ToDo mAIstro: A Personalized AI To-Do List Manager

ToDo mAIstro is a smart chatbot assistant designed to help you manage your tasks and personal information. It uses a sophisticated AI backend to understand your conversations, automatically update your ToDo list, and remember your personal details (like your name, location, and interests) to provide a truly personalized experience.

The entire system is built with a separate Streamlit frontend and FastAPI backend, featuring speech-to-text input and end-to-end AI observability with Langfuse.

✨ Features

• Personalized Memory: Remembers your user profile (name, job, etc.), a detailed ToDo list, and custom preferences for how you like your tasks managed.
• Automatic Updates: Uses Trustcall to intelligently and reliably extract information from your conversation and update your long-term memory.
• Multi-Threaded Chat: Manage multiple, separate chat sessions, each with its own context.
• User Authentication: A simple username-based login system keeps memories and threads separate for each user.
• Voice Input: Speak your commands directly to the app using the built-in speech-to-text recognition.
• AI Observability: Fully integrated with Langfuse to trace and debug every step of the AI's reasoning and memory updates.
• Scalable Architecture: Decoupled frontend and backend allows for independent scaling and development.

🏗️ Architecture

The application operates on a client-server model:

1. Frontend (frontend.py): A Streamlit application provides the user interface. It handles user login, thread management, and captures both text and voice input (using streamlit-webrtc and SpeechRecognition). It then sends user messages to the FastAPI backend.

2. Backend (api.py): A FastAPI server exposes the /chat and /memory/{user_id} endpoints.

3. AI Core (api.py): The backend uses LangGraph to create a stateful graph (a state machine). This graph:

   • Receives the user's message.
   • Loads the user's current Profile, ToDo list, and Instructions from memory.
   • Uses a Google Gemini model (gemini-2.5-pro) to decide if a memory update is needed.
   • Routes to specific nodes (update_profile, update_todos, update_instructions) if an update is required.
   • Uses Trustcall to perform structured, reliable updates to the memory schemas.
   • Generates a natural response for the user.

4. Memory: The InMemoryStore from LangGraph is used to persist user data (profile, todos, instructions) across different sessions, namespaced by user_id.

🛠️ Tech Stack

• Backend: FastAPI, LangGraph, Trustcall
• Frontend: Streamlit, Streamlit-WebRTC
• LLM: Google Gemini (via langchain_google_genai)
• Observability: Langfuse
• Voice: SpeechRecognition, PyAudio, pydub, gTTS
• Dependencies: Pydantic, python-dotenv

---

🚀 Setup and Installation

Follow these steps to get the application running locally.

1. Prerequisites

• Python 3.9+
• A Google API Key (for the Gemini model)
• A Langfuse account (for observability)

2. Clone the Repository

git clone 
replace your environment variables
run bat file