Backup and upgrade

What to back up#

Everything lives in the data folder: data/ inside the mindbaton folder (or wherever MINDBATON_DATA points), or the mindbaton-data volume with Docker. The one file that matters is mindbaton.db; ai_keys holds your optional AI keys. Your mindbaton.env (if you changed it) is worth keeping too.

Make a backup#

Mindbaton can keep running. This uses SQLite's own backup, so the copy is consistent even mid-write:

cd ~/mindbaton
python3 -c "import sqlite3; sqlite3.connect('data/mindbaton.db').backup(sqlite3.connect('mindbaton-backup.db'))"

With Docker:

docker compose exec mindbaton python3 -c "import sqlite3; sqlite3.connect('/data/mindbaton.db').backup(sqlite3.connect('/data/backup.db'))"
docker compose cp mindbaton:/data/backup.db ./mindbaton-backup.db
docker compose exec mindbaton rm /data/backup.db

Keep the backup somewhere private — it contains everything you've told your AIs (secrets are removed, but it is still your life). To back up every night, put the first command in a cron job or systemd timer.

Want your memory in another tool instead? Export it from the app, or with a full token: curl -H "Authorization: Bearer mb_xxxxxxxx" "http://localhost:3004/export?format=json" (also cypher for Neo4j and graphml for Gephi).

Restore#

Stop Mindbaton, put the backup in place, start it again:

systemctl --user stop mindbaton                        # macOS: launchctl bootout gui/$(id -u)/ai.mindbaton
cp mindbaton-backup.db data/mindbaton.db
rm -f data/mindbaton.db-wal data/mindbaton.db-shm
systemctl --user start mindbaton                       # macOS: ./install.sh

With Docker:

docker compose stop
docker compose run --rm -v "$PWD:/backup" mindbaton \
  sh -c "cp /backup/mindbaton-backup.db /data/mindbaton.db && rm -f /data/mindbaton.db-wal /data/mindbaton.db-shm"
docker compose start

(Removing the old -wal/-shm files matters: they belong to the database you're replacing.)

Upgrade#

Upgrades keep all your data. Mindbaton saves every message as it arrived and builds the memory graph from those, so when a new version understands messages better it simply rebuilds the graph on its first start — and anything you told it to forget stays forgotten. A backup first never hurts.

Installed with install.sh:

mindbaton update   # from any folder: downloads, runs the self-test (goes back if it fails), restarts the service

The installer adds the mindbaton command (to ~/.local/bin, or Homebrew's folder on a Mac). On an install older than that, run cd ~/mindbaton && python3 mindbaton.py update once: the update adds the command, and from then on mindbaton update is all you need. mindbaton doctor checks everything; mindbaton --help lists the rest.

Docker (Windows, macOS, Linux, NAS), in the folder with compose.yaml:

docker compose pull
docker compose up -d

That fetches the newest published image (ghcr.io/dkshbyte/mindbaton) and restarts on it. Your data stays in the mindbaton-data volume. Started with plain docker run instead? Then docker pull ghcr.io/dkshbyte/mindbaton, remove the old container (docker rm -f mindbaton) and run the same docker run command again: the volume keeps everything.

Automatic updates#

Let Mindbaton update itself every night. Each update is tested first, and if the new version fails, you stay on (or go back to) the one you had, and that release is skipped until a newer one comes out. Your memories are never touched.

• Installed with install.sh (Linux, macOS): python3 mindbaton.py autoupdate on (a systemd timer or launchd job, around 4:00 each night; a night the computer was off runs at the next start). autoupdate off stops it; autoupdate status shows it.
• Docker on Windows (PowerShell, in the folder with compose.yaml): powershell -ExecutionPolicy Bypass -File scripts\auto-update.ps1 -Install adds a Task Scheduler job for 04:00 (-Remove takes it away; without a switch it updates once, now).
• Docker on Linux, macOS or a NAS: sh scripts/auto-update.sh --install adds one line to your crontab (--remove takes it away).

What each run did is in auto-update.log next to compose.yaml (Docker) or ~/.mindbaton/install.log.

Why not Watchtower? It was archived in December 2025, and it needs full control of Docker to work. These scripts only update Mindbaton's own container.

Your folder is from before 25 September 2026? The project's history was rewritten once that day, so git pull refuses in older copies. Run this once in the folder (your data and mindbaton.env aren't touched):

git fetch origin
git reset --hard origin/main

After upgrading, reload the browser extension if it changed: download /mindbaton-extension.zip again, unzip it over the old folder, and press the reload button on it in chrome://extensions.

Moving to another computer#

Install Mindbaton on the new computer, stop it, copy mindbaton.db (and ai_keys) into its data folder, and start it. Your password, device tokens and memories all come along — just update the address in your extension and apps if it changed.