Use feedback tracking for:
  • user’s reactions to your chatbot’s responses directly on the frontend.
  • score LLM outputs directly yourself
You can then use this to filter llm calls and fine-tune your own models based on the data. Feedback tracking can be done in the backend or directly on the frontend if it’s easier for you. 
const msgId = thread.trackMessage({
  role: 'assistant',
  content: 'Hello! How can I help you?',
})

lunary.trackFeedback(msgId, { thumb: 'up' })

​
Example of a React Feedback component

You can also use the feedback method to track user reactions to your chatbot’s responses directly on the frontend. 
<Button onClick={() => lunary.trackFeedback(message.id, { thumb: "up" })}>
  👍
</Button>
The trackFeedback method takes two arguments:
  • runId: the ID of the message or run you want to track the feedback on.
  • feedback: an object containing the feedback data. You can use any key/value pair you want.

​
Feedback data

You can send any feedback data you want, as long as it’s a valid JSON object. We recommend using the following keys to ensure that data is displayed correctly in the dashboard.
KeyValuePreview
thumbup or down👍 / 👎
commentarbitrary stringeg. “This is not correct.”

​
Removing feedback

To remove feedback, simply pass null as the feedback data. 
lunary.trackFeedback(message.id, { thumb: null })