A Step is a fundamental building block within a Thread that represents a single unit of interaction or operation.

Steps are versatile and can encompass various types of activities, such as sending a message, making a call to a large language model (LLM), or executing a custom function.

By organizing interactions into Steps, the Literal platform enables users to structure, visualize, and analyze the flow of conversations or processes in a granular and meaningful way.

​
Types of Steps

There are several types of Steps, each serving a distinct purpose:

​
Run

A run step signifies the execution of an agent or a chain that takes multiple steps. Unlike other step types, it does not have to be part of a Thread.

This is useful if you are monitoring non conversational use cases.

​
Message

A message step represents a message exchange within a Thread. There are two primary subtypes of message steps:

  • User Message: This type of step is used to log messages sent by the user. It helps in tracking user inputs and requests.
  • Assistant Message: This step logs messages generated by the assistant or application in response to user inputs.

​
Tool

A tool step is used to document interactions with external tools or services. This step type is particularly useful for applications that interact with external APIs, databases, or services.

​
LLM

An llm step is used to document interactions with large language models (LLMs). This step type is particularly useful for applications that leverage AI models, such as GPT-3, for generating responses, content, or performing analysis.

​
Implementing Steps

import os
from literalai import LiteralClient


client = LiteralClient(api_key=os.getenv("LITERAL_API_KEY"))

@client.step(type="tool")
def my_tool():
    # Implement your custom logic here
    return "Success"

# You can change the step type
@client.step(type="run")
def my_run():
    # Implement your custom logic here
    my_tool()
    return "Success"

with client.thread() as thread:
    client.message(content="Hello World", type="assistant_message", name="My Assistant")
    my_run()
    
# Network requests by the SDK are performed asynchronously.
# Invoke flush_and_stop() to guarantee the completion of all requests prior to the process termination.
# WARNING: If you run a continuous server, you should not use this method.
client.flush_and_stop()

​
Visualize the Steps on Literal

Output on the platform