Model Output Insights AI Agent

Overview

The Model Output Insights Agent helps users investigate and analyze Cosmic Frog model outputs by turning analytical questions into structured, data-backed strategic reports. It breaks down complex questions into a step-by-step exploration plan, executes targeted queries, synthesizes findings, and produces a professional report - complete with visualizations and actionable recommendations.

This documentation describes how this specific agent works and can be configured. Please see the “AI Agents: Architecture and Components” Help Center article if you are interested in understanding how the Optilogic AI Agents work at a detailed level.

Why It's Useful

Extracting meaningful insights from large databases typically requires exploring and analyzing many output tables which can take a lot of time and effort. The Model Output Insights Agent streamlines the process, helping users get to the insights quicker than ever before.

• Enhances productivity by automating complex research, analysis, and reporting tasks.
• Delivers high-quality, data-backed executive reports suitable for decision-makers.
• Adaptable to a wide range of analytical domains, from business strategy to technical investigations.

Key Capabilities

Structured Exploration

• Operates using a to-do list approach, breaking down analytical questions into discrete, manageable tasks.
• Progresses methodically by selecting and addressing one task at a time, ensuring thoroughness and clarity.

Deep Analytical Reasoning

• Utilizes the Analyzer skill (see table below) to query databases, analyze data, and synthesize findings.
• Provides evidence-based insights and actionable recommendations.

Executive Reporting

• When sufficient findings are gathered, it invokes the Report Builder skill (see table below) to generate comprehensive, professional-style executive reports.
• Reports include sections such as Executive Summary, Situation Assessment, Problem Statement, Analytical Framework, Methodology, Key Findings, Scenario Analysis, Data Summary, Insights & Recommendations, Validation, Strategic Implications, and Next Steps.

Iterative and Adaptive

• Addresses one to-do per turn, allowing for focused analysis and adaptability based on interim findings.
• Can conclude early if enough information is obtained, optimizing efficiency.

Main skills the Model Output Insights Agent uses:

Supporting capabilities:

How To Use It

The agent can be accessed through the Run AI Agent task in DataStar. Once a Run AI Agent task is added to the macro, first the Model Output Insights Agent needs to be selected from the list of available agents and utilities in the "Select Utility" section:

Next, the inputs and settings for the task can be specified in the Configure Utility, Run Configuration, and Notes sections:

1. Cosmic Frog Model Name - The model that the user wants to analyze.
2. Analysis Questions - This is where the user asks the agent what they would like to know from the model outputs. It can be as simple as "Give me cost and flow comparison between Scenario A and Scenario B", or more involved/a longer list of questions.‍
3. Knowledge Folder (optional) - Provide additional context for the agent to understand business background, modeling strategy, or even a request for a certain report style. The accepted file formats are md, csv, and txt. To use these files, we recommend creating a folder in the Explorer application, uploading the files to it, and entering the Folder Path as the Knowledge Folder input of the Run AI Agent Task. The screenshot below shows how to get a Folder Path.‍
4. Report File Name - This field is populated with "exploration_report" by default, but users can specify their own name for the report file.‍
5. Output Directory - Specify the folder name where the report and charts will be saved (default = Model Output Insights Agent). The system assumes that this is a folder name under My Files, so you do not need a full folder path here. This is different from what is needed for Knowledge Folder. For example, if the full folder path is /projects/My Files/Model Output Insights Agent_Report/ABC, only "Model Output Insights Agent_Report/ABC" is needed in this box
6. The run Configuration Section contains the following settings, which can optionally be set/adjusted:
   
   • Tags: Enter the tags for this task for easy filtering in the Run Manager application.
   • Timeout: The time allowed for the task to run until being stopped.
   • Resource Size: Different resource sizes offer different memory (RAM in Gb) and number of CPU cores to handle various task complexities. This only impacts ability to load the work into memory. The recommendation is to start with the default setting, monitor memory usage in the Run Manager application > Job Usage, and scale up if needed. The key principle is that a bigger resource does not always result in faster agent response time.
7. In the Notes section users can optionally add information about the task - especially useful when working in teams.

This next screenshot shows how to get a Folder Path while in the Explorer application: 1) right-click on the folder in the Explorer, 2) hover over Copy in the context menu, and 3) click on Folder Path:

After the run, a report in both markdown (.md) and pdf (.pdf) format and charts (if any) are created and can be found in the Explorer with the specified file name and folder. Once clicked, the file is opened in the Lightning Editor application for review.

Note that currently the charts are only included in the markdown file as a file name. Users can look for the charts in the Charts folder in the targeted output directory.

Other Helpful Notes

• If the Report File Name already exists in the Report Directory, a numbered suffix will be added to avoid overwriting the existing file - e.g., exploration_report (1).
• Runtime for the agent varies based on the amount of data to analyze and the complexity of the question(s). Expect at least 10 minutes of runtime.
• Just like many other DataStar tasks, it is possible to run multiple tasks in parallel with the Model Output Insights Agent.
• Additional info on the run can be found in the Task Logs tab underneath the Macro Canvas. This includes steps that the agent takes, tools it calls, as well as a work summary. The AI Response sections are typically the most useful as they explain the exploration plan, the work it has done, and the results after exploration. This is generally a response to the user, while all others are more about internal processes.

• The following screenshot shows part of a report instructions file, which can be placed in the Knowledge Folder specified in a Run AI Agent task that uses the Model Output Insights Agent. It contains info on target audience for the report, expected report length and tone. If you would like to review it and optionally use as a starting point for your own usage with this Agent, you can download it here. After downloading, please rename the .txt extension to .md. You can then upload it to your Optilogic account using the Explorer application and then view it in the Lightning Editor application.

Other Helpful Resources

• Learn more about Cosmic Frog
• Learn more about DataStar
• An overview of Optilogic's AI Agents and Utilities
• The Data Cleansing AI Agent
• The Data Profiler AI Agent
• The Modeler AI Agent