vakovalskii/gitmark-memory-bank

GitMark Memory Bank is a Python tool for building a searchable knowledge base out of plain Markdown files stored in a git repository. The idea is that a project's documentation should live as simple text files you can read in any editor, with a README in each folder acting as that folder's index. Everything else, including the search index, an HTML overview, and a visual graph of links between documents, gets generated fresh from those files on demand and is never stored in git itself. The core component is a command-line tool called gitmark that uses only Python's built-in standard library and requires no external packages. It builds a local SQLite database with a full-text search index that supports ranked results, substring matching, and fuzzy matching to handle typos and non-English text. You can search by keyword and get back results showing the file, the heading, and a snippet of context. The map command produces a self-contained HTML file with a collapsible document tree, rendered Markdown, and an interactive graph showing how documents link to each other. A lint command can optionally check for broken links, missing folder indexes, and other structural problems. There is also a plugin for Claude Code, the AI coding assistant tool. Once installed, it gives the AI agent two skills: searching the knowledge base instead of blindly searching files, and following a set of rules when editing documents. This is aimed at teams that use AI assistants for coding and want the AI to have reliable, fast access to the project's documentation rather than guessing from source code alone. The tool is intentionally minimal. The README positions it against wikis, Notion, and vector-database memory systems by arguing that plain Markdown plus git plus a single auditable Python file is simpler to own, works offline, and shares nothing with external services. The project is licensed under MIT and requires Python 3.7 or newer.