Track Your PAI Personal Data in Git

[ecielam]

PAI generates valuable data as you use it—learnings, preferences, research, work history. By default this lives in MEMORY/, USER/, and WORK/ inside ~/.claude.

I wanted a clean way to version control this data separately, so I set up a simple pattern: keep personal data in its own git repo and symlink it into PAI.

Why this is nice:

• Your learnings have history — see how your PAI knowledge evolves over time
• Easy backup — just git push to your private repo
• Portable — clone to any machine, run one script, done
• Clean separation — PAI code stays separate from your data

I made a template with a setup script:

👉 https://github.com/ecielam/pai-personal-data

Quick start (obviously use whatever you want for my-pai-data):

git clone https://github.com/ecielam/pai-personal-data.git ~/my-pai-data
rm -rf ~/my-pai-data/.git
cd ~/my-pai-data
./setup.sh

The script creates symlinks from PAI to your repo. After that, init git and push to your own private repo:

git init
git add -A
git commit -m "Initial PAI personal data"
git remote add origin git@github.com:YOU/your-private-repo.git
git push -u origin main

Now all your PAI personal data is version controlled and backed up.

[ollie-dev]

I wonder what @danielmiessler proposes regarding git. He said in one Video, that he has everything in git, but it is not part of the setup nor documentation, as fas as my research shows.

I like the way here as solution, but wonder, what the "official" way is.

[jogi-poikola]

This question of the "official" or even "community best practice" way is very relevant. I know, I should be versioning in git more consistently, but I don't quite know from where to start. I am not skilled to designing my own hooks and the PAI memory/learnings are a bit fuzzy to me.

Mostly I worry of breaking the skills that I created and not being able to go back because I forgot to commit something.

[danielmiessler]

I definitely have pretty much everything in Git. And would recommend it. Make sure it's a private repo, and you don't accidentally end up putting that stuff somewhere.

…

On Tue, Jan 20, 2026 at 1:15 PM, Ollie < ***@***.*** > wrote: I wonder what @ danielmiessler ( https://github.com/danielmiessler ) proposes regarding git. He said in one Video, that he has everything in git, but it is not part of the setup nor documentation, as fas as my research shows. I like the way here as solution, but wonder, what the "official" way is. — Reply to this email directly, view it on GitHub ( #436 (comment) ) , or unsubscribe ( https://github.com/notifications/unsubscribe-auth/AAAMLXSBVIQEDIS4X7FUIPD4H2LINAVCNFSM6AAAAACSBBEE4KVHI2DSMVQWIX3LMV43URDJONRXK43TNFXW4Q3PNVWWK3TUHMYTKNJVGM2DAMI ). You are receiving this because you were mentioned. Message ID: <danielmiessler/Personal_AI_Infrastructure/repo-discussions/436/comments/15553401 @ github. com>

[ShaneDugas]

I use this philosophy as well. One piece of wisdom that I can offer here is to do this one very slowly and make sure you "hook" your security skill very carefully. Verify, verify, verify that it's catching all of the commit output. After that, you're golden.

[jgmontoya]

The first thing I did with PAI was set up a hook that autocommits and pushes the entire ~/.claude folder. I'm pushing to a private repo to a locally run gitea instance so I can be a bit more relaxed regarding security

[ecielam]

I took that approach with 2.0, but it was too dangerous. It would commit things I didn't want committed, and often you'd end up with sensitive data in ~/.claude/projects (since it's the message transcripts from claude code)

And so when Daniel moved to the whole 'USER + MEMORY' model it made a lot more sense to be explicit with what gets tracked since now all of the custom stuff is isolated to specific locations. So I no longer have to track all of .claude.

Also a factor, I can use this same data/repo (or sub-sections of it) for different PAI installs (think personal, and 2-3 business installs)

[wojteksbt]

Thanks for your work! I'm looking for a good way of syncing personal data and, after having a repo in .claude directory with most of the files ignored, I'm happy to try your approach. I'm wondering why does the setup script creates backups of most of the directories instead of just moving the files to the personal data repo? And why is CORE_USER special in this context?

[ecielam]

CORE_USER is ~/.claude/skills/CORE/USER ... this is where we can override skills and other PAI stuff, and the CORE skills automatically check it. Since that's customization, I want that tracked.

It gets backed up because I had my PAI assistant create the script, and that's something that PAI setup scripts do, so it does it automatically. Plus, moving stuff is destructive, so PAI won't generally do it.

I personally don't use the setup script, it's provided as an example.

[ecielam]

I should add as a general comment that I don't autocommit in my personal repository. I may get to that point, but for now, I want a bit more control over when stuff gets committed.

[klittle32]

Forgive me if this has been discussed before. I have been thoroughly enjoying and benefiting from PAI since installing it for the first time two short weeks ago. This is a really exciting and high-impact project and I am grateful to Daniel and all other contributors.

I've been thinking about ways to separate some concerns (user data vs code) and perhaps modularize the PAI codebase a bit.

Idea 1: Put the user data into a user-specified directory. So we would separate $PAI_DIR from $PAI_DATA_DIR

Idea 2: Distribute PAI as plugins

Here is a mockup of how it might end up looking (courtesy of claude):

~/pai-data/ — User's Portable Identity and State

~/pai-data/                             # $PAI_DATA_DIR - Never touched by plugin updates
│
├── TELOS/                              # Deep goal system - who you are
│   ├── MISSION.md
│   ├── GOALS.md
│   ├── PROJECTS.md
│   ├── BELIEFS.md
│   ├── MODELS.md
│   ├── STRATEGIES.md
│   ├── NARRATIVES.md
│   ├── LEARNED.md
│   ├── CHALLENGES.md
│   └── IDEAS.md
│
├── MEMORY/                             # Three-tier memory system
│   ├── Hot/                            # Active session context
│   │   ├── current-session.md
│   │   └── recent-decisions.md
│   ├── Warm/                           # Recent history (days/weeks)
│   │   ├── sessions/
│   │   └── summaries/
│   └── Cold/                           # Long-term archive
│       ├── 2025/
│       └── 2026/
│
├── SIGNALS/                            # Learning feedback loop
│   ├── ratings.jsonl
│   ├── corrections.jsonl
│   ├── successes.jsonl
│   └── failures.jsonl
│
├── IDENTITY/                           # DA personality config
│   ├── name.txt
│   ├── voice.json
│   ├── personality.md
│   └── principal.md
│
├── PREFERENCES/                        # User customizations
│   ├── tech-stack.md
│   ├── workflows.md
│   ├── notifications.json
│   └── security-policy.json
│
└── config.json                         # PAI_DATA_DIR settings

~/.claude/plugins/ — PAI as Plugin Series

~/.claude/
├── settings.json
│
└── plugins/
    │
    ├── pai-core/                       # PLUGIN: Core infrastructure
    │   ├── plugin.json
    │   ├── skills/
    │   │   └── core/
    │   │       ├── SKILL.md            # Main routing skill
    │   │       └── templates/
    │   ├── hooks/
    │   │   └── hooks.json              # SessionStart, Stop hooks
    │   └── scripts/
    │       ├── load-telos.sh           # Loads $PAI_DATA_DIR/TELOS/*
    │       ├── load-context.sh
    │       └── session-end.sh
    │
    ├── pai-memory/                     # PLUGIN: Memory & learning system
    │   ├── plugin.json
    │   ├── skills/
    │   │   └── memory/
    │   │       └── SKILL.md
    │   ├── hooks/
    │   │   └── hooks.json              # PostToolUse signal capture
    │   └── scripts/
    │       ├── capture-signal.sh       # Writes to $PAI_DATA_DIR/SIGNALS/
    │       ├── archive-session.sh      # Hot → Warm → Cold rotation
    │       └── query-memory.py
    │
    ├── pai-security/                   # PLUGIN: Security validation
    │   ├── plugin.json
    │   ├── hooks/
    │   │   └── hooks.json              # PreToolUse validation
    │   └── scripts/
    │       ├── validate-command.sh
    │       └── audit-log.sh
    │
    ├── pai-voice/                      # PLUGIN: Voice notifications (optional)
    │   ├── plugin.json
    │   ├── hooks/
    │   │   └── hooks.json
    │   └── scripts/
    │       ├── speak.sh
    │       └── prosody-enhance.py
    │
    ├── pai-notifications/              # PLUGIN: Push notifications (optional)
    │   ├── plugin.json
    │   ├── hooks/
    │   │   └── hooks.json
    │   └── scripts/
    │       ├── notify-ntfy.sh
    │       └── notify-discord.sh
    │
    ├── pai-agents/                     # PLUGIN: Dynamic agent composition
    │   ├── plugin.json
    │   ├── agents/
    │   │   ├── researcher.md
    │   │   ├── critic.md
    │   │   └── implementer.md
    │   └── skills/
    │       └── agents/
    │           └── SKILL.md
    │
    ├── pai-telos/                      # PLUGIN: Goal management skill
    │   ├── plugin.json
    │   ├── skills/
    │   │   └── telos/
    │   │       └── SKILL.md
    │   └── commands/
    │       ├── goals.md                # /goals
    │       ├── mission.md              # /mission
    │       └── reflect.md              # /reflect
    │
    ├── pai-research/                   # PLUGIN: Research skill (optional)
    ├── pai-osint/                      # PLUGIN: OSINT skill (optional)
    └── pai-redteam/                    # PLUGIN: Red team skill (optional)

Environment Variables

export PAI_DATA_DIR="$HOME/pai-data"
export PAI_DIR="$HOME/.claude"
export DA="Aria"
export PRINCIPAL="Kyle"

Installation Flow

# 1. Install core plugins
/plugin install pai-core
/plugin install pai-memory
/plugin install pai-security


First run detects missing $PAI_DATA_DIR → setup wizard creates ~/pai-data/


Install optional plugins as desired

/plugin install pai-voice

/plugin install pai-research

Cross-Platform Portability

~/.claude/plugins/pai-*      → Claude Code
~/.cursor/plugins/pai-*      → Cursor  
~/.windsurf/plugins/pai-*    → Windsurf
All reference the same ~/pai-data/ — your identity is platform-independent.

Key Benefits

Current Approach	Plugin + External Data
PAI takes over entire ~/.claude	PAI installs as discrete plugins
User data mixed with infrastructure	User data fully separated
Upgrades require careful merging	rm plugin && reinstall is safe
Single-platform identity	Same identity across all AI platforms

[danielmiessler]

Hey @ecielam, thanks for raising this, and sorry it sat for a while.

We're changing how LifeOS ships. Instead of cloning a full ~/.claude directory and running it as a complete system, LifeOS is becoming a skill you install through an agentic installer. The installer hands integration to your own AI, which reads your actual machine (your OS, your paths, your harness) and wires the hooks and system prompt in where they belong.

That's aimed right at what you hit here. The old "one directory, one layout, hope it matches your setup" approach is exactly what broke for so many people, and the new model should handle it far better because your AI does the integration per machine instead of us guessing.

So we're closing this in prep for that release. If it still bites you once the skill-based version is out, reopen or file a fresh one and we'll jump on it. Appreciate you taking the time.