A newer unicorn release can add tables and columns to your database. npm run upgrade applies those changes first and then deploys the new code. Your history, sources and connected apps carry over.

​
Upgrade

1

Back up the database

The free Cloudflare plan has no point-in-time restore, so this export is your only way back. Keep the file until you have run the checks below.
2

Get the new release

From your unicorn folder:
If git pull stops because of changes in wrangler.jsonc, that is your own database ID. Keep your line when you resolve it.
3

Run the upgrade

This runs two commands in order: it applies the database migrations to your remote database, then deploys the Worker.
4

Check it

Open https://<your-worker>.workers.dev/health. You want "status":"ready", "database":true and "scheduler":"running". Then ask your agent to check unicorn’s status, as in Connect your agent.
You do not need to start the scheduler again. It keeps running through a deploy unless you had stopped it.

​
Why the database changes first

The hourly sync in the new code reads tables and columns that the new migrations add. If you deploy first, every sync fails until the migration lands. If you migrate first, the old code keeps working against a database that only gained something. For the same reason, do not run wrangler deploy by itself to upgrade. Always use npm run upgrade, or the two commands it contains, in that order. To see which migrations your database has applied:
A migration that has already run is never run again.

​
Check the sync after an upgrade

To run a sync now rather than wait for the hour, use the settings page: open Maintenance and click Sync now. From a terminal, use your admin token:
A 200 answer means every source pulled cleanly. A 207 means at least one source reported an error, and the response names it.

​
If you deployed an earlier version

Releases before the memory layer ran their own AI agent inside the Worker and had a different database. If your database is at migration 0011, the upgrade to the current release also:
  • Deletes the old agent’s conversation history, notification queue and job records. They are not recoverable without your backup.
  • Keeps your saved corrections.
  • Copies your existing change history into the new format, and later removes the noisiest legacy entries, such as view counts and vote counts, while keeping real changes like deadlines and state changes. Your agent’s saved position in the change feed keeps working.
  • Removes the old agent class from the Worker.
The Telegram, email and Discord settings are no longer read by anything. You can delete those secrets:
Delete only the ones secret list shows. Keep ADMIN_TOKEN, MCP_TOKEN and every source credential.

​
Roll back

Database migrations only go forward, and your backup cannot be replayed over a migrated database. To go back, restore the backup into a new database and point the older release at it. The unicorn repository’s docs/UPGRADING.md has the exact steps. Your original database stays untouched, so you can move forward again later.