CompanionAI

CompanionAI is a prototype AI system designed to provide friendly companionship, adaptive support, and practical assistance. The project is built with Flask to showcase its features through a simple web application.

---

Key Features

• Fully on-device inference — no cloud dependency, no data exfiltration
• RAG-based long-term memory — personalized knowledge base built from caregiver-provided biography and dynamically updated from conversations
• Short-term memory (STM) — rolling conversation summarization for coherent multi-turn dialogue
• Proactive conversation management — LLM-based intent routing and personalized topic pool
• Event management — natural language scheduling with recurring event support
• Speech I/O — speech-to-text (STT) input and text-to-speech (TTS) output for accessibility
• Caregiver dashboard — password-protected interface for biography management, event entry, and conversation monitoring

Project Presentation: CompanionAI Presentation

Project Documentation: CompanionAI Documentation

---

Functionalities

1. Friendly Companionship: Engage in warm, reassuring conversations that make you feel supported and less alone. The system is designed to respond naturally and provide a comforting presence.

2. Smart Adaptation: CompanionAI continuously learns from your interactions, adjusting to your habits, preferences, and emotional needs. Over time, it becomes more personalized and aligned with your lifestyle.

3. Event & Reminder Tracking: Stay on top of your daily life. CompanionAI remembers important events, medications, and appointments, sending reminders so you never miss what matters.

4. Family Access Dashboard: Relatives and caregivers can securely access a dedicated dashboard (example password: 1234). From there, they can add personal details, schedule reminders, and review chat history to stay updated on the user’s wellbeing.

5. Works Completely Offline: All features are available without internet. No cloud services are involved — your data remains on your device only.

6. Guaranteed Privacy: Privacy is at the core of CompanionAI. Nothing is transmitted or shared; your conversations and personal information stay fully private.

---

Project Structure

• All source code is inside the app/ folder.
• The main entry point is app.py.

Getting Started

1. Clone the Repository

git clone https://github.com/andreabochicchio02/CompanionAI.git
cd CompanionAI

2. Set Up Python Environment

Using a Conda environment with Python 3.11 is recommended:

conda create -n companionai python=3.11
conda activate companionai

3. Install Dependencies

Install all required packages from requirements.txt:

pip install -r requirements.txt

4. Install and Configure Ollama

This project relies on Ollama to run local LLMs. Make sure it is installed and then download the models:

ollama pull llama3.2:3b
ollama pull gemma3:1b

5. Run the Application

Start Ollama, then launch the web app:

python app.py

6. Access the Web Application

Once running, open: http://127.0.0.1:5000

You will find:

• User Page: chat directly with CompanionAI.
• Family Dashboard: access with password 1234 to add events, manage reminders, and view chat history.

Notes

• The dashboard password is hardcoded as 1234 for demonstration only.
• All information is stored locally. Deleting the project folder will erase your data.

---

License

This project is released under the MIT License.