chatModel": "llama3"
  }
}

Make sure Ollama is running and the models are pulled:

ollama pull nomic-embed-text
ollama pull llama3

What gets indexed?

By default, easy-rag-cli indexes:

• Source code: .js, .ts, .jsx, .tsx, .py, .go, .rs, .java, .c, .cpp
• Docs: .md, .txt, .pdf
• Config: .json, .yaml, .yml, .html, .css, .sh

Files in node_modules, .git, dist, build are skipped automatically.

How it works

1. Scan — globs your project for matching files
2. Chunk — splits code at function/class boundaries, docs at paragraph boundaries
3. Embed — generates vector embeddings via OpenAI or Ollama
4. Index — builds a hybrid BM25 + vector store with symbol graph locally (no external DB)
5. Search — fuses vector + keyword results via Reciprocal Rank Fusion, expands with import graph
6. Answer — passes top chunks as context to the LLM and streams the answer

License

MIT