🧠 Blueprint RAG Agnostic

A Highly Modular RAG Architecture Blueprint — LLM-agnostic, Vector Database-agnostic, built with LangChain, and powered by a deterministic Template-Driven Auto-Configuration engine.

---

✨ Features

• 🌐 Provider Agnostic — 100% decoupling from specific LLMs (OpenAI, Anthropic, Gemini, Deepseek, Bedrock, Vertex AI).
• 🗄️ Database Agnostic — Swap your Vector Store seamlessly (Chroma, Qdrant, Pinecone, pgvector).
• 🧙‍♂️ Interactive TUI Wizard — Zero-friction configuration. A beautiful terminal UI generates your entire setup dynamically without manual file editing.
• 🛡️ 100% Idempotent — No configuration drift. Re-run the setup 100 times and get the exact same deterministic state.
• 🔒 Secure Injection — Masked input for API keys safely injected straight into your .env file.
• 📚 Rich Documentation — Dedicated, self-contained markdown guides for every architectural decision.

---

🚀 Quick Start

Prerequisites: Python 3.10+

git clone https://github.com/vandre-sales/blueprint-rag-agnostic.git
cd blueprint-rag-agnostic

# Create and activate your virtual environment
python3 -m venv .venv
source .venv/bin/activate

# Install the wizard dependencies
pip install -r requirements.txt

# Run the 1-Click Interactive Setup
python3 bin/init.py

The elegant CLI will ask for your preferred LLM and Database, inject your API Keys safely, and instantly build your architecture using Jinja2 templates.

---

📐 How It Works

flowchart LR
    subgraph INPUT["💻 Developer Input"]
        A1["Run bin/init.py<br/>(TUI Wizard)"]
        A2["Choose LLM &<br/>Vector Store"]
    end

    subgraph ENGINE["⚙️ Auto-Configuration"]
        B1["my-setup.json<br/>Generated"]
        B2["bin/configure.py<br/>Triggered"]
        B3["Jinja2 Templates<br/>Rendered"]
    end

    subgraph OUTPUT["🚀 Output"]
        C1["src/core/<br/>Python Factories"]
        C2[".env<br/>Credentials Injected"]
        C3["Ready for<br/>RAG Execution"]
    end

    A1 --> A2 --> B1 --> B2 --> B3
    B3 --> C1 & C2 --> C3

Loading

The Setup Workflow

Step	Action	Result
1	Run Wizard (python3 bin/init.py)	Interactive terminal prompts for selections
2	Select Providers (LLM, Embeddings, DB)	Deterministic generation of my-setup.json
3	Inject Secrets (API Keys)	Safe population of your .env file
4	Compile Templates (bin/configure.py)	Production-ready factory classes created in src/core/

---

🌟 Philosophy: From Meta-Commenting to Templating

Historically, this project used a Meta-Commenting pattern to avoid opaque abstraction layers, letting you manually uncomment the providers you wanted. We have now evolved this into a fully automated TUI. The project maintains its transparency—generating clean, readable, copy-paste-friendly Python code into src/core/—but completely eliminates manual effort and cross-contamination (Zero Drift).

---

📖 Documentation Index

All extensive documentation, setup instructions, and architectural decisions have been decentralized into detailed markdown files.

🏁 00 - Getting Started

Learn how to run your first agnostic chain after setup.

🧠 01 - LLM Providers

Detailed configuration details for OpenAI, Anthropic, Google Vertex AI, Google AI Studio (Gemini), AWS Bedrock, Deepseek.

🗄️ 02 - Vector Stores

Detailed database guides for Chroma, Qdrant, Pinecone, pgvector.

🔍 03 - RAG Strategies

Swap between Standard Naive RAG, Multi-Query, Hybrid, and Contextual Compression.

🤖 04 - Agent Templates

Take your RAG to the next level with Tool Calling, ReAct, and LangGraph Workflows.

🧩 05 - Meta-Commenting Guide

Explains the historical design pattern used to engineer the templates.

🧪 06 - Dryrun Testing Guide

A reusable framework to validate any new LLM Provider + Vector Store combination.

✅ 07 - Dryrun Checklist

An interactive checklist to ensure no steps are missed. (Now largely automated by bin/init.py).

📋 08 - Automation Audit Report

A formal report summarizing the 100% resilience and idempotency score of our Auto-Configuration Engine.