🧠 memvid-mind
Claude Code finally remembers.
One file. Instant recall. Zero config.
Installation • How it Works • Commands • FAQ
https://github.com/user-attachments/assets/b57cb3db-576b-4c1f-af92-95796ba3fb5b
The Problem
You: "Hey Claude, remember when we fixed that auth bug?"
Claude: "I don't have memory of previous conversations."
You: "We literally spent 3 hours on this yesterday"
Claude: "I'd be happy to help you debug it from scratch!"
200K context window. Zero memory between sessions.
You're paying $200/month for a goldfish with a PhD.
The Solution
You: "What did we decide about the auth system?"
Claude: "Last week we chose JWT over sessions because of your
microservices architecture. I also remember we had issues with
the refresh token rotation - here's what we fixed..."
One plugin. One file. Claude remembers everything.
🚀 Installation
30 seconds. No config.
# One-time setup (if you haven't used GitHub plugins before)
git config --global url."https://github.com/".insteadOf "git@github.com:"
Then in Claude Code:
/plugin marketplace add memvid/memvid-mind
/plugin install memvid-mind
Restart Claude Code. Done.
🔮 How it Works
After install, Claude stores memories in one portable file:
your-project/
└── .claude/
└── mind.mv2 # Claude's brain. That's it.
What Gets Captured
| When | What |
|---|---|
| Session start | Injects relevant context from past sessions |
| While working | Decisions, bugs found, solutions, file changes |
| Session end | Session summary for future recall |
Why One File?
| Action | Result |
|---|---|
git commit | Version control Claude's memory |
scp file user@server: | Transfer to any machine |
| Send to teammate | Instant project onboarding |
No database. No background service. No API keys. No cloud.
📦 Commands
/mind stats # memory statistics
/mind search "authentication" # find past context
/mind ask "why did we choose X?" # ask your memory
/mind recent # what happened lately
⚡ Endless Mode
Claude hits context limits fast. memvid-mind compresses tool outputs ~20x:
Before: Read (8K) + Edit (4K) + Bash (12K) = 24K tokens gone
After: Read (400) + Edit (200) + Bash (600) = 1.2K tokens
Keeps errors, structure, key functions. Drops the noise.
Work longer without hitting limits.
❓ FAQ
Why is the .mv2 file ~4MB even when empty?
The .mv2 format is designed for instant operations on files of any size. It pre-allocates:
- Vector index for semantic search (find memories by meaning, not just keywords)
- Write-ahead log (WAL) for crash-safe writes
- Block-aligned storage for O(1) random access
This is the same architecture used by SQLite, LevelDB, and other production databases. The 4MB overhead enables sub-millisecond operations whether you have 10 memories or 10 million.
Think of it like an empty filing cabinet - it takes up space, but it's ready to organize thousands of documents instantly.
How big does the file get?
~1KB per memory. 1000 memories ≈ 1MB additional. The base 4MB is fixed overhead.
Is my data private?
100% local. Nothing leaves your machine. No telemetry. No cloud sync. You own your data.
The .mv2 file is just a file - encrypt it, back it up, delete it, share it. Your choice.
How fast is it?
Native Rust core via memvid. Sub-millisecond operations:
- Search: < 1ms for 10K+ memories
- Insert: < 0.5ms
- Load context: < 5ms
Can I reset Claude's memory?
rm .claude/mind.mv2
Or use /mind clear (coming soon).
Does it work with multiple projects?
Yes! Each project gets its own .claude/mind.mv2 file. Memories are project-scoped by default.
🔧 Configuration (Optional)
Create .claude/mind.config.json:
{
"memoryPath": ".claude/mind.mv2",
"maxContextObservations": 20,
"endlessMode": true
}
Most users don't need to configure anything.
🤝 Contributing
We love contributions! See CONTRIBUTING.md for guidelines.