ragkit-sdk 0.1.1

Universal RAG framework SDK for Python developers

RAGKit

Universal RAG (Retrieval-Augmented Generation) framework SDK for Python.

Build AI-powered assistants, document search, and FAQ systems in minutes — without rewriting the pipeline for every project.

---

What is RAGKit?

RAGKit is an open-source Python SDK that gives you a complete Retrieval-Augmented Generation pipeline in a few lines of code.

You plug in your data sources, RAGKit handles the rest: chunking, embedding, vector storage, retrieval, and LLM generation via OpenRouter.

Your Data → RAGKit → Answer

---

Quick Start

pip install ragkit-sdk
rag init

Add your API key to .env:

OPENROUTER_API_KEY=your-key-here

Build your first agent:

from ragkit import Agent, MarkdownSource

agent = Agent()
agent.use(MarkdownSource("./docs"))
agent.index()

print(agent.ask("How does authentication work?"))

That's it. No vector database setup. No embedding model download configuration. No prompt engineering.

---

Features

• 6 built-in sources — Markdown, TXT, JSON, CSV, REST API, Directory
• Automatic pipeline — load → clean → chunk → embed → index
• Semantic search — Chroma or Qdrant vector store
• 100+ LLMs — via OpenRouter (GPT-4o, Claude, Gemini, Llama…)
• Streaming — async token-by-token output
• Cache — embeddings and query results cached on disk
• Hooks — intercept before/after generate and response
• Plugin system — pip install ragkit-pdf to add PDF support
• CLI — rag init, rag index, rag ask, rag dev, rag serve

---

Installation

pip install ragkit-sdk

Optional extras:

pip install ragkit-sdk[serve]    # HTTP server (FastAPI + uvicorn)
pip install ragkit-sdk[qdrant]   # Qdrant vector store

---

Usage Examples

Multiple sources

from ragkit import Agent, MarkdownSource, CsvSource, RestSource

agent = Agent()
agent.use(MarkdownSource("./docs"))
agent.use(CsvSource("./faq.csv", content_column="answer"))
agent.use(RestSource("https://api.example.com/kb", content_key="body"))

agent.index()
print(agent.ask("What is the return policy?"))

Streaming

import asyncio

async def main():
    agent = Agent()
    agent.use(MarkdownSource("./docs"))
    agent.index()

    async for token in agent.stream("Summarize the onboarding guide"):
        print(token, end="", flush=True)

asyncio.run(main())

Custom model

from ragkit import Agent
from ragkit.models import OpenRouterProvider

agent = Agent()
agent.set_model(OpenRouterProvider(
    model="anthropic/claude-sonnet-4-5",
    temperature=0.3,
))

Hooks

@agent.on("before_generate")
def log_prompt(prompt: str) -> None:
    print(f"Sending {len(prompt)} chars to LLM")

@agent.on("after_response")
def save_answer(response: str) -> None:
    with open("answers.log", "a") as f:
        f.write(response + "\n")

Extend with a custom source

from ragkit.core.base import SourceAdapter, Document

class NotionSource(SourceAdapter):
    def load(self) -> list[Document]:
        # fetch from Notion API
        ...

agent.use(NotionSource())

---

CLI

rag init                        # Initialize a new project (rag.yaml + .env)
rag index                       # Index all configured sources
rag sync                        # Sync sources and re-index
rag ask "What is X?"            # One-shot question
rag dev                         # Interactive REPL
rag serve --host 0.0.0.0        # HTTP API on /ask

---

Documentation

Full documentation: hustlerlabs.github.io/ragkit

• Quick Start
• Configuration reference
• Sources guide
• API reference
• Plugin development

---

Roadmap

Version	Features
V0.1 ✅	Core pipeline, 6 sources, Chroma, OpenRouter, CLI, cache, hooks, plugins
V0.2	Hybrid search (BM25 + vector), streaming CLI, plugin PDF
V0.3	PostgreSQL, MongoDB sources
V0.5	Observability (cost tracking, latency dashboard)
V1.0	Stable API, full docs site

---

Contributing

Contributions are welcome. See CONTRIBUTING for guidelines.

git clone https://github.com/HustlerLabs/ragkit
cd ragkit
pip install -e ".[dev]"
pytest tests/

---

License

MIT — see LICENSE.