FriendForge AI — Build for a Friend

This submission for the Hacktoberfest Weekend Challenge introduces FriendForge AI, a private personal AI companion designed to assist a friend in improving English, organizing daily tasks, and practicing conversations. The application combines goals, tasks, private notes, vocabulary practice, AI chat, and study plans into a single web app. FriendForge securely stores all data locally, ensuring the user's information never leaves their computer.

The application operates on a local web server at http://localhost:8080 when initialized using Docker Compose. Upon launching the local demo, users can interact with the AI chat and observe the model's response in real-time. The GitHub repository contains an animated SVG project overview that visually represents the app's dashboard layout, AI chat, and the local Ollama model flow.

To run the application, users must install Docker Desktop, clone the repository, copy the .env.example file to .env, run docker compose up -d --build, download the default model with docker compose exec ollama ollama pull qwen2.5:3b, and finally, open http://localhost:8080 in their browser. The demo is limited to local development; however, a public demonstration is available via the video linked on the GitHub repository.

FriendForge is a PHP 5.6 web application packaged with Docker Compose. It utilizes MySQL to store application data and hosts an open-weight qwen2.5:3b model within a separate Ollama container. The PHP app communicates with the Ollama model over a private Docker network, eliminating the need for a hosted AI API. The PHP app retrieves prompts and receives responses from the Ollama container, all while maintaining the user's data privacy.

The project's source code is hosted on GitHub (github.com/mah-shamim/friend-forge) and is released under the MIT License. FriendForge's design enables open innovation by allowing users to keep the AI model, application, and data on their personal computers. Since the open-weight model is downloaded locally, the inference process can run without an internet connection, provided the computer remains disconnected from the web.

The configuration of the AI model, Ollama endpoint, and database connection is managed through environment variables, ensuring flexibility for future adjustments without the need to modify the PHP application. The MIT License enables users to inspect and adapt the code to their needs.