Build a deterministic MockLLM, a small tool interface, a calculator tool, and a one-step agent loop.
Add multiple tools, tool routing, input validation, and clearer action schemas.
Introduce thought-action-observation loops and multi-step reasoning traces.
Add short-term memory, simple long-term memory, and memory inspection examples.
Separate high-level planning from execution and track task progress.
Create a small multi-agent system with simple roles, message passing, and coordination rules.
Add deterministic task cases, scoring functions, failure analysis, and result reports.
Introduce a real LLM API behind a clean interface after the from-scratch loop is understood.
Add simple document retrieval with local files and lightweight indexing.
Compare the from-scratch implementation with LangChain, LangGraph, AutoGen, and CrewAI conceptually and experimentally.