Checkpoint & Rollback
AI agents make mistakes. Hermes Agent creates checkpoints before every significant operation, so recovery is always one command away.
What is Agent Checkpointing?
Agent checkpointing is the practice of saving the complete state of an AI agent before executing potentially destructive or risky operations. If the operation fails or produces unintended results, the agent can be restored to its pre-operation state. This transforms autonomous agents from fragile experiments into reliable production tools.
What Gets Checkpoined
Memory State
All conversation history, learned preferences, and active context are preserved in the checkpoint.
Filesystem
File contents, directory structures, and permissions are snapshotted for complete restoration.
Configuration
Environment variables, tool settings, and model parameters are included in the checkpoint.
Skill Versions
The exact versions of all loaded skills are preserved, ensuring consistent behavior after restoration.
Learn More
Frequently Asked Questions
What is a checkpoint in Hermes?
A checkpoint is a snapshot of the agent's state — including memory, filesystem, and configuration — captured before a potentially risky operation. If something goes wrong, Hermes can restore this snapshot instantly.
When are checkpoints created automatically?
Hermes creates checkpoints before file modifications, shell command execution, package installations, and deployments. You can also trigger manual checkpoints at any time.
How fast is rollback?
Rollback is near-instantaneous for memory and configuration. Filesystem rollback depends on the snapshot mechanism (Docker layer restore or filesystem copy) but typically completes in under a second.
Can I review changes before rolling back?
Yes. Hermes shows a diff of all changes made since the last checkpoint, allowing you to selectively revert specific modifications while keeping others.