EasyRAG

RAG in 10 lines. No PhD required.

A dead-simple Python library and CLI that lets you build Retrieval-Augmented Generation pipelines in under 10 lines of code. Chat with your documents, codebase, or any text data.

---

Get Started in 30 Seconds

pip install easyrag

from easyrag import RAG

rag = RAG()
rag.ingest("./docs")
answer = rag.ask("How does authentication work?")
print(answer)

That's it. Five lines. No API keys. Runs locally.

---

Use Case: Onboard Developers Instantly

Problem: A new developer joins your team. Your project has 200+ pages of docs, runbooks, API specs, and inline code comments spread across dozens of files. They spend 2 weeks just reading before they can contribute.

Solution: Feed everything into EasyRAG. Now they ask questions and get instant, cited answers.

from easyrag import RAG

# One-time setup (takes ~60 seconds)
rag = RAG(llm="ollama:llama3")
rag.ingest("./docs")          # Markdown, PDFs, runbooks
rag.ingest("./src")           # Source code with comments

# New developer asks questions on day one
result = rag.ask("How do I deploy to production?")
print(result.answer)
# => "Merge to main (auto-deploys to staging), verify on staging,
#     then trigger the 'Deploy to Production' GitHub Actions workflow.
#     Requires 2 approvals. ECS auto-rolls back within 5 min on failure. [1][2]"

# Every answer cites its sources
for src in result.sources:
    print(f"  - {src.metadata['file_name']} (score: {src.score:.3f})")
#  - deployment.md (score: 0.575)
#  - ci_cd.md (score: 0.545)

Other use cases: internal Q&A bots, support knowledge bases, research paper chat, codebase exploration.

---

CLI

easyrag ingest ./docs                         # Ingest documents
easyrag ask "What is the retry policy?"       # Ask a question
easyrag chat                                  # Interactive chat
easyrag search "authentication flow"          # Semantic search (no LLM)
easyrag serve --port 8000                     # Start API server

---

Why EasyRAG?

Feature	EasyRAG	LangChain	LlamaIndex
Lines to first RAG	5	50+	30+
Zero-config setup	Yes	No	No
Built-in vector store	Yes	No	No
Local-first (no API keys)	Yes	No	No
CLI included	Yes	No	No
API server included	Yes	No	No

---

Supported Formats

Category	Formats
Documents	PDF, Markdown, Plain text, CSV, JSON
Code	Python, JavaScript, TypeScript, Go, Rust, Java, C/C++, Ruby, PHP, and more
Config	YAML, TOML, XML, HTML, INI

---

Configuration

from easyrag import RAG

# Minimal -- everything has sensible defaults
rag = RAG()

# Full control
rag = RAG(
    embedding_model="openai:text-embedding-3-small",
    llm="openai:gpt-4o-mini",
    chunk_size=512,
    chunk_overlap=50,
    top_k=5,
    db_path="./my_vectors.db",
    rerank=True,
)

Embedding Models

Provider	Model	API Key
Local (default)	all-MiniLM-L6-v2	No
OpenAI	text-embedding-3-small, text-embedding-3-large	Yes

LLM Providers

Provider	Models	API Key
Ollama (default)	llama3, mistral, phi3, etc.	No
OpenAI	gpt-4o, gpt-4o-mini	Yes
Anthropic	claude-sonnet-4-20250514, claude-haiku-4-5-20251001	Yes

Vector Stores

Store	Best For	Dependencies
NumPy (default)	Most use cases, portable, single file	None
ChromaDB	Large datasets, production	pip install easyrag[chroma]

---

Streaming

for token in rag.ask_stream("What is the retry policy?"):
    print(token, end="", flush=True)

---

API Server

easyrag serve --port 8000

Method	Path	Description
GET	/health	Health check
GET	/info	Database info
POST	/ingest	Ingest documents
POST	/ask	Ask a question
POST	/search	Semantic search
WS	/ws/ask	Streaming answers via WebSocket

curl -X POST http://localhost:8000/ask \
  -H "Content-Type: application/json" \
  -d '{"question": "How does auth work?", "top_k": 5}'

---

Docker

docker build -t easyrag .
docker run -p 8000:8000 -v ./data:/data easyrag

---

Development

git clone https://github.com/aymenhmaidiwastaken/easyrag.git
cd easyrag
pip install -e ".[dev,all]"
pytest

---

License

MIT