When you set up an AI agent for your product, the first version is never quite right. The tone is a little off. The fallback message sounds robotic. The confidence threshold is too aggressive or too lenient.

Fixing any of these used to mean filing a ticket with your dev team and waiting.

The AI Behavior Editor changes that.

What it controls

The AI Behavior Editor is a visual interface in the Companin dashboard that gives non-technical team members direct control over how the agent behaves — without touching configuration files or writing a single line of code.

Personality and tone. Choose from preset tones (professional, friendly, concise, detailed) or write custom instructions. "Always acknowledge the user's frustration before answering" is a valid instruction.

Fallback message. What the agent says when it genuinely can't answer. You control the exact wording.

Confidence threshold. The minimum confidence score required before the agent answers. Raising this makes the agent more conservative. Lowering it makes it answer more questions.

Escalation triggers. Specific phrases that always route to your human handoff flow, regardless of confidence. "Talk to a human", "I need a refund", "cancel my account".

Topic restrictions. Questions the agent should decline to answer, even if the answer is technically in your knowledge base.

Changes take effect immediately

When you update a setting in the AI Behavior Editor, it applies to the next conversation. No deployment required. No cache to clear.

Every change can be previewed before it goes live. The test panel lets you send sample questions and see exactly how the agent will respond with the new configuration — including tone, fallback behavior, and escalation routing.