Chats & Threads

Record and replay chat conversations in your chatbot app. Helps you understand where your chatbot falls short and how to improve it.

Chats integrate seamlessly with traces by reconciliating messages with LLM calls and agents.

You can record chats in the backend or directly on the frontend if it's easier for you.

Start by opening a thread.

JavaScript

Python

const thread = lunary.openThread()

thread = lunary.open_thread()

You can resume an existing thread by passing an ID from an existing thread.

JavaScript

Python

// Save `thread.id` somewhere
const thread = lunary.openThread({
  id: 'your-thread-id'; // Replace with your actual thread ID
})

# Save `thread.id` somewhere
existing_thread_id = 'your-thread-id'  # Replace with your actual thread ID
thread = lunary.open_thread(existing_thread_id)

You can also add tags to a thread by passing a object with a tags param:

JavaScript

Python

const thread = lunary.openThread({
  tags: ['support']
})

thread = lunary.open_thread(existing_thread_id, tags=['support'])

Now you can track messages. The supported roles are assistant, user, system, & tool.

JavaScript

Python

thread.trackMessage({
  role: 'user',
  content: 'Hello, please help me'
})

thread.trackMessage({
  role: 'assistant',
  content: 'Hello, how can I help you?'
})

thread.track_message({
  "role": "user",
  "content": "Hello, please help me"
})

thread.track_message({
  "role": "assistant",
  "content": "Hello, how can I help you?"
})

Finally, you can track user feedback on bot replies:

The ID is the same as the one returned by trackMessage.

JavaScript

Python

const msgId = thread.trackMessage({
  role: "assistant",
  content: "Hope you like my answers :)"
})

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

msg_id = thread.track_message({
  "role": "assistant",
  "content": "Hope you like my answers :)"
})

lunary.track_feedback(msg_id, { "thumb": "up" })

To remove feedback, pass null as the feedback data.

JavaScript

Python

lunary.trackFeedback(msgId, { thumb: null })

lunary.track_feedback(msg_id, { "thumb": None })

To take full advantage of Lunary's tracing capabilities, you can reconcile your LLM and agents runs with the messages.

We will automatically reconciliate messages with runs.

JavaScript

Python

const msgId = thread.trackMessage({ role: "user", content: "Hello!" })

const res = await openai.chat.completions
  .create({
    model: "gpt-4o",
    temperature: 1,
    messages: [message],
  })
  .setParent(msgId)

thread.trackMessage({
  role: "assistant",
  content: res.choices[0].message.content
})

msg_id = thread.track_message({ "role": "user", "content": "Hello!" })

chat_completion = client.chat.completions.create(
    messages=[message],
    model="gpt-4o",
    parent=msg_id
)

thread.track_message(
    {"role": "assistant", "content": chat_completion.choices[0].message.content})

If you're using LangChain or agents behind your chatbot, you can inject the current message id into context as a parent:

JavaScript

Python

const msgId = thread.trackMessage({ role: "user", content: "Hello!" })

// In your backend, inject the message id into the context

const agent = lunary.wrapAgent(function ChatbotAgent(query) {
  // your custom code...
})

await agent("Hello!").setParent(msgId)

msg_id = thread.track_message({ "role": "user", "content": "Hello!" })

# In your backend, inject the message id into the context

with lunary.parent(msg_id):
    # your custom code...
    pass

Note that it's safe to pass the message ID from your frontend to your backend, if you're tracking chats directly on the frontend for example.