Improvement 2.0

Improvement is now a conversation with an AI agent, not a text field. Previously, you had to describe the issue yourself, and the system would blindly apply changes. Now the AI agent investigates the situation first — asking clarifying questions, reviewing context — and only then suggests a specific fix. Before anything is applied, you can see exactly what will change and roll it back if needed.

The AI agent can now edit both agent instructions and the knowledge base. Previously, the Improvement feature only worked with instructions. Now the AI agent has access to the knowledge base as well and can add, edit, or delete information directly in the conversation.

Rollbacks are now safer. Each fix is independent — if you undo one, the rest remain intact. We also added the ability to re-apply a change if you accidentally rolled back something you needed.