Hello Agents · Build AI Agents from Scratch

Hello Agents is an educational tutorial project that provides a free, open-source guide on building AI agents from scratch. The content covers agent principles, the history of agents, and LLM fundamentals, followed by practical instruction on no-code platforms, framework development, and building a custom agent framework using the OpenAI native API. Advanced sections address memory and retrieval, context engineering, agent communication protocols, agentic reinforcement learning, and agent evaluation.

The tutorial includes case studies such as a Smart Travel Assistant and a Cyber Town simulation, along with interview preparation material for careers in AI. It is especially suited for AI developers, software engineers, and students with a basic programming background, as well as self-learners with a strong interest in cutting-edge AI.

All content is completely free and open source, allowing users to grow together with the community without any cost. The course balances theory and practice, helping learners master the design and development of single-agent to multi-agent systems end-to-end.

The curriculum covers building your own Agent framework from scratch using the OpenAI native API and mastering Agentic RL, including the full pipeline from SFT to GRPO for LLM training. It also details context engineering, memory systems, RAG, vector storage, and core metrics for agent evaluation.

The tutorial provides step-by-step instructions for context understanding for continuous interaction and building a Smart Travel Assistant that utilizes MCP and multi-agent collaboration in practice. It also covers combining agents with games to simulate social dynamics in the Cyber Town Simulation.

The course covers no-code platforms like Coze, Dify, and n8n, as well as framework development tools such as AutoGen, AgentScope, and LangGraph. It also includes hands-on practice with classic agent paradigms like ReAct, Plan-and-Solve, and Reflection.

The curriculum includes a dedicated section on model training, covering the full pipeline from supervised fine-tuning (SFT) to Group Relative Policy Optimization (GRPO) for LLM training. It also addresses agentic reinforcement learning as a key advanced skill.

The course is designed to help users build a solid foundation by deep diving into agent concepts, history, and classic paradigms before moving to practical application. It also prepares learners for careers in AI through specific interview preparation material.

The tutorial includes a section on agent communication protocols, analyzing MCP, A2A, and ANP protocols. It also covers memory systems, RAG, and vector storage for handling memory and retrieval tasks.

The course covers the history of agents, tracing the evolution from symbolic AI to LLM-powered agents, and provides an introduction to agents, types, paradigms, and applications. It also includes LLM fundamentals covering transformers, prompting, major LLMs, and limitations.

The tutorial includes a capstone project where users build their own complete multi-agent application, integrating all previously learned concepts. It also features an automated deep research agent reproduction and analysis.

The course is written in English and is available on the website helloagents.online, where users can follow the author on X and star the project on GitHub. It is an open-source initiative that encourages community growth.

The course covers context engineering for continuous interaction and provides a comprehensive guide on building a custom agent framework using the OpenAI native API. It also includes a section on building a Smart Travel Assistant as a practical example.

The course includes a section on agent evaluation, covering core metrics, benchmarks, and evaluation frameworks to ensure the developed agents meet performance standards. It also addresses the limitations of major LLMs as part of the fundamentals.

The official website of Hello Agents · Build AI Agents from Scratch is helloagents.online.