Claude Code has a goldfish memory. This is the bowl.

What it does

Claude Code starts every conversation with a blank slate, no matter how much context you fed it yesterday. This plugin captures decisions, bugs, and solutions into a single .mv2 file inside your project, then auto-injects that history when Claude restarts. It is essentially a long-term memory graft for an LLM that otherwise has none.

The interesting bit

The entire state lives in one file—no SQLite, no ChromaDB, no cloud sync, no API keys. That means you can treat Claude’s memory like source code: diff it, branch it, or hand it to a teammate so they inherit months of project context instantly. The core is native Rust, and the README claims sub-millisecond searches across 10,000+ memories.

Key highlights

• One .mv2 file per project; grows roughly 1 KB per memory and stays under 5 MB even after a year of use.
• 100% local; nothing leaves the machine.
• Optional standalone CLI (memvid-cli) for direct file access without opening Claude Code.
• Built on the separate memvid engine, which treats single-file embedding storage as a first-class primitive.

Verdict

If you live in Claude Code and are tired of re-explaining your architecture every morning, this is worth the install. If you use Copilot, Cursor, or vanilla Claude, it is irrelevant.