Agent Brain

Memvid Shoutout

Memvid's creative approach to data storage—encoding text in video frames and using decades of video codec R&D for superior compression—inspired me to fork their Claude Brain project into an agent-agnostic version. None of this would be possible without their dedication to open source. If you're working on AI memory, check them out at github.com/memvid —it's seriously clever engineering.

The Problem

You: "Remember that auth bug we fixed?"
Agent 1: "I don't have memory of previous conversations."
Agent 2: "I did not work on this code yesterday.
You: "We spent 3 hours on it yesterday"
Agent 1: "I'd be happy to help debug from scratch!"
Agent 2: "Let me review the codebase to get an understanding" *sound of tokens burning*

small context window. Zero memory between sessions.

You're paying for a dory with a PhD.

The Fix

You: "What did we decide about auth?"
Agent 1: "We chose JWT over sessions for your microservices.
        The refresh token issue - here's exactly what we fixed..."
Agent 2: " JWT overs sessions is a more secure implementation, Here is the code we fixed"

One file. All your agents remember everything.

Installation

Claude Code (Marketplace)

# Optional one-time setup (if GitHub plugin URLs fail)
git config --global url."https://github.com/".insteadOf "git@github.com:"

# In Claude Code
/plugin add marketplace brianluby/Agent-brain

Then in Claude Code:

1. Open /plugins
2. Go to Installed
3. Enable mind
4. Restart Claude Code

On first run, memory is created at:

.agent-brain/mind.mv2

If you already have a legacy file at .claude/mind.mv2, migrate it safely to .agent-brain/mind.mv2:

if [ ! -f ".agent-brain/mind.mv2" ]; then
  mkdir -p ".agent-brain" && mv ".claude/mind.mv2" ".agent-brain/mind.mv2"
else
  echo "Destination .agent-brain/mind.mv2 already exists. Back up both .claude/mind.mv2 and .agent-brain/mind.mv2, then reconcile manually."
fi

OpenCode

Add this plugin package to your OpenCode config:

{
  "$schema": "https://opencode.ai/config.json",
  "plugin": ["@brianluby/agent-brain"]
}

Or use a local checkout while developing:

{
  "$schema": "https://opencode.ai/config.json",
  "plugin": ["file:///absolute/path/to/agent-brain"]
}

Then restart OpenCode. The plugin will:

• inject memory context on the first message of each session
• capture tool outputs to persistent memory
• expose a mind tool (search, ask, recent, stats, remember)

Optional: install OpenCode slash commands globally for all projects:

mkdir -p ~/.config/opencode/commands
cp .opencode/commands/mind-*.md ~/.config/opencode/commands/

Project-local command files are already included in .opencode/commands/.

How it Works

After install, Your Agent's memory lives in one file:

your-project/
└── .agent-brain/
    └── mind.mv2   # Agent's brain. That's it.

No database. No cloud. No API keys.

What gets captured:

• Session context, decisions, bugs, solutions
• Auto-injected at session start
• Searchable anytime

Why one file?

• git commit → version control Agent's brain
• scp → transfer anywhere
• Send to teammate → instant onboarding

Commands

In Claude Code:

/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

In OpenCode (slash commands):

/mind-stats
/mind-search authentication
/mind-ask "why did we choose X?"
/mind-recent
/mind-remember "Project uses pnpm, not npm"

These are provided in .opencode/commands/ for project-local usage. To use them in every repo, copy them to ~/.config/opencode/commands/.

Or just ask naturally: "mind stats", "search my memory for auth bugs", etc.

OpenCode Support

Agent Brain supports the same core memory lifecycle through a platform adapter model.

• Claude and OpenCode sessions can share project memory continuity.
• Unknown or incompatible platforms fail open (session continues, memory capture safely skips).
• Adapter contracts are SemVer-checked and validated through regression and contract tests.
• OpenCode packaging is published through the npm package @brianluby/agent-brain.

CLI (Optional)

No separate CLI install is required. Use the built-in memory commands in Claude Code (/mind ...) or OpenCode (/mind-* ...) for direct access.

FAQ

How big is the file?

Empty: ~70KB. Grows ~1KB per memory. A year of use stays under 5MB.

Is it private?

100% local. Nothing leaves your machine. Ever.

How fast?

Sub-millisecond. Native Rust core. Searches 10K+ memories in <1ms.

Reset memory?

rm .agent-brain/mind.mv2

---