agents-from-scratch and ai-agents-from-scratch
These two tools are ecosystem siblings, with "ai-agents-from-scratch" being a more comprehensive and updated version of "agents-from-scratch," incorporating advanced concepts like function calling, memory, and ReAct patterns, while both share the educational goal of building AI agents from first principles using local LLMs.
About agents-from-scratch
pguso/agents-from-scratch
Build AI agents from first principles using a local LLM - no frameworks, no cloud APIs, no hidden reasoning.
Structured around 12 progressive lessons, it teaches agent fundamentals by evolving a single `Agent` class through observable steps: routing logic, tool integration, observe-decide-act loops, memory management, atomic action execution, and dependency graphs (Atom of Thought). Built in Python with local GGUF models, it includes evaluation frameworks for regression testing and telemetry for runtime observability—emphasizing explicit constraints and state management over prompt engineering.
About ai-agents-from-scratch
pguso/ai-agents-from-scratch
Demystify AI agents by building them yourself. Local LLMs, no black boxes, real understanding of function calling, memory, and ReAct patterns.
Implements core agent patterns with local LLMs through **node-llama-cpp** and **JSON Schema tool definitions**, progressing through function calling, iterative ReAct loops, and plan-based reasoning without external dependencies. The curriculum separates concept explanations from code walkthroughs, enabling learners to understand architecture before exploring implementations. Targets JavaScript/Node.js developers seeking hands-on mastery of agent foundations prior to adopting production frameworks like LangChain.
Scores updated daily from GitHub, PyPI, and npm data. How scores work