small-rag 0.1.2

A minimal, dependency-free Retrieval-Augmented Generation (RAG) engine using SQLite as a local vector store.

small-rag

A lightweight, dependency-free Retrieval-Augmented Generation (RAG) engine for Python.

small-rag is a minimal, easy-to-understand vector store + RAG toolkit.
It is designed for developers who want a simple, transparent, hackable RAG library that can be fully understood in minutes.

No FAISS, no Qdrant, no heavy dependencies — just pure Python + SQLite.

---

Features

• Pure Python implementation (no external dependencies)
• SQLite-backed vector store
• Pluggable embedding function (OpenAI, HF models, local LLMs, custom functions)
• Text preprocessing + chunking
• Metadata filtering
• Import/export database to JSON
• Command-line interface (smallrag-cli)
• Easy to read, extend, and embed inside other projects

---

Installation

pip install small-rag

Verify:

python -m smallrag --help

---

Quick Start (Python)

1. Create a RAG instance

from smallrag import SmallRAG

rag = SmallRAG("my.db")

2. Set an embedder (required)

rag.set_embedder(lambda t: [float(ord(c) % 255) for c in t[:64]])

3. Add documents

rag.add_document("Python is a programming language created by Guido van Rossum.")
rag.add_document("Retrieval augmented generation improves LLM accuracy.")

4. Query

res = rag.query("What is RAG?")
print(res)

Example output:

[
  {
    "id": 2,
    "text": "retrieval augmented generation improves llm accuracy.",
    "metadata": {},
    "score": 0.87
  }
]

---

Realistic Example (HuggingFace Embeddings)

from smallrag import SmallRAG
from sentence_transformers import SentenceTransformer

model = SentenceTransformer("all-MiniLM-L6-v2")
rag = SmallRAG("docs.db")

rag.set_embedder(lambda t: model.encode(t).tolist())

rag.add_document("Paris is the capital of France.", metadata={"topic": "geography"})
rag.add_document("The Eiffel Tower is in Paris.", metadata={"topic": "landmark"})

print(rag.query("Where is the Eiffel Tower?"))

---

CLI Usage

Add a document

smallrag-cli --db my.db add "Python is easy to learn"

With metadata:

smallrag-cli --db my.db add "Paris is beautiful" --meta '{"country": "France"}'

Query

smallrag-cli --db my.db query "What is Python?"

Export DB

smallrag-cli --db my.db export dump.json

Import DB

smallrag-cli --db my.db import dump.json

---

Filtering by Metadata

rag.query("Paris", filter={"country": "France"})

---

Chunking Behavior

Default:

• chunk size: 512
• overlap: 64

rag.add_document(big_text, chunk=True, chunk_size=256)

---

Using OpenAI Embeddings

from smallrag import SmallRAG
from openai import OpenAI

client = OpenAI()

rag = SmallRAG("openai.db")

def openai_embed(t: str):
    emb = client.embeddings.create(
        model="text-embedding-3-small",
        input=t
    )
    return emb.data[0].embedding

rag.set_embedder(openai_embed)
rag.add_document("The Amazon rainforest is the largest rainforest on Earth.")

print(rag.query("What is the largest rainforest?"))

---

Limitations (Intentional)

• Linear search (no ANN index)
• Not optimized for millions of vectors
• Retrieval only — no built-in LLM generation
• Minimalistic by design

---

Contributing

Pull requests welcome.
If you use small-rag, share feedback via issues.

---

License

MIT License.