GitHub Repository Teaches AI Agents from First Principles to Production

Summary

AI Agents: Zero to Hero is a source-available, code-first educational repository by Tanmay Sah that explains how agent systems work without requiring an agent framework. Its central distinction is that an agent is not simply an LLM and a prompt: it combines a model or policy with a runtime control loop, state and memory, tools, and an external environment. The curriculum presents an operational spectrum from raw LLMs and chatbots through RAG pipelines, workflows, tool-using models, iterative autonomous agents, and multi-agent systems. It then builds the Observe-Decide-Act feedback loop, explaining that the model requests or selects an action while an execution layer performs it and returns the result. Modules cover tool schemas and function calling, MCP, first-agent construction, SQLite memory, planning and ReAct, context engineering, runtime budgets, multi-agent handoffs, failure handling, trajectory evaluation, safety verification, production durability, coding agents, and self-improvement with rollback. The repository runs on standard Python 3.11+ with no mandatory third-party packages or API keys; deterministic simulators and mock models support offline learning and testing. Real providers, including OpenAI, Anthropic, Gemini, and local Ollama instances, can be added through the modular examples. The repository also includes runnable demonstrations, an 88-test unittest suite, and capstone systems for coding assistance and human-reviewed Reddit automation.