Claude Brain, a new GitHub plugin for Claude Code, tackles an annoying problem: Anthropic's coding assistant forgets everything between sessions despite having a 200K token context window. Everything gets stored in a single file called mind.mv2 in the project's .claude directory. You can commit it to git, transfer it between machines, or share with teammates for quick onboarding. Its approach mirrors distributed versioned storage systems that prioritize Git-based state management. Built on Memvid, a single-file memory engine written in Rust, the tool achieves sub-millisecond search across thousands of memories. This underscores how Claude can implement Rust solutions. Its "Smart Frames" architecture borrows from video encoding concepts, using an append-only sequence of immutable frames that store content with timestamps and checksums. This enables time-travel debugging and crash safety, according to the project's documentation. Storage stays small, starting around 70KB and growing roughly 1KB per memory entry. Not everyone is convinced this fills a real gap. One Hacker News commenter pointed out that Claude Code has since added native memory features, which could make this plugin redundant. Another suggested using Jira integrated via Model Context Protocol instead, arguing that familiar project management patterns are well-represented in LLM training data and provide human-readable documentation that survives even if the AI tools disappear. Installation raised concerns too. One commenter questioned the recommended git config change, which rewrites GitHub SSH URLs to HTTPS, warning it could interfere with pushing to GitHub for users with certain setups. If you're already using SSH keys for GitHub authentication, check your config before running that command.