star the repo

Provides a hierarchically-organized collection of 30+ production-ready and educational LLM application templates spanning seven architectural categories (starter agents, advanced single agents, multi-agent systems, RAG tutorials, MCP agents, voice agents, and memory-augmented apps). Templates are organized by complexity level (beginner to expert) and include complete working implementations with dependencies, configuration examples, and framework-specific patterns, enabling developers to clone, customize, and deploy reference architectures without building from scratch.

Demonstrates implementation patterns across three major agent frameworks (Agno, LangChain/LangGraph, and MCP) with explicit code examples showing how the same architectural goal (e.g., multi-agent coordination, RAG integration) is achieved differently in each framework. Includes pattern documentation for tool calling, state management, context passing, and agent composition, allowing developers to understand framework trade-offs and migrate between ecosystems.

Demonstrates a production-ready research agent using Google Gemini's Interactions API for advanced reasoning and multi-turn interactions. Shows how to structure research tasks (planning, execution, synthesis), integrate web search and document retrieval, and use Gemini's reasoning capabilities for complex analysis. Enables developers to build sophisticated research and analysis agents that can decompose complex questions into research subtasks.

Provides production-ready implementations of AI agents for investment analysis and financial decision-making. Shows how to integrate financial data APIs (stock prices, company fundamentals, market data), implement financial reasoning patterns, and generate investment recommendations. Demonstrates domain-specific prompting for finance, risk assessment, and portfolio analysis. Enables developers to build financial advisory agents with real-time market data integration.

Demonstrates web scraping agents that combine LLM reasoning with browser automation (Selenium, Playwright) to extract and analyze information from websites. Shows how agents can navigate complex websites, extract structured data, handle dynamic content, and synthesize information across multiple pages. Enables developers to build agents that can autonomously gather information from the web for analysis or monitoring.

Provides implementations of seven distinct RAG patterns (Gemini Agentic RAG, Database Routing RAG, Deepseek Local RAG, Corrective RAG, Hybrid RAG, Cohere RAG Agent, Autonomous RAG with Reasoning) with complete code examples showing retrieval strategy, vector database integration, prompt engineering, and response generation. Each pattern includes architectural diagrams and trade-off analysis, enabling developers to select and implement the RAG approach best suited to their data characteristics and latency requirements.

Demonstrates multi-agent architectures through two production examples: SEO Audit Team (specialized agents for technical SEO, content analysis, backlink analysis coordinating results) and Home Renovation Agent (agents for budgeting, design, contractor coordination). Implementations show agent communication patterns (message passing, shared state, hierarchical coordination), task decomposition, and result aggregation using frameworks like Agno and LangGraph, enabling developers to build team-based AI systems where agents specialize in subtasks.

Demonstrates Model Context Protocol (MCP) integration patterns through three implementations: Travel Planner and GitHub Agents (using MCP servers for external tool access), Notion and Multi-MCP Agents (coordinating multiple MCP servers), and Browser Automation Agent (MCP for browser control). Shows how MCP's server-client architecture enables agents to access external tools and data sources through standardized protocol bindings rather than direct API calls, improving modularity and enabling tool composition.

Provides implementations of voice-enabled agents combining speech-to-text (STT) and text-to-speech (TTS) with LLM reasoning. Templates show integration patterns with speech APIs (Google Cloud Speech, OpenAI Whisper, ElevenLabs TTS), real-time audio streaming, and voice interaction loops. Enables developers to build conversational agents that accept spoken input, process it through LLM reasoning, and respond with synthesized speech.

Demonstrates persistent conversation memory patterns enabling LLM applications to maintain context across multiple interactions. Implementations show memory storage strategies (in-memory, database, vector store), context window management, and conversation history retrieval. Enables developers to build stateful LLM applications where agents remember previous interactions, user preferences, and learned information across sessions.

Provides specialized chat application templates for interacting with specific data sources: GitHub repositories (code search, issue analysis), PDF documents (document Q&A), and YouTube videos (transcript analysis, content summarization). Each template shows data ingestion, indexing, retrieval, and LLM-based response generation specific to that data type. Enables developers to quickly build domain-specific chatbots without implementing data source integration from scratch.

Demonstrates Streamlit-based UI development patterns for LLM applications, showing how to build interactive web interfaces for agents, RAG systems, and multi-agent teams. Templates include chat interfaces, real-time streaming responses, file upload handling, and agent execution visualization. Enables developers to quickly create web UIs for LLM applications without frontend expertise.

Demonstrates integration patterns for multiple vector databases (Pinecone, Weaviate, Chroma, Faiss) in RAG and memory systems. Shows data ingestion, embedding generation, semantic search, and result retrieval with different database backends. Enables developers to choose and integrate vector databases based on deployment constraints (cloud vs. local, managed vs. self-hosted, scale requirements).