alsgur9865-sketch/second-brain-engine

Second Brain Engine is a personal knowledge system that stores notes as Markdown files and lets you search them by meaning rather than exact keywords. It is designed to work alongside AI coding tools: an agent like Claude Code or Cursor can automatically save facts, decisions, and reminders into the system while you work, and you can later recall them through a visual graph in your browser. The core idea is that the agent writes and the human reads. When Claude Code is connected, it can call remember to save a note and recall to search for related ones. On the human side, opening localhost:8000 shows a browser graph where notes appear as colored nodes connected by links. Green edges show direct note-to-note links. Gray edges show notes that are semantically similar even without an explicit link. You can click any node to read its content. The search works on meaning, not keywords. Asking how many days to get a refund will surface notes about return policies even if they never use those exact words. A separate ask endpoint goes further: it searches your notes and passes the results to a local AI model, which then answers your question using only what is in your notes. If the answer is not there, it says so rather than guessing. The engine keeps notes clean over time. It detects near-duplicate notes by comparing their meanings, merges them on request, and rewires any internal links that pointed to the old version. When you edit a note, any relationship labels connecting it to other notes are automatically re-evaluated so outdated tags do not pile up. Setup requires Python, Ollama for a local embedding model, and cloning the repo. The first embedding model download is about 2 gigabytes. You can swap the embedding backend for OpenAI or other services with a single environment variable. The system also includes a sample note vault to try before setting up your own.