Git Version Control

Your agent's identity files, memory, skills, cron jobs, and workspace should all be in a Git repository. This gives you version history (so you can roll back if something breaks), backup to GitHub, and the ability to sync or restore your setup on a new machine.

The way I have it set up: the entire ~/.openclaw/ directory is a Git repo. The workspace, skills, cron jobs, and config all live inside it. A carefully crafted .gitignore keeps secrets and ephemeral data out of the repo.

What Gets Committed

Workspace files -- AGENTS.md, SOUL.md, USER.md, IDENTITY.md, MEMORY.md, TOOLS.md, VOICE.md, HEARTBEAT.md
Memory files -- memory/YYYY-MM-DD.md (your agent's daily logs and long-term memory)
Custom skills -- anything in skills/ (SKILL.md files, reference docs, scripts)
Cron jobs -- cron/jobs.json and run history
Scripts and resources -- utility scripts, reference material

What Gets Excluded

This is the critical part. Your .gitignore needs to exclude anything sensitive or ephemeral:

# Secrets and credentials -- NEVER commit these
openclaw.json          # Contains API keys, bot tokens
openclaw.json.bak
identity/
credentials/
agents/*/agent/auth-profiles.json

# Session data -- large, ephemeral, contains full conversations
agents/*/sessions/

# Runtime and ephemeral
logs/
telegram/
update-check.json
*.lock

# Large binary files
models/              # Embedding models (multi-GB)
browser/             # Browser profiles with saved sessions
media/               # Inbound/outbound media files

# Backup files
*.bak
*.bak.*

# Nested git repos
workspace/.git/

# Projects directory -- each project is its own git repo
workspace/projects/

# OS files
.DS_Store

The big ones to never commit:

openclaw.json -- this has your Anthropic API key, Telegram bot token, and any other secrets. Committing this to a public repo would be very bad.
credentials/ -- auth tokens and similar
browser/ -- saved browser profiles may contain logged-in sessions (Amazon, etc.)
models/ -- multi-gigabyte embedding model files that don't belong in git

Setup

Fill in .gitignore before your first commit

If you commit first and add the .gitignore later, your API keys, bot tokens, and credentials will already be in the git history -- and removing them after the fact is a pain. Create and populate your .gitignore before staging anything.

cd ~/.openclaw
git init

# FIRST: create and fill in .gitignore with everything from the
# template above. Don't skip this step.
nano .gitignore   # or whatever editor you prefer

# Verify that sensitive files are ignored
git status
# openclaw.json, credentials/, models/, browser/ should NOT appear

# Now it's safe to stage and commit
git add .gitignore
git add -A
git commit -m "Initial commit"

# Create a PRIVATE repo on GitHub
gh repo create your-username/openclaw-config --private --source . --push

Make the repo private. Even with a good .gitignore, your memory files, personality profile, and workspace contain personal information you probably don't want public.

Ongoing

I have my agent commit and push periodically as part of its normal workflow. You can also add it to heartbeat tasks or do it manually:

cd ~/.openclaw
git add -A
git commit -m "Update memory and workspace"
git push

Setup Prompt

Set up git version control for the OpenClaw directory at ~/.openclaw.
Initialize a git repo if one doesn't exist. Create a .gitignore that
excludes:
- openclaw.json and any backup copies (contains API keys)
- credentials/ and identity/ directories
- auth-profiles.json files
- session data (agents/*/sessions/)
- logs, telegram data, lock files
- models/ (multi-GB binary files)
- browser/ (contains saved login sessions)
- media/ (inbound/outbound files)
- .bak files
- workspace/projects/ (each project is its own git repo)
- .DS_Store

Create a private GitHub repo and push the initial commit.

Periodically commit and push changes as part of your workflow --
especially after memory updates, skill changes, or config tweaks.

IMPORTANT: Double-check that openclaw.json is NOT tracked before
pushing. Run `git status` and verify it's listed as ignored.

Restoring on a New Machine

If you need to set up on a new machine:

git clone git@github.com:your-username/openclaw-config.git ~/.openclaw
# Then manually create openclaw.json with your API keys
# Install OpenClaw: npm install -g openclaw
# Download your embedding model to models/
# Start the gateway: openclaw gateway start

Your agent's entire identity, memory, skills, and workspace restore instantly. The only things you need to recreate manually are the secrets (API keys, bot tokens) and large binary files (models).