Correct It Once, and the AI Stays Corrected

The AI quality loop: flag off-target responses from live chats with a thumbs down, test answers in the AI Playground, and manage the whole evaluation history from one page.

From a live conversation, hover over the Resolve section and click the thumbs-down icon on the response you want corrected. The evaluation is recorded without leaving the chat.

A place to test the AI before its answers reach customers. Ask it questions, inspect the responses, and judge the results in an environment separate from real conversations.

Yes. The Evaluation page lists the feedback history for review, editing when a correction needs refining, and removal once it is no longer relevant.

Whoever knows the right answer best, usually CS or a supervisor. Because it is one click from the chat, correcting becomes a daily habit rather than a special project.

The knowledge base is the source of what the AI knows; evaluation corrects how it answers. They complement each other: feedback often reveals which part of the knowledge needs sharpening.

Evaluations are recorded and become managed corrections. You test the change in the Playground first, so no correction ships unreviewed.

No. Feedback is one click, the Playground is an ordinary conversation, and the evaluation page is a searchable list. All of it is built for non-technical teams.