TODO

A simple Todo app, built to learn LangChain and LangGraph. Starts as a plain FastAPI + MySQL CRUD app with a vanilla HTML/CSS/JS frontend, then extended with a tool-calling LLM agent (Ollama, llama3.1:8b) that can manage todos through natural language.

Architecture (before LangChain+LangGraph)

Frontend (HTML/CSS/JS)  -->  Backend (FastAPI)  -->  Database (MySQL)

Plain REST CRUD: the frontend calls GET /todos, POST /todo, DELETE /todo/:id, DELETE /todos; the backend does plain CRUD against MySQL via SQLAlchemy.

Architecture (current)

Frontend (HTML/CSS/JS)  -->  Backend (FastAPI)  -->  Database (MySQL)
                                      ‖
                                    Agent
                          (Ollama · llama3.1:8b, tool-calling)

The original REST stack is unchanged. A separate agent (backend/agents/agent.py) shares the same CRUD tool layer (backend/app/crud.py) to manage todos through natural-language chat, with multi-turn conversation memory. The agent currently runs standalone (a CLI chat loop) and is not yet wired into the FastAPI backend.

Future Plan

Next 2 steps:

1. Rebuild the hand-rolled agent loop using LangGraph (StateGraph, nodes/edges, ToolNode).
2. Wire the agent into the FastAPI backend as a real endpoint (e.g. POST /chat), so the frontend can talk to it directly.

TBD