Kaguya: A Modular AI Agent Framework Built on LangGraph, Making Intelligent Assistant Development Simple and Intuitive

Kaguya is an open-source AI Agent framework built on LangGraph and OpenAI, designed to lower the barrier to intelligent assistant development. It addresses challenges developers face when building Agents, such as state management, tool calling, and scalability. With its clean and clear code structure, detailed Chinese comments, and flexible tool extension mechanism, it provides developers with an entry-level platform and a scalable infrastructure.

With the improvement of LLM capabilities, AI Agents have become a development hotspot. However, developers face many challenges when building fully functional Agent systems: How to manage dialogue states? How to design tool calling mechanisms? How to ensure scalability and maintainability? The Kaguya project was born to solve these problems. As an open-source framework based on LangGraph, it lowers the learning and usage threshold with clean code and Chinese comments.

Kaguya's core features include: LangGraph-based Agent flow (clear and controllable dialogue flow), OpenAI dialogue reasoning, custom tool extension, simple state management (retaining dialogue history), and detailed Chinese comments. The architecture adopts a modular design with clear organization of core files: agent/graph.py defines the workflow diagram, agent/state.py manages dialogue states, tools/custom_tools.py is for tool extension, etc. Each component has clear responsibilities, facilitating development and maintenance.

Tools are key to extending an Agent's capabilities. Kaguya defines tools using the @tool decorator, which requires including the function name, type annotations, and docstring (functionality, parameters, examples). For example, the temperature conversion tool celsius_to_fahrenheit. After definition, register it in the get_tools() function, and the LLM will autonomously recognize and call the tool to handle user questions.

Kaguya supports flexible model configuration: it uses the OpenAI GPT model by default, and can switch to GPT-3.5 Turbo (low cost) or GPT-4 Turbo (complex reasoning); the temperature parameter adjusts output randomness (low temperature 0.0-0.3 is suitable for factual calculations, high temperature 0.7-1.0 for creative scenarios). Quick start steps: Clone the repository → Install dependencies → Configure OpenAI API key → Run main.py; To add custom tools, write them in tools/custom_tools.py and register.

Application scenarios include intelligent customer service systems, personal assistants, educational tutoring tools, data analysis assistants, creative writing partners, etc. Compared to other frameworks, Kaguya's advantages are: clean code structure (no over-abstraction), automatic tool calling (LLM makes autonomous decisions), complete dialogue history, high scalability, detailed Chinese comments, and production readiness (basic error handling).

As a lightweight framework, Kaguya has room for improvement: incomplete logging (difficult to track in production environments), insufficient unit test coverage, performance optimization (need to introduce caching to avoid repeated LLM calls), multi-language support (currently mainly for Chinese), and multi-Agent collaboration (current single-Agent architecture).

Kaguya builds a fully functional AI Agent system with clean code, focusing on providing a clear and usable infrastructure, suitable for entry-level developers or as a starting point for lightweight projects. With the development of the LangGraph ecosystem and the improvement of LLM capabilities, intelligent applications of such frameworks will demonstrate value in more fields, and its concise design concept also provides a reference for building easy-to-use and maintainable AI systems.