Blog / operations
How to Upgrade, Back Up, and Roll Back Hermes Agent
An agent that holds memory and channel credentials is stateful, and stateful systems need a backup story before they need an upgrade story. This guide covers what to copy, when to upgrade, how to prove an upgrade worked, and how to get back to the previous version when it does not.
- 01
Inventory what is stateful
Configuration, memory files, session history, and provider credentials. Anything in that list has to survive the upgrade, so it has to be in the backup.
- 02
Take a restorable backup
Copy the configuration and memory directories somewhere outside the host. A backup you have never restored is a hypothesis, not a backup.
- 03
Record the current version
Write down the exact version or image tag you are running. Without that, rolling back becomes guesswork.
- 04
Upgrade and run a verification message
After upgrading, send a real message in a real channel and confirm the reply, then check that memory from before the upgrade is still present.
- 05
Decide to keep or roll back
Give the new version a defined window, such as a week of normal use, before deleting the previous one. If a workflow breaks, roll back rather than debug under pressure.
Four-step path
How do you upgrade Hermes Agent safely?
Back up first, upgrade deliberately, verify with a real message, and keep the old version available until the new one has proved itself.
- Step 01
Inventory what is stateful
Configuration, memory files, session history, and provider credentials. Anything in that list has to survive the upgrade, so it has to be in the backup.
- Step 02
Take a restorable backup
Copy the configuration and memory directories somewhere outside the host. A backup you have never restored is a hypothesis, not a backup.
- Step 03
Record the current version
Write down the exact version or image tag you are running. Without that, rolling back becomes guesswork.
- Step 04
Upgrade and run a verification message
After upgrading, send a real message in a real channel and confirm the reply, then check that memory from before the upgrade is still present.
- Step 05
Decide to keep or roll back
Give the new version a defined window, such as a week of normal use, before deleting the previous one. If a workflow breaks, roll back rather than debug under pressure.
What exactly should you back up?
Configuration, memory and profile files, session history, and the list of credentials needed to run, kept somewhere that is not the host itself.
- The configuration directory, including provider and channel settings.
- Memory and user profile files, which are the accumulated value of the deployment.
- Session history if you rely on older conversations.
- A written list of required secrets, without the secret values in the same file.
How often should you upgrade?
On a schedule you choose, not whenever a release appears. For an agent people depend on, a monthly upgrade with a backup is saner than upgrading weekly.
- Upgrade for a specific reason: a fix, a feature, or a security notice.
- Avoid upgrading right before a dependent event such as a launch.
- Upgrade one environment first if you run more than one.
- Keep the rollback path documented next to the upgrade command.
How do you roll back?
Redeploy the previously recorded version and restore the configuration and memory you copied. Both halves are needed; a version rollback alone loses anything the new version wrote.
On managed hosting this is a redeploy to a known image tag. On a self-hosted install it is checking out the recorded version and rebuilding, which is why the version number belongs in your notes and not in your memory.
How do you prove an upgrade worked?
One real message in one real channel, plus confirmation that pre-upgrade memory is still there and that a tool call still executes.
- Send a message in the exact channel users rely on.
- Ask something only the agent's memory could answer.
- Trigger one tool call so the integration surface is exercised.
- Check that scheduled runs still fire after the restart.
What else belongs in the runbook?
Restart behaviour, provider failover, and what to do when a credential expires.
Which primary sources support this page?
Product behavior is checked against the current deployment flow. These external links provide the upstream project or channel documentation used for setup details.
Direct answers
Frequently asked questions
Does upgrading Hermes Agent erase memory?
Not by itself, but anything unexpected is survivable only if you copied configuration and memory first. Back up before every version change.
Can I skip versions when upgrading?
Usually, since the runtime is replaced rather than patched in place. The risk is in configuration changes, not in the version jump.
What is the fastest way to roll back?
Record the exact version or image tag before upgrading, then redeploy that version and restore the configuration and memory you copied.
Key takeaways
- An agent that holds memory and channel credentials is stateful, and stateful systems need a backup story before they need an upgrade story. This guide covers what to copy, when to upgrade, how to prove an upgrade worked, and how to get back to the previous version when it does not.
- Start with one model, one channel, and one controlled test conversation.
- Use managed hosting when deployment speed matters more than operating the server yourself.
Continue the setup
Related Hermes Agent guides
Blog guide
How to Restart or Redeploy Hermes Agent Safely
Restart or redeploy Hermes Agent safely by checking runtime state, active sessions, credentials, and message flow before returning to production use.
Read nextBlog guide
Hermes Agent Deployment Checklist
Use this Hermes Agent deployment checklist to choose a model, connect Telegram, Discord, or WhatsApp, test replies, secure credentials, and go live.
Read nextBlog guide
Hermes Agent Credential Vault and Key Pools
Store model and tool credentials in the Hermes Agent credential vault, build key pools for rate limits, and rotate secrets without breaking the agent.
Read nextBlog guide
Hermes Agent Troubleshooting: A Recovery Toolkit
Diagnose the common Hermes Agent failures in order: install, provider access, session storage, channel wiring, and tool calls. Fix one layer at a time.
Read next