Lyzr ADK — RAG Cookbooks

A collection of hands-on Python cookbooks demonstrating Retrieval-Augmented Generation (RAG) pipelines using the Lyzr ADK SDK. Each cookbook is a standalone script that progressively introduces more advanced capabilities — from basic vector search to full end-to-end pipelines with knowledge graphs and persistent memory.

Cookbooks

#	File	Description
01	01_basic_rag.py	Basic RAG Pipeline — Knowledge base creation, document ingestion (PDF, DOCX, TXT, website), and querying with multiple retrieval strategies
02	02_agentic_rag.py	Agentic RAG — AI agents with single/multi-KB configurations, custom retrieval settings, multi-turn sessions, and streaming
03	03_knowledge_graph.py	Knowledge Graph (Neo4j) — Entity/relationship extraction, schema-guided ingestion, and graph-powered agent answers
04	04_multimodal_parsing.py	Multi-Modal Parsing — Side-by-side comparison of all 3 PDF parsers, chunking strategies, and parser selection guide
05	05_cognis_memory.py	Cognis Memory — Persistent semantic memory across sessions, scoped retrieval, memory management, and async patterns
06	06_full_pipeline.py	Full End-to-End Pipeline — Combines all capabilities: multi-modal parsing, knowledge graph, agentic RAG, Cognis memory, and streaming

Architecture

                        ┌──────────────┐
                        │  Lyzr Studio │
                        │   (Cloud)    │
                        └──────┬───────┘
                               │ LYZR_API_KEY
          ┌────────────────────┼────────────────────┐
          │                    │                     │
   ┌──────▼──────┐     ┌──────▼──────┐      ┌──────▼──────┐
   │  Knowledge  │     │    Agent    │      │   Cognis    │
   │    Base     │     │   Engine    │      │   Memory    │
   └──────┬──────┘     └──────┬──────┘      └─────────────┘
          │                   │
    ┌─────┼─────┐             │
    │     │     │             │
 ┌──▼─┐ ┌▼──┐ ┌▼────┐  ┌────▼────┐
 │Qdr.│ │Neo│ │Parse│  │Sessions │
 │ant │ │4j │ │ rs  │  │Streaming│
 └────┘ └───┘ └─────┘  └─────────┘

Key Capabilities

Retrieval Strategies: basic · mmr (maximal marginal relevance) · hyde (hypothetical document embeddings) · time_aware

Document Parsers: pymupdf (fast text extraction) · llmsherpa (layout-aware) · lyzr_parse (multi-modal)

Supported Formats: PDF · DOCX · TXT · Websites · Raw text

Vector Stores: Qdrant (default) · Neo4j (knowledge graph)

Prerequisites

• Python 3.8+
• lyzr-adk — pip install lyzr-adk
• LYZR_API_KEY — obtain from Lyzr Studio
• Neo4j Aura — required for cookbooks 03 and 06 (console.neo4j.io)
• Cognis — used in cookbooks 05 and 06 (included in lyzr-adk, no extra setup)

Note: No OPENAI_API_KEY is needed — LLM access is managed through Lyzr Studio.

Quick Start

# Install the SDK
pip install lyzr-adk

# Set your API key
export LYZR_API_KEY="your-api-key"

# Run any cookbook
python 01_basic_rag.py

License

This project is licensed under the MIT License.