Examples Gallery

Check out these runnable examples to learn how to use Maticlib for different use cases. You can find all these scripts in the examples/ directory of the repository.

---

Basic Chat & Multi-Provider

Demonstrates how to initialize and use LLM clients from different providers.

• Basic Sync Chat: A simple script to get text responses from OpenAI, Mistral, and Gemini.
• Focused OpenAI Example: Detailed usage of OpenAI-specific features like reasoning tokens.
• Async Concurrent Chat: Sending requests to multiple models at the same time using asyncio.gather.

---

Graph Workflows

Learn how to build intelligent agents using the MaticGraph engine.

• Sequential Graph: A basic A → B → C pipeline.
• Conditional Routing: Branching logic based on model outputs (e.g., sentiment analysis).
• Parallel Node Execution: Running multiple independent analysis nodes in parallel for maximum performance.

---

Structured Output & Parsing

• Pydantic Response Parsing: Force models to generate validated JSON that maps directly to your Pydantic schemas.
• Typed Graph State: Using Pydantic models to enforce strict state schemas throughout your graph workflows.

---

Native Tool Calling

• Tool Definition & Usage: Use the @tool decorator to let LLMs call your Python functions directly.

RAG & Text2SQL

• Context Engineering Basics: How to extract text from PDFs, Webpages, and TXTs, and chunk them correctly.
• Secure Text2SQL Engine: Auto-reflecting databases and generating strictly validated, injection-proof SQL queries.
• RAG Pipeline & In-Memory Vector Store: A complete end-to-end RAG pipeline using InMemoryVectorIndex and Query Transformers.
• Chroma Vector Store: Using the ChromaVectorIndex for persistent local embedding storage.
• Milvus Vector Store: Using the MilvusVectorIndex (Lite mode) for scalable vector search.
• Milvus Server Vector Store: Connecting to a standalone Milvus Server at http://localhost:19530.

---

Observability & Resilience

• Traces, Retry & Memory: Using the Prompt Registry, rolling window memory buffers, handling external API failures with exponential backoff, and tracing exact token usages via Callbacks.
• Formatting & Evaluation: Converting raw SQL results into Markdown tables and computing context relevance and answer accuracy scores without external dependencies.

---

Running the Examples

1. Set up your environment: Make sure you have your API keys in a .env file.

2. Execute a script:

   # From the project root
   python examples/01_basic_chat.py