Talk to Your Past Self: The AI Journal That Chats Back

The practice of journaling is rapidly vanishing. At one time, the leather-bound diary or the composition book served as a repository for everything from mundane daily observations to the deepest existential crises. Recording these thoughts and observations required a person to sit in silence, reflect on their experiences, and commit their internal world to paper.

Today, the examined life has largely migrated from the private page to the public social media feed. With this change, writing has become more performative than reflective. The deliberate pace of handwriting has also been sped up by the fast pace of digital tools. In short, the practice of self-reflection is fading, and that isn’t good for any of us.

GitHub user cristal-orion still sees journaling as an important practice, and has developed a distraction-free digital tool to get the most from it. Called Reminor, it is an AI-powered personal diary. As you write your thoughts, Reminor analyzes them, remembers them, and chats with you about them. All of this is to help you get more from your journal entries — things that may be lost in the hundreds of pages that accumulate over the years.

Unlike most modern AI journaling platforms, Reminor is not a website or a subscription service. It is designed to run locally on a Raspberry Pi. The system hosts its own web interface and backend, storing every entry directly on the device. Users type on an external keyboard, and cristal-orion used a 4-inch Waveshare LCD mounted inside a 3D-printed case, creating a dedicated writing appliance that sits quietly on a desk and consumes under five watts of power.

Under the hood, the Pi runs a Docker-based software stack composed of a Svelte frontend and a FastAPI backend. Journal entries are saved as simple dated text files. Reminor processes each entry using local embedding models that convert writing into mathematical vectors — numerical representations of meaning. Those vectors are stored using a system called Memvid, which compresses them into video frames to conserve storage while enabling semantic search. Instead of searching for specific words, a user can ask questions like “when was I worried about work?” and retrieve related passages written months or years apart.

The system also tracks emotional tone across eight dimensions, visualizing long-term mood trends and writing streaks. An integrated chat interface allows the user to ask questions about their own past; the software retrieves relevant entries and uses an AI model to summarize patterns or connections. Importantly, all of this can run offline using local models. Remote large language model APIs from providers such as OpenAI, Anthropic, or Google are optional and only used if the user explicitly enables them.

If you’d like to try out Reminor for yourself, start by looking over the GitHub repository.—