دانلود کتاب ساخت یک سیستم چندعاملی

Multi-agent systems and the LLM agents that comprise them are some of the most discussed and worked on topics in AI today. Everyday tasks like searching the web, as well as more complex tasks like building entire codebases of software applications, are some examples where LLM agents have already been deployed. While they are by no means perfect systems, many recognize the great potential for agent-to-agent interactions to reshape the way many tasks are done in our society today. There are more than a handful of LLM agent frameworks that exist today, which I have used extensively and even contributed to building. In fact, assuming some prior familiarity with LLMs and having past experiences programming with Python or JavaScript/TypeScript, you can probably already create LLM agents as well as multi-agent systems and have them autonomously perform tasks with any one of these frameworks. This book, however, employs a hands-on approach to help you gain a deeper understanding of the inner workings of a multi-agent system and the LLM agents that comprise them by having you build these from scratch. This is not so different from how you might deepen your knowledge in LLMs by learning how to implement attention and transformers yourself. To build a multi-agent system from scratch, our journey starts with the building of a foundational LLM agent. We’ll also incorporate a few significant enhancements to it, such as making it MCP-ready. The vast network of tools and resources made available through MCP increases the potential of LLM agents that can leverage them. It also helps to offload some of the responsibilities for developing robust tooling for LLM agents to external teams and organizations. We’ll also consider how to implement human-in-the-loop capabilities and memory for LLM agents, before finally taking on the step of assembling multi-agent systems through the Agent2Agent protocol. We’ll package all our code, which includes the required infrastructure, such as interfaces for tools and LLMs, in the book’s very own LLM agent framework that you’ll get to develop for yourself. This framework is primarily designed for educational purposes, rather than being deployed in production settings. Nevertheless, it will give you the foundation to work more confidently and effectively with any other LLM agent and multi-agent frameworks of your choosing or even to develop your own specialized solutions.