Building a Privacy-First AI Companion with Next.js, FastAPI and Ollama

Most AI applications send your conversations to a cloud LLM. I wanted to experiment with a different approach: What if an AI companion could run entirely on your own machine? That idea led me to build MindMirror — an open-source AI companion for emotional reflection that supports local LLMs through Ollama, removing the requirement for paid AI APIs. GitHub: https://github.com/HASSANFARYAD/MindMirror Live Demo: https://mindmirror-neon-tau.vercel.app/ What is MindMirror? MindMirror combines journaling with AI-powered reflection. Users can write about what is on their mind and explore: 🧠 Emotional patterns The goal is not to replace therapy or pretend that AI can do that. The goal is to create a structured and private space for reflection. Why Local AI? Privacy was one of the main reasons behind this project. Many AI applications rely entirely on cloud APIs, which means user conversations are sent to third-party services. With Ollama, MindMirror can run models locally. This means users can experiment with AI while keeping their journal data on their own machine. This approach combines: Local LLMs MindMirror is built using: Frontend The application follows a simple architecture: User For voice journaling: Voice Input The conversation flow is inspired by CBT-style reflection patterns. The process is roughly: Detect → Validate → Examine → Ground → Next Step Detect Identify emotions, themes, and thought patterns. Validate Acknowledge feelings without judgment. Examine Explore assumptions and possible cognitive distortions. Ground Focus on practical perspectives and context. Next Step Suggest small actions or reflections. This structure helps create more useful interactions than simple question-answer chat systems. Running MindMirror Locally Clone the repository: git clone https://github.com/HASSANFARYAD/MindMirror.git Start the application: docker compose up Install Ollama: ollama pull llama3 After setup, the AI can run locally without requiring paid APIs. What I Learned Building MindMirror taught me several things: Local LLMs are becoming practical for real applications. Some ideas I am exploring: Better emotional trend visualization MindMirror is fully open source. If you are interested in: Local AI Feel free to explore the code, fork the project, open issues, or contribute. GitHub: https://github.com/HASSANFARYAD/MindMirror Live Demo: https://mindmirror-neon-tau.vercel.app/ I would appreciate feedback from developers building local AI and privacy-focused applications.