My AI Has a Brain Now

The problem: Every session, my AI wakes up fresh. No memory of yesterday's decisions, no context on my preferences, no clue that I hate cilantro or that my wife knits. It's Groundhog Day with better autocomplete.

The worse problem: Even when AI tools do remember, they trap it in proprietary black boxes. Cursor's memory doesn't talk to Claude Code. OpenClaw's context doesn't sync to my phone. I wanted a brain that follows me across tools.

So I built one.

The Architecture

Two layers. Not complicated.

Layer 1: Files (GitHub)

Structured source of truth. Lives in ~/.openclaw/workspace/ and pushes to a repo every morning at 6 AM.

MEMORY.md — Long-term facts, project context, lessons learned
WATCHLIST.md — TV shows and movies to watch
AGENTS.md, SOUL.md, USER.md — Who I am, who the AI is, how we work
Project specs, skills, heartbeat handlers

Why files: Portable. Diffable. Works offline. Any AI can read them. I own them completely.

Layer 2: Cloud (mem0)

Conversational memory layer. Semantic search. Cross-session, cross-tool.

When I say "my wife knits" or "ship it means I trust you" — that goes to mem0. Next time I mention "my wife" in Claude Code or Cursor, the context is there. No file bloat, no manual curation.

Why cloud: Fuzzy retrieval. Natural language. Ephemeral facts that don't deserve a file.

How It Works

Daily sync: OpenClaw cron fires brainpack_push at 6 AM CST. Stages, commits, pushes. If the machine explodes, my brain is on GitHub.

Real-time memory: During conversations, key facts auto-sync to mem0. User ID: mager. Searchable from any session.

Skill integration: My beginner-japanese skill uses both — mem0 for conversational progress, local file for structured lesson state. Falls back gracefully if the cloud is down.

The Tools

brainpack — Platform-agnostic CLI for managing AI brains (16 platforms supported)
OpenClaw — Gateway daemon handling cron, messaging, session routing
mem0 — Memory layer for AI agents (cloud or self-hosted)

What Changed

Before: "What was that show you recommended?" → Blank stare.

After: I mention "my wife" and the AI knows the yarn metaphor in Loooom is personal. I say "ship it" and it knows not to ask for confirmation. I ask about Japan and it remembers I'm learning Japanese for a trip in two months.

The AI stopped being a chatbot. It became a partner with continuity.

The Code

# Daily brain sync
openclaw cron add \
  --name "brain-sync" \
  --cron "0 6 * * *" \
  --tz "America/Chicago" \
  --system-event "brainpack_push"

# HEARTBEAT.md handles the event:
# 1. git add -A
# 2. git commit -m "🧠 daily brain sync $(date +%Y-%m-%d)"  
# 3. git push

mem0 is just HTTP:

curl -X POST https://api.mem0.ai/v1/memories/ \
  -H "Authorization: Token $MEM0_API_KEY" \
  -d '{"messages": [...], "user_id": "mager"}'

The Philosophy

Your AI's personality shouldn't be trapped in a vendor's database. The core of what makes your agent yours — its memory, instructions, context — should be portable, versionable, and platform-independent.

Files for structure. Cloud for context. Own your brain.

brainpack: npm install -g @mager/brainpack