Getting Started

Build your first RAG pipeline locally in a few minutes.

Getting Started

This guide helps you go from zero setup to a working local RAG pipeline in just a few minutes.

RAGLight is designed to be simple, explicit, and fast to experiment with. You can start either from the CLI or directly from Python.

---

Prerequisites

Before getting started, make sure you have:

• Python 3.9+

• A local or remote LLM provider

  • Recommended for local use: Ollama
  • Alternatives: LMStudio, vLLM, OpenAI, Mistral

Using Ollama (recommended)

Install Ollama and pull a model:

ollama pull llama3

Make sure Ollama is running:

ollama serve

---

Installation

Install RAGLight from PyPI:

pip install raglight

That’s it. No additional setup required.

---

Option 1 — Instant RAG with the CLI

The fastest way to get started is using the interactive CLI wizard.

raglight chat

The wizard will guide you through:

• Selecting a local folder containing your documents
• Choosing an embedding model
• Choosing a vector store
• Selecting an LLM provider and model
• Configuring ignore folders (e.g. .venv, node_modules)

Once completed, RAGLight will:

1. Ingest your documents
2. Build the vector store
3. Start an interactive chat session

No Python code required.

---

Option 2 — Your First RAG Pipeline in Python

If you prefer explicit code, here is the minimal Python example.

1. Define your knowledge sources

from raglight.models.data_source_model import FolderSource

knowledge_base = [
    FolderSource(path="./data")  # Folder containing your documents
]

2. Create the RAG pipeline

from raglight.rag.simple_rag_api import RAGPipeline
from raglight.config.settings import Settings

pipeline = RAGPipeline(
    knowledge_base=knowledge_base,
    model_name="llama3",
    provider=Settings.OLLAMA,
    k=5
)

3. Build the pipeline

pipeline.build()

This step:

• Parses your documents
• Generates embeddings
• Stores them in the vector database

4. Query your documents

response = pipeline.generate(
    "What is this project about?"
)

print(response)

You now have a fully working local RAG system.

---

What happens under the hood?

RAGLight keeps everything explicit:

1. Documents are ingested from your data sources
2. Embeddings are generated using the selected model
3. Vectors are stored in a vector database
4. Relevant chunks are retrieved at query time
5. The LLM generates an answer using retrieved context

Nothing is hidden or automatic unless you configure it.

---

Going further

Once you have a basic RAG running, you can explore:

• Agentic RAG for multi-step retrieval and reasoning
• RAT (Retrieval-Augmented Thinking) with reflection loops
• Custom pipelines with the Builder API
• Multimodal document ingestion (PDFs with images)
• MCP integration for tool-augmented agents

Check the next sections of the documentation to dive deeper.

---

Next steps

• Learn how each component works in the Core Concepts section
• Explore ready-to-run examples in the examples/ folder
• Customize your pipeline step by step

RAGLight is built for experimentation — start simple, then iterate.