easy-rag

RAG in 3 lines of code. Zero dependencies. Load documents, ask questions, get answers.

from easy_rag import EasyRAG

rag = EasyRAG("./my_docs/")
answer = rag.ask("What does the contract say about termination?")
print(answer)

That's it. No API keys, no vector databases, no configuration.

Why easy-rag?

Installation

pip install easy-rag

Optional extras

pip install easy-rag[pdf]      # PDF support (pdfplumber)
pip install easy-rag[openai]   # OpenAI embeddings
pip install easy-rag[local]    # Local embeddings (sentence-transformers)
pip install easy-rag[all]      # Everything

Quick Start

Search your documents

from easy_rag import EasyRAG

# Load an entire folder of documents
rag = EasyRAG("./contracts/")

# Ask a question
result = rag.ask("What is the payment schedule?")
print(result)
# [contract_v2.pdf | Score: 0.87]
# Payment is due within 30 days of invoice date...

Add documents on the fly

rag = EasyRAG()

# Add individual files
rag.add_document("report.md")
rag.add_document("notes.txt")

# Add raw text directly
rag.add_text("The meeting is scheduled for Friday at 3pm.", source="email")

# Search
results = rag.search("meeting schedule", top_k=3)
for r in results:
    print(f"[{r.score:.2f}] {r.chunk.text[:100]}")

CLI usage

# Search documents from the command line
easy-rag ./docs/ "What are the key findings?"

# Show index statistics
easy-rag ./docs/ --stats

# Customize chunk size
easy-rag ./docs/ "query" --chunk-size 1000 --top-k 5

Check index stats

rag = EasyRAG("./docs/")
print(rag.stats())
# {
#   'num_chunks': 142,
#   'num_sources': 12,
#   'num_unique_terms': 3847,
#   'chunk_size': 500,
#   'overlap': 50
# }

Supported File Formats

How It Works

1. Load — Reads all supported files from a directory
2. Chunk — Splits text into overlapping pieces at smart boundaries (paragraphs, sentences)
3. Index — Builds a TF-IDF index using only Python's standard library
4. Search — Finds the most relevant chunks using cosine similarity
5. Return — Gives you the best matching text with source and score

No neural networks, no API calls, no GPU needed. Just fast, reliable text search.

API Reference

EasyRAG(path=None, chunk_size=500, overlap=50)

Create a new RAG index.

• path — Directory or file to load (optional, can add later)
• chunk_size — Max characters per chunk (default: 500)
• overlap — Character overlap between chunks (default: 50)

.ask(question, top_k=3) -> RAGResult

Ask a question and get a formatted answer with sources.

.search(query, top_k=5) -> List[SearchResult]

Search for relevant chunks (raw results).

.add_document(filepath) -> int

Add a single file to the index. Returns number of chunks created.

.add_text(text, source="inline") -> int

Add raw text to the index. Returns number of chunks created.

.stats() -> dict

Get index statistics.

License

MIT