AI agent

The AI agent resolves conversations on its own and hands over to a human with full context when it should not answer.

How a message is handled

1. Deterministic triggers — handover keywords (human, refund, …) or the max-replies cap escalate immediately.
2. Knowledge grounding — relevant articles from the enabled knowledge sources are added to the prompt.
3. Model call with tools — Claude (claude-opus-5 by default) may call your webhook tools or the built-in handover_to_human tool.
4. Confidence — the reply carries a confidence score. Below the threshold the agent hands over instead of replying.
5. Reply or handover — the reply is sent on the channel; a handover routes the case to a queue, assigns an agent, leaves an internal summary and publishes ai.handover.

Status

Before live answers are enabled

Until your administrator configures the model provider, the agent runs a deterministic demo mode that still exercises knowledge grounding, tool calls, guardrails and handover, so the product can be evaluated end to end.