claude-brain

Sync your Claude Code brain across machines — portable, automatic, intelligent.

🇨🇳 中文

claude-brain is a Claude Code plugin for brain sync — sync Claude Code memory, skills, agents, rules, and settings across all your machines with semantic merge. Cross-machine, portable, zero daily effort.

---

# Machine A (work laptop) — first time setup
> /brain-init git@github.com:you/my-brain.git
✓ Brain exported: 42 memory entries, 3 skills, 5 rules
✓ Pushed to remote

# Machine B (home desktop) — join existing brain
> /brain-join git@github.com:you/my-brain.git
✓ Pulled brain: 42 memory entries, 3 skills, 5 rules
✓ Merged with local state
✓ Auto-sync enabled

# Later... check sync status from any machine
> /brain-status
=== Claude Brain Status ===
Machine: home-desktop (b7f2...)
Remote:  git@github.com:you/my-brain.git
Last push: 2 minutes ago
Last pull: 2 minutes ago
Status:  Clean

Two commands. Zero daily effort. Works forever.

---

The Problem

You use Claude Code on multiple machines. Your laptop has learned your coding patterns. Your desktop has custom skills. Your cloud VM has different rules. None of them talk to each other.

Every time you switch machines, you lose context. You re-teach Claude the same things. Your carefully crafted CLAUDE.md stays behind.

The Solution

# Machine A (work laptop)
> /brain-init git@github.com:you/my-brain.git
✓ Brain exported: 42 memory entries, 3 skills, 5 rules
✓ Pushed to remote

# Machine B (home desktop)
> /brain-join git@github.com:you/my-brain.git
✓ Pulled brain: 42 memory entries, 3 skills, 5 rules
✓ Merged with local state
✓ Auto-sync enabled

# That's it. Every session start/end syncs automatically.
# Your brain follows you.

Two commands. Zero daily effort. Works forever.

Why claude-brain?

Tool	What it does	What claude-brain adds
claude-mem	Enhances memory on one machine	Syncs your entire brain across all machines
chezmoi / dotfiles	Copies config files	Intelligently merges knowledge — resolves contradictions, deduplicates
Manual CLAUDE.md copy	Works but tedious	Auto-syncs silently on every session start/end

No other tool does cross-machine semantic merge of Claude Code's brain state.

Key Features

• Auto-sync — hooks run on every session start/end, zero effort
• Semantic merge — LLM-powered deduplication of memory and CLAUDE.md (not just overwrite)
• N-way merge — laptop + desktop + cloud VM all converge intelligently
• Encryption — optional age encryption for brain snapshots at rest
• Team sharing — share skills, agents, and rules with teammates
• Auto-evolve — promotes stable patterns from memory to durable config
• Security-first — secrets stripped, env vars excluded, private repo enforced
• Dirt cheap — ~$0.50-2.00/month typical usage via claude -p

Quick Start

Install

# Add the marketplace and install
/plugin marketplace add toroleapinc/claude-brain
/plugin install claude-brain-sync

Initialize (first machine)

/brain-init git@github.com:you/my-brain.git

Join (other machines)

/brain-join git@github.com:you/my-brain.git

With encryption

/brain-init git@github.com:you/my-brain.git --encrypt

Done. Auto-sync handles everything from here.

Commands

Command	Description
/brain-init <remote>	Initialize brain network with a Git remote
/brain-join <remote>	Join an existing brain network
/brain-status	Show brain inventory and sync status
/brain-sync	Manually trigger full sync cycle
/brain-evolve	Promote stable patterns from memory to config
/brain-conflicts	Review and resolve merge conflicts
/brain-share <type> <name>	Share a skill, agent, or rule with the team
/brain-shared-list	List all shared artifacts in the network
/brain-log	Show sync history

What Gets Synced

Component	Synced?	Merge Strategy
CLAUDE.md	Yes	Semantic merge
Rules	Yes	Union by filename
Skills	Yes	Union by name
Agents	Yes	Union by name
Auto memory	Yes	Semantic merge
Agent memory	Yes	Semantic merge
Settings (hooks, permissions)	Yes	Deep merge
Keybindings	Yes	Union
MCP servers	Yes	Union (env vars stripped)
Shared team artifacts	Yes	Union via shared namespace
OAuth tokens	Never	Security
Env vars	Never	Machine-specific
API keys	Never	Stripped automatically

Architecture

Machine A              Machine B              Machine C
┌──────────┐          ┌──────────┐          ┌──────────┐
│ claude-   │          │ claude-   │          │ claude-   │
│ brain     │          │ brain     │          │ brain     │
│ plugin    │          │ plugin    │          │ plugin    │
└─────┬─────┘          └─────┬─────┘          └─────┬─────┘
      │                      │                      │
      └──────────┬───────────┴──────────┬───────────┘
                 │     Git Remote       │
                 │  (your private       │
                 │       repo)          │
                 └──────────────────────┘

No central server. Git handles transport. Each machine merges on pull.

Merge strategy:

• Structured data (settings, keybindings, MCP) → deterministic JSON deep-merge (free)
• Unstructured data (memory, CLAUDE.md) → LLM-powered semantic merge via claude -p (~$0.01-0.05)

Security

claude-brain is designed with security as a first-class concern:

• Secrets are never exported — OAuth tokens, API keys, env vars, .claude.json are all excluded
• Pattern-based secret scanning — warns if potential secrets are detected in memory
• MCP env vars stripped — server configs sync without credentials
• Private repo enforced — warns if public repo detected
• Automatic backups — every import creates a backup in ~/.claude/brain-backups/
• Machine trust model — only add machines you fully control
• Optional encryption — age encryption for snapshots at rest

See the full Security Notice below.

API Costs

Operation	Cost	When
Structured merge	Free	Every sync
Semantic merge	~$0.01–0.05	Only when content differs
Auto-evolve	~$0.02–0.10	At most once per 7 days
Export / import	Free	Every sync

Typical monthly cost: $0.50–2.00 for active multi-machine use. Budget cap: $0.50/call (configurable).

Platform Support

Platform	Status
Linux	Fully supported
macOS	Fully supported (Apple Silicon + Intel)
WSL	Fully supported (WSL2 recommended)
Windows native	Not supported (use WSL)
Claude Code Cloud	Not supported

claude-brain runs as a local plugin that hooks SessionStart/SessionEnd to shell out to git and read/write ~/.claude/. Cloud sessions are ephemeral, sandboxed, and don't have a persistent ~/.claude to sync to, so the current architecture doesn't apply. See #32 for the discussion of what a future read-only Cloud variant could look like.

Dependencies

• git — sync transport
• jq — JSON processing (apt install jq / brew install jq)
• claude CLI — semantic merge (already installed with Claude Code)
• age — optional, for encryption

Team Sharing

Share skills, agents, and rules with teammates:

/brain-share skill my-useful-tool.md
/brain-share agent debugger.md
/brain-share rule security.md

Shared artifacts live in shared/ in the brain repo. Memory is never shared — personal only. Team members receive shared artifacts on their next sync.

Auto-Evolve

The brain runs evolution analysis every 7 days (configurable). It:

• Analyzes memory for stable, repeated patterns
• High-confidence promotions (>0.9) are applied automatically
• Lower-confidence suggestions are queued for manual review via /brain-conflicts

Trigger manually anytime with /brain-evolve.

Encryption

/brain-init git@github.com:you/my-brain.git --encrypt

• Generates an age keypair per machine
• Snapshots encrypted before push, decrypted on pull
• Recipients file controls access: meta/recipients.txt
• Backward compatible with unencrypted snapshots

Security Notice

Read this before using claude-brain. This plugin syncs your Claude Code configuration via a Git remote. Understand what data leaves your machine:

What IS exported

• CLAUDE.md, rules, skills, agents
• Auto memory and agent memory
• Settings (hooks, permissions — NOT env vars)
• MCP server configurations (env vars stripped)
• Keybindings
• Machine hostname and project directory names

What is NEVER exported

• OAuth tokens and API keys
• ~/.claude.json (credentials)
• Environment variables from settings
• MCP server env fields
• .local config files
• Session transcripts

Important considerations

1. Use a PRIVATE Git repository. Plugin warns if public repo detected.
2. Memory may contain sensitive context. Review before initializing.
3. Git history is permanent. Use git-filter-repo to purge if needed.
4. Auto-sync runs silently. Backups created before each import.
5. Semantic merge uses Claude API. Memory content is sent to claude -p.
6. Trust all machines in your network. Imported skills execute with Claude's permissions.

Contributing

Contributions are welcome! See CONTRIBUTING.md for guidelines.

License

MIT