Claude Code has a memory problem. Every new session starts from scratch, even if you spent hours debugging something the day before. A new open-source project called Claude Brain fixes this by giving the AI coding assistant persistent memory in a single local file.
Built on Memvid, a Rust-based memory engine, Claude Brain stores session context, decisions, bugs, and solutions in a .mv2 file inside your project's .claude/ directory. No database required. No cloud service. No API keys either. The file starts around 70KB, grows roughly 1KB per memory entry, and stays under 5MB after a year of typical use, according to the project's GitHub page. Retrieval supposedly runs sub-millisecond even with 10,000+ memories, though that's the project's own claim, not an independent benchmark.
Installation happens through Claude Code's plugin marketplace. Run /plugin add marketplace memvid/claude-brain, enable the plugin, restart, and you're set. You get commands like /mind search, /mind ask, and /mind recent to query accumulated context. Or just ask naturally in conversation. Everything runs locally, and since it's one file, you can commit it to git for version control or hand it to a teammate for instant project context.
Important caveat: this isn't an official Anthropic product, despite the name. Hacker News commenters flagged that quickly, and some raised concerns about potentially problematic git configuration advice in the setup instructions. If you're already tracking decisions in Jira with MCP, adding another memory layer might just create opacity. But for developers tired of re-explaining their codebase to Claude every morning, Claude Brain addresses a real frustration with a simple approach. Your codebase doesn't care how it got written.