Awesome AI Agents › Human-in-the-Loop Agents

langchain-ai/agents-from-scratch

⭐ 2219 Jupyter Notebook added to this list on 2025-07-30 repository created 2025-03-31

The project "Agents From Scratch" is a comprehensive guide and implementation repository for building intelligent agents, specifically focusing on an email assistant that integrates human-in-the-loop and memory capabilities. The repository is structured into four main sections, each containing a Jupyter notebook and corresponding source code, which progressively build the agent from basic concepts to advanced features such as evaluation, human-in-the-loop interaction, and memory integration. The email assistant is designed to manage emails by connecting to the Gmail API, enabling functionalities like email triage, response handling, and scheduling through an agent framework. The project begins with an introduction to LangGraph, a framework that supports chat models, tool calling, and agent workflows, providing foundational knowledge necessary for understanding the agent's architecture. The building phase demonstrates how to create the email assistant by combining email triage with agent-driven email responses. Evaluation is conducted using an email dataset and tools like Pytest and LangSmith's evaluate API, allowing for automated assessment of the agent's performance in generating email responses and making triage decisions. A significant feature of the project is the human-in-the-loop (HITL) mechanism, which allows users to review and approve specific tool calls such as sending emails or scheduling meetings, enhancing control and reliability. This is facilitated through integration with the Agent Inbox interface. The memory section introduces long-term memory capabilities using LangGraph Store, enabling the assistant to learn from user feedback and adapt over time to user preferences. The repository also provides detailed instructions for environment setup, including Python version requirements, API key configuration for OpenAI and LangSmith, and package installation. It supports deployment on the LangGraph platform and includes automated testing suites to ensure functionality and quality. Future extensions suggest incorporating LangMem for managing collections of background memories, further enhancing the assistant's contextual understanding and responsiveness. Overall, this project offers a practical, modular, and extensible approach to building sophisticated AI agents for email management, combining state-of-the-art techniques in agent design, evaluation, human interaction, and memory integration.

https://github.com/langchain-ai/agents-from-scratch

agentagent-inboxagentsautomated-testingdeploymentemail-assistantemail-triageevaluationgmail-apihuman-in-the-looplanggraphlanggraph-storelangmemlangsmithmemorypython