If your application requires more flexibility and you can’t use the decorators, you can use the track_event method directly.

Here’s a brief overview of how to use it:

import lunary

lunary.track_event(
    run_type,
    event_name,
    run_id,
    parent_run_id=None,
    name=None,
    input=None,
    output=None,
    error=None,
    token_usage=None,
    user_id=None,
    user_props=None,
    tags=None,
    params=None,
    metadata=None,
)

The parameters are as follows:

• run_type: The type of the run. It can be “llm”, “agent”, “tool”, “chain” or “embed”.
• event_name: The event. It can be “start”, “end”, or “error”.
• run_id: A unique identifier for the run.
• parent_run_id: (Optional) The unique identifier of the parent run in case of nested run.
• name: (Optional) The name of the model, agent or chain.
• input: (Optional) The input data for the event. Can be any JSON-safe value.
• output: (Optional) The output data for the event. Can be any JSON-safe value.
• error: (Optional) Any error information.
• token_usage: (Optional) The number of tokens used.
• user_id: (Optional) The user ID.
• user_props: (Optional) Any user properties.
• tags: (Optional) Any tags associated with the event.
• params: (Optional) The params passed to entity you’re tracking (for example, the temperature for an LLM call).
• metadata: (Optional) Any metadata.

Here’s an example of how to use it:

import uuid
import lunary

run_id = uuid.uuid4()

def my_tool(input):

  lunary.track_event(
      run_type="tool",
      event_name="start",
      run_id=run_id,
      name="my_tool",
      input=input,
      user_id="123",
      tags=["tag1", "tag2"],
  )

  # ...
  # do something
  # ...

  lunary.track_event(
      run_type="tool",
      event_name="end",
      run_id=run_id,
      output={"result": "success"},
  )