Introduction
Why do we need events and events-based analytics?
LLM applications generate thousands of logs and traces every day. Analyzing them manually and making sense of it is hard, if not impossible, considering the amount of data generated.
That’s where events-based analytics comes in.
- Catch failures in real time
Use Laminar’s Semantic Events to catch failures in your agents in real-time based on the online evaluations.
- Evaluate and gather metrics on agent behavior
Use Laminar’s Semantic Events and Regular Events to gather metrics on the user or agent behavior based on the semantic meaning.
Overview
Events are the building blocks of the analytics in Laminar.
Laminar Events are OpenTelemetry-compatible.
All events belong to a span. In other words, every event must be sent within a function instrumented with Laminar.
Event properties
Each event has the following properties:
- Name
Name is needed to identify the event. Event name must be unique within the project.
For example, name can be user_sentiment
.
- Value type
There are three types of event values:
Boolean
String
Number
- Value
Based on the event value type, each event will have a value.
For example, an event of Boolean
value type will be true
or false
.
- Timestamp
Timestamp is the time when the event has been created. If you don’t specify the timestamp, the time at which the event was sent will be used.
Sending events
You can send Laminar events and their values from code.
Events are useful when you want to make assertions, record known actions which agent takes, or track metrics important for your use case.
For example, you can record if the response matches regular expressions, if the chat assistant has found the answer to the user’s question, or which tool has been used how many times.
Learn more on how to send events and the use cases.