Git Markdown Files Proposed as Simpler Alternative to Vector RAG for AI Coding Agents

Developers using AI coding tools like Cursor and Claude Code often face a recurring problem: agents forget architectural decisions between sessions, forcing repeated explanations. A proposed solution replaces vector-based memory systems with structured Markdown files stored directly in the repository under a dedicated directory such as '.opencontext/'. Rather than embedding past conversations into a vector database, agents query a lightweight index of around 100 tokens and fetch only the relevant document for the current task. When architectural decisions change, the agent updates the existing Markdown file in place, eliminating the risk of retrieving outdated or conflicting versions that vector similarity searches can produce. The approach requires no external APIs or cloud subscriptions, keeps AI memory visible in standard Git pull request diffs, and allows different coding agents to share the same context seamlessly.