Application Workflow

Learn how to design and manage agentic application workflows. Improve traceability and reliability in testing.

Application Workflow Overview

The Application Workflow provides a detailed view of the complete execution flow of a dataset datapoint, showcasing all the interactions, inputs, outputs, and operations executed during the process. This feature captures a step-by-step trace of how the application processes data, including interactions with tools, models, APIs, and files.

Interaction Types

The Application Workflow tracks the following interaction types:

• agent_call_start: Indicates the start of an agent's execution.

• agent_call_end: Marks the completion of an agent's execution.

• input: Logs user or system-provided input to the application.

• output: Captures the final output generated by the application.

• tool_call_start: Represents the initiation of a tool or external service call.

• tool_call_end: Signals the completion of the tool or external service call.

• llm_call: Logs the interaction with a Language Model (LLM) during execution.

• file_read: Records any file-reading operations performed by the application.

• file_write: Logs file-writing operations carried out by the application.

• network_call: Tracks external API or network requests made by the application.

How to View the Application Workflow

1. Open a Dataset:

   • Navigate to the dataset where the datapoint is located.

2. Select a Datapoint:

   • Click on a specific datapoint within the dataset to open the trace view.

3. Access the Application Workflow:

   • In the trace view, locate and click on the "Application Workflow" tab to view the detailed flow of interactions.

   • User can click on a interaction and it will open the span and view the complete trace.