Agent versioning for uninterrupted production updates
Key Takeaway
An activated agent version is locked, so you update a live agent by creating a new version, testing it in the Preview panel, and then activating it. Activating the new version automatically deactivates the old one, so users keep using the old version until the new one is in service.
Agent versioning lets you keep a working agent live while you build and test changes in a separate, editable copy. Each version holds its own copy of the agent's setup. Because you can't change a version after you activate it, versions are how you update an agent that is already in production.
What each version holds
A version is an editable copy of an agent. Each version gets its own copy of the agent's assets:
- System messages
- Subagents, including standard subagents, custom subagents, and your customizations to standard subagents
- Actions, including standard and custom agent actions, and your customizations to standard actions
- Variables, output mappings, and input assignments
Subagents and actions don't support versioning of their own. A change you make to a custom subagent or action applies only to the agent and version it is assigned to.
A new version inherits the name of the agent it was created from, plus the next sequential number. For example, a version created from My Agent - 3 is named My Agent - 4. You can change the name with "Rename Version".