Getting Started with Leapfrog AI

Leapfrog helps Cosmic Frog users explore and use their model data via natural language. View data, make changes, create & run scenarios, analyze outputs, learn all about the Anura schema that underlies Cosmic Frog models, and a whole lot more!

Leapfrog combines an extensive knowledge of PostgreSQL with the complete knowledge of Optilogic’s Anura data schema, and all the natural language capabilities of today’s advanced general purpose LLMs.

For a high-level overview and short video introducing Leapfrog, please see the Leapfrog landing page on Optilogic’s website.

In this documentation, we will first get users oriented on where to find Leapfrog and how to interact with it. In the section after, Leapfrog’s capabilities will be listed out with examples of each. Next, the Tips & Tricks section will give users helpful pointers so they can get the most out of Leapfrog. Finally, we will step through the process of building, running, and analyzing a Cosmic Frog model start to finish by only using Leapfrog!

Dive in if you’re ready to take the leap!

Interacting with Leapfrog

Getting Started

Start using Leapfrog by opening the module within Cosmic Frog:

1. We are in Optilogic’s Cosmic Frog application.
2. Click on the Module Menu icon (3 horizontal bars) to open the list of Cosmic Frog modules.
3. Select Leapfrog from the list.

Once the Leapfrog module is open, users’ screens will look similar to the following screenshot:

1. While inside the Leapfrog module, users will see the jumping frog icon at the top, to the right of the Module Menu icon.
2. Leapfrog will greet you with a short introduction of how it works and how it can help users. This message is shown as the first one in each new conversation between Leapfrog and the user.
3. User can add questions / prompts to the conversation by typing in the “Enter a question” free type text box at the bottom of the screen. Below the question box, user can select the large language model they want to use to answer the question. Currently there are 2 models available to choose from which are explained in more detail in the next section “Leapfrog’s Large Language Models”. As the welcome message for new conversations also indicates, choose the one to use as follows:
   1. Text2SQL: use this for hands-on data work, such as performing analysis on input or output data and model operations. The responses will mostly be either SQL queries which user can execute or optionally run tasks such as creating and running models & scenarios. Hovering over the button also shows a brief description of the Text2SQL LLM.
   2. Anura Help (also referred to as Anura Aficionado or A2): use this for schema guidance. Anura Help can explain table structures, column relationships, validation rules, and which components work with different Cosmic Frog engines. Responses will be text-based and will include context. Hovering over the button also shows a brief description of the Anura Help LLM.
4. After typing in a prompt, hit the Enter key or click on the send icon on the right to submit it to Leapfrog.
5. Several example prompts are shown here to help users get started with Leapfrog. The ones for the Text2SQL LLM are shown here, those for the Anura Help LLM are shown in the next screenshot. Clicking on an example prompt will submit the prompt to Leapfrog and Leapfrog will then respond to it.
   1. The special “Prompt Library” prompt has a different font color and will take users to the Prompt Library on the Frogger Pond community in a new tab of the browser. Users can use this to get ideas for what types of prompts can be answered, so they can get more out of Leapfrog.
6. Under the question mark users can find following 3 resources to learn more about Leapfrog:
   1. Show Screen Tips – this will highlight parts of the Leapfrog user interface while explaining Leapfrog functionality; user can step through the available tips by clicking on the Next and Back buttons.
   2. Open Leapfrog Help – this will open this Help Center article in a new tab of the browser.
   3. Prompt Library – this will open the Prompt Library on the Frogger Pond community in a new tab of the browser. Users can use this to get ideas for what types of prompts can be answered, so they can get more out of Leapfrog.

The example prompts when using the Anura Help LLM are shown here:

1. We are using the Anura Help LLM here, which can be seen from the color of the LLM toggle: the active one is blue.
2. The example prompts give users an idea of the types of questions to ask Anura Help. Again, the Prompt Library special prompt is available here too, recognizable by the different font color. Clicking on it will take users to the Prompt Library on the Frogger Pond community.

When first starting to use Leapfrog, users will also see the Privacy and Data Security statement, which reads as follows:

“Leapfrog AI Training: Optilogic does not use your model data to train Leapfrog. We do collect and store conversational data so it can be accessed again by the user, as well as to understand usage patterns and areas of strength/weakness for the LLM. Included in this data: natural language input prompts, text and SQL responses, as well as feedback from users. This information is maintained by Optilogic, not shared with third parties, and all of the conversation data is subject to the data security and privacy terms of the Optilogic platform.”

This message will stay visible within Leapfrog whenever it is being used, unless user clicks on the grey cross button on the right to close the message. Once closed, the message will not be shown again while using Leapfrog.

Conversation history is stored on the platform at the user level – not in the model database – so it does not get shared when a model is shared. Note that if you are working in a Team rather than in your My Account (see documentation on Teams on the Optilogic platform here), the Leapfrog conversations you are creating will be available to the other team members when they are working with the same model.

Leapfrog’s Large Language Models

As mentioned in the previous section, Leapfrog currently makes use of 2 large language models (LLMs): Text2SQL and Anura Help (also referred to as Anura Aficionado or A2). They will be explained in some more detail here. There is also an appendix to this documentation where for a few example personas Leapfrog questions and responses are listed, which showcases how some users may predominantly use one model, while others may switch back and forth between them. Of course, when unsure, users can try a specific prompt using both LLMs to see which provides the most helpful response.

Please note that in future users will not need to indicate which LLM they want to run a prompt against as Leapfrog will recognize which one will be most suitable to use based on the prompt.

Text2SQL LLM

The Text2SQL LLM combines extensive knowledge of PostgreSQL with Optilogic’s Anura data schema, and all the natural language capabilities of today’s advanced general purpose LLMs. It has been further fine-tuned on a large set of prompt-response pairs hand-crafted by supply chain modeling experts. This allows the Text2SQL model to generate SQL queries from natural language prompts.

Prompts for which it is best to use the Text2SQL model often imply an action: “Show me X”, “Add Y”, “Delete Z”, “Run scenario A”, “Create B”, etc. See also the example prompts listed when starting a new conversation and those in the Prompt Library on the Frogger Pond community.

Leapfrog responses using this model are usually actionable: run the returned SQL query to add / edit / delete data, create a scenario or model, run a scenario, geocode locations, etc.

A full list of the capabilities of both LLMs is covered in the section “Leapfrog Capabilities” further below.

Anura Help LLM

Anura Help (also referred to as Anura Aficionado or A2) is a specialized assistant that leverages advanced natural language processing to help users navigate and understand the Anura schema within Optilogic’s Cosmic Frog application. The Anura schema is the foundational framework powering Cosmic Frog’s optimization, simulation, and risk assessment capabilities. Anura Help eliminates traditional barriers to schema understanding by providing immediate, authoritative guidance for supply chain modelers, developers, and analysts.

Anura Help’s architecture uses the Retrieval Augmented Generation (RAG) approach: based on the natural language prompt, first the most relevant documents of those in its knowledge base are retrieved (e.g. schema details or engine awareness details). Next, it uses them to generate a natural language response.

Use the Anura Help model when wanting to learn about specific fields, tables or engines in Cosmic Frog. Its core capabilities include:

• Anura schema understanding:
  • Table details: provides descriptions, purposes, and complete column listings for all tables.
  • Column details: explains the usage, data types, and validation rules for individual columns.
  • Relationships: maps connections between tables and their dependencies.
  • Technical requirements: details primary keys, required fields, and default values.
  • SCG model conversion: maps fields from SCG models to the Anura schema.
  • Anura versioning: provides schema version and change details.
• System integration:
  • Engine awareness: identifies which tables and columns are used by different Cosmic Frog engines.
  • Template models: provides information about various Cosmic Frog template models.

Responses from Leapfrog when using the Anura Help model are text-based and generated from retrieved documents shown in the context section. This context can for example be of the category “column info” where all details for a specific field are listed.

A full list of the capabilities of both LLMs is covered in the section “Leapfrog Capabilities” further below.

LLM Comparison

The following table shows a side-by-side comparison of the 2 LLMs available in Leapfrog today:

	Text2SQL	Anura Help
LLM description	Combines PostgreSQL with the Anura data schema and natural language capabilities. Fine-tuned on prompt-response pairs created by supply chain modelling experts.	Expert on the Anura data schema with access to expansive knowledge base of 5k+ documents. Uses RAG approach to retrieve relevant documents and combines these with natural language capabilities to formulate response.
Core capabilities	Producing SQL and several other actionable modeling tasks from text.	Anura schema understanding, system integration awareness.
Types of Prompts (Questions)	Perform actions on model inputs / outputs & modeling tasks. (“Show me…”, “Add…”, “Delete…”, “Change X to Y”, “Increase/decrease A by B”, “Create…”, “Run…”.)	Understanding the schema, engines, system integration. (“What are the valid options for X field?”, “Which tables are required to run Y engine?”.)
Types of Responses (Answers)	Actionable: SQL (Insert, Delete), SQL Select to show data. Create scenarios / groups, creating & running models, geocoding. Text.	Text, with context that can be expanded.
Example Prompt	Increase transportation costs by 5%	Which engines use the Unit Cost field on the Transportation Policies table?
Example Response	SQL Query: UPDATE TransportationPolicies SET UnitCost = UnitCost::DOUBLE PRECISION * 1.05 Scenarios: Option to create a new scenario containing a scenario item that multiplies the Unit Cost field on the Transportation Policies table by 1.05	Text: The Unit Cost field on the Transportation Policies table is used by the following engines: NEO, THROG. Context: Provides 5 documents for context all of type Column Info for 1) Unit Cost field on Transportation Policies table, 2) Unit Cost field on Transportation Policies Multi-Time Period table, 3) Unit Cost UOM field on the Transportation Policies table, 4) Unit Cost field on the Transportation Modes table, 5) Unit Cost field on the Transportation Rates table
Visual Difference	Responses have a Frog avatar and a white background	Responses have a Frog avatar with “A2” written beneath and a purple background

 

Leapfrog Responds to Questions

Depending on the type of question, Leapfrog’s response to it can take different forms: text, links, SQL queries, data grids, and options to create models, scenarios, scenario items, groups, run scenarios, or geocode locations. We will look at several examples of questions that result in these different types of responses in this section. This is not an exhaustive list; the next section “Leapfrog Capabilities” will go through the types of prompt-response pairs Leapfrog is capable of today.

For our first question, we used the first Text2SQL example prompt “What are the top 3 products by demand?” by clicking on it. After submitting the prompt, we see that Leapfrog is busy formulating a response:

1. The prompt that was submitted to Leapfrog.
2. While Leapfrog is busy formulating a response, this “Creating new message…” message will be showing.
3. We can also tell Leapfrog is busy answering by the spinning wheel and the “Sending…” text in the question box.
4. We can tell the prompt was submitted to the Text2SQL LLM as that one is colored blue.

And Leapfrog’s response to the prompt is as follows:

1. The example prompt that was submitted to Leapfrog’s Text2SQL model.
2. The top part of Leapfrog’s response shows a SQL Query, which can answer the question that was posed. By default, the SQL Query is expanded, so the SQL statement is visible. User can collapse the SQL statement by clicking on the button with the arrowhead pointing up. In this case, the SQL Statement uses a SELECT query, applied to the Customer Demand input table where it sums the Quantity field for each unique Product Name. The result is sorted by this summed Total Quantity in descending order (high to low), and the top 3 products are then filtered out by using the Limit 3 clause.
3. If user wants to use the SQL Statement elsewhere, it can be copied to the clipboard by clicking on the button with 2 squares at the bottom of the query.
4. When prompts result in a response from Leapfrog in the form of a SQL SELECT statement, a Data Grid is shown beneath the SQL Query. This Data Grid shows the result of the SQL Query, and it is by default expanded too. It can be collapsed by clicking on the button with the arrowhead pointing up. Please note that this preview of the data grid is limited to 50 rows.
5. There are several options for the user on how to use the data grid by clicking on one of the 3 icons at the bottom of it. These perform following actions, from left to right: export the data grid to an .xlsx Excel file, export the data grid to a .csv file, and save the data grid as a table or view. These options are explained in more detail in the section “Data Grid Options” further below.
6. For Leapfrog’s continuous improvement, it is helpful when users rate Leapfrog’s responses by clicking on the thumbs up or thumbs down buttons at the top right of a response, and, optionally, include more details in this feedback. Giving Leapfrog feedback is explained in more detail in the section “Feedback for Continuous Improvement” further below.
7. Clicking on the icon with 3 dots opens up an additional section at the bottom named “Response Metadata”:

The metadata included here are:

1. Model – which of Optilogic’s LLM models was used to generate the response. This will say Leapfrog when Text2SQL was used and A2 when Anura Help was used.
2. Version – the version of the model that was used to generate the response at the time of response generation.
3. Timestamp – the time and date the response was generated.

Clicking on the icon with 3 dots again will collapse the response metadata.

This first prompt asked a question about the input data contained in the Cosmic Frog model. Let us now look at a slightly different type of prompt, which asks to change model input:

1. The prompt asks Leapfrog to Increase demand by 20%. The Text2SQL model is used to run this prompt.
2. This prompt results in an UPDATE SQL Statement, where the Quantity field on the Customer Demand input table will be set to its current value multiplied by 1.2, resulting in the asked for 20% increase.
3. Prompts that result in actionable SQL queries (i.e. UPDATE, INSERT, and DELETE Statements) will show a Run SQL button at the bottom of the SQL Query section of the response. If user wants to make this change in the input table directly (changing this table permanently), user can click on the Run SQL button to perform this action. See the next screenshots below which step through this, including looking at the input table before and after the query is run.
4. For this type of prompt that results in an UPDATE SQL query, a Scenarios section is also part of the response, giving user the option to create a scenario that makes the change to the input data when that specific scenario is run, rather than making the change permanently in the master data in the input table. We will explore how this works below after walking through the Run SQL option.

We are going to run the SQL query of the above response to our “Increase demand by 20%” prompt. Before doing so, let’s review a subset of 10 records of the Customer Demand input table (under the Data Module, in the Input Tables section):

Next, we will run the SQL query:

1. Click on the Run SQL button.
2. In the message confirming user wants to run the SQL query against the model that comes up, user can opt to cancel out if deciding not to make this permanent change to the Quantity field on the Customer Demand input table.
3. User can click on the Run SQL button in the message when ready to make this change.

After clicking the Run SQL button at the bottom of the SQL Query section in Leapfrog’s response, it becomes greyed out so it will not accidentally be run again. Hovering over the button also shows text indicating the query was already run:

Note that closing and reopening the model or refreshing the browser will revert the Run SQL button’s state so it is clickable again.

Opening the Customer Demand table again and looking at the same 10 records, we see that the Quantity field has indeed been changed to its previous value multiplied by 1.2 (the first record’s value was 643, and 643 * 1.2 = 771.6, etc.):

Running the SQL query to increase the demand by 20% directly in the master data worked fine as we just saw. However, if we do not want to change the master data, but rather want to increase the demand quantity as part of a scenario, this is possible too:

1. We are looking at the same “Increase demand by 20%” prompt and response pair again.
2. The SQL Query section has been collapsed now and can be expanded by clicking on the button with the arrowhead pointing down.
3. The Scenarios section has been expanded here. It contains details for creating a new scenario in the Cosmic Frog model based on the prompt:
4. A name for the new scenario to be created has been auto generated based on the prompt and is called “increase_customer_demand_by_20_percent”.
5. Similarly, a name for the new scenario item that will be assigned to the new scenario has been auto generated too and it is called “increase_quantity_by_20_percent”.
6. The details of the scenario item are listed here:
   1. The name of the table to which the change is made: CustomerDemand.
   2. The action that represents the change made: quantity = quantity * 1.2.
   3. The filter condition that determines which records of the table the action will be applied to: (None), meaning that the change is made to all records in the table.
7. User can click on the Create Scenario button to create the scenario and item using these names and details. After clicking on the Create Scenario button, it becomes greyed out and the hover over text also indicates that the scenario has already been created:

After navigating to the Scenarios module within our Cosmic Frog model, we can see the scenario and its item have been created:

1. There is now a scenario named “increase_customer_demand_by_20_percent” in our list of scenarios.
2. This scenario has 1 item assigned to it, named “increase_quantity_by_20_percent”. The details of which match those listed out in the Scenarios section of the Leapfrog response:
   1. Table Name is set to CustomerDemand.
   2. The actions change the quantity field to its current value multiplied by 1.2.
   3. There are no conditions for the scenario item, so the actions are applied to all records in the Customer Demand table.

Note that if desired, the scenario and scenario item names auto-generated by Leapfrog can be changed in the Scenarios module of Cosmic Frog: just select the scenario or item and then choose “Rename” from the Scenario drop-down list at the top.

As a final example of a question & answer pair in this section, let us look at one where we use the Anura Help LLM, and Leapfrog responds with text plus context:

1. The prompt used here asks the Anura Help LLM to explain optimization policy options on the Transportation Policies table.
2. Note the Leapfrog avatar in the response has A2 written underneath it to indicate the LLM used was the Anura Help one. The background color of the response is also different (purple) as compared to responses by the Text2SQL LLM, which have a white background. These are visual reminders for users to know which LLM was used to respond to the prompt.
3. This is the text-based answer from the Anura Help model explaining the 3 options of what the Optimization Policy field on the Transportation Policies table can be set to.
4. This bottom section is labelled “Context”, and this can be expanded by clicking on the button with the arrowhead pointing down. This will then show the information that was retrieved from Anura Help’s knowledge base in order to help generate the response:

1. The Context section is expanded and can be collapsed again by clicking on the button with the arrowhead pointing up.
2. There are 5 sections with information from which the response was generated and each can be expanded / collapsed individually by clicking on the buttons with the arrowheads pointing down / up. The information sections are labelled with the information category. Here we see 2 of them: Table Info (2a) and Column Info (2b).
3. The Column Info for the Optimization Policy field on the Transportation Policies table seems interesting to us, so we click to expand this section, which will then show following details:

There is a lot of information listed here; we will explain the most commonly used information:

1. Table – in which table is this field located.
2. Field – the name of the field in question.
3. Type – is the table an input or an output table.
4. Required – is the field required to be populated (if not and the field is blank, the default value will be used).
5. Engine – which Cosmic Frog engine(s) use this field. To understand where the engine names come from, see this Help Center article.
6. Purpose and Usage – a text explanation of the field of what value(s) the field can contain and how that impacts the model.
7. Data Type – what type of data should be entered into the field. If it is a list of values in between square brackets, separated by commas (like it is here), it means these are the allowable values and in this case the field in Cosmic Frog will show a drop-down list with these options when user puts the cursor in this field.
8. Default value – if the field is left blank, this is the value that will be used.

Leapfrog Conversations

Prompts and their responses are organized into conversations in the Leapfrog module:

1. We are in the Leapfrog module, looking at the list of Conversations.
2. All conversations that were created within the currently active Cosmic Frog model are listed here. By default, they are named the same as the first prompt in the conversation.
3. When hovering over a conversation, text will pop up telling the user how many prompts and responses that specific conversation contains.
4. Use the Search box to quickly find conversations that have certain text in their name.
5. The conversations can be sorted by clicking on this icon with the 3 horizontal bars. Users can choose from sorting in the Default order, which lists the conversations in the order in which they were created, or to sort them alphabetically by conversation name (in ascending order when first selected, in descending order if clicked on again).

Users can organize their conversations with Leapfrog by using the options from the Conversations drop-down at the top of the Leapfrog module:

1. The Conversations drop-down menu, which has following options:
   1. Choose the New option to start a new conversation. Users may want to do this to keep conversations shorter and organized. One could for example have a conversation that is only about querying output data, another conversation that revolves around creating new scenarios, and a third one that focuses on creating views to be used for Analytics dashboards. This way user does not need to scroll through one very long conversation to return to specific prompts.
   2. Rename can be used to change the names of conversations, to for example be more concise and more descriptive of the whole conversation rather than just referring to the first prompt of the conversation.
   3. To keep things organized, conversations that were not giving the desired responses or those that have become irrelevant can be deleted by using this Delete option.

Feedback for Continuous Improvement

Users can rate Leapfrog responses by clicking on the thumbs up (like) and thumbs down (dislike) buttons and, optionally, providing additional feedback. This feedback is used to continuously improve Leapfrog. Giving a thumbs up to indicate the response is what you expected helps reinforce correct answers from Leapfrog. When a response is not what was expected or wrong, users can help improve Leapfrog’s underlying LLMs by giving the response a thumbs down. Especially thumbs down ratings & additional feedback will be reviewed so Leapfrog can learn and become more useful all the time.

When a response is not as expected as was the case in the following screenshot, user is encouraged to click the thumbs down button:

1. This prompt asks the Anura Help LLM which tables are required to run network optimization.
2. Leapfrog responds with a list of 2 output and 2 input tables. However, user expected a list of only input tables and more than the 2 that are listed.
3. User has clicked on the thumbs down (Dislike) button, which is now a darker shade of blue and increased in size. This ‘dislike’ feedback is submitted immediately after clicking the thumbs down button.
4. After clicking the thumbs down button, a form where detailed feedback can optionally be given is opened. There are 5 tags here that user can click to indicate why the response was given a thumbs down (note that multiple tags can be selected if desired):
   1. Table/Field Issue: use this when Leapfrog’s SQL response was targeting a different table and/or field than user intended or the text based response covered tables/fields other than those expected.
   2. Query Type: use this when Leapfrog’s SQL response was of a different query type than expected (e.g. UPDATE instead of DELETE).
   3. Incomplete: use this when any part of the response seems incomplete.
   4. Slow Response: use this if Leapfrog took an unusually long time to provide a response.
   5. Other: use this if the other 4 tags do not cover the issue user sees with the response.
5. In the “Type your feedback here” textbox, user can type in a brief description of why the response was given a thumbs down (or thumbs up).
6. If user decides not to send the detailed feedback, they can click on the Close button.
7. When user has selected the appropriate tag(s) and typed in their feedback, they can submit it by clicking on the Send button. Note that user needs to type something in the “Type your feedback here” textbox for the Send button to become active.

After clicking on the Send button, the detailed feedback is automatically added to the conversation:

1. A message thanking user for the feedback is shown and the thumbs down icon is shown in a darker shade of blue while also having been increased in size.
2. The tag(s) and text of the detailed feedback are added to the conversation and labelled “Feedback”.

The next screenshot shows an example where user gave Leapfrog’s response a thumbs up as it was what user expected. This feedback can then be used by Leapfrog to reinforce correct answers. User also had the option to provide detailed feedback again, using any of the following 4 optional tags: Showcase Example, Surprising, Fun, and Repeatable Use Case. In this example, user decided not to give detailed feedback and clicked on the Close button after the detailed feedback form came up:

If you have any additional Leapfrog feedback (or questions) beyond what can be captured here, you can feel free to send an email to Leapfrog@Optilogic.com. You are also very welcome to ask questions, share your experiences, and provide feedback on Leapfrog in the Frogger Pond Community.

Data Grid Options

We will now go back to our first prompt “What are the top 3 products by demand” to explore some of the options users have when Data Grids are included in a Leapfrog response, which is the case when Leapfrog’s SQL Query response is a SELECT statement.

1. Clicking on the grid icon exports the data grid to an .xlsx Excel file. Following 2 messages will appear below the Data Grid when exporting the Data Grid, these can be closed by clicking on the grey cross on the far right in the messages:

When clicking on the Download File button, a zip file with the name of the active Cosmic Frog model appended with an ID, is downloaded to the user’s Downloads folder. The zip contains:

1. 1. An .xlsx file of the same name as the zip file. This Excel file contains a query_results worksheet which contains the data that was the result of the SQL Query of the Leapfrog response in Cosmic Frog. This is the same data as seen in the Data Grid, or possibly more if the result is more than 50 rows (the Data Grid preview is limited to 50 rows). The other worksheets in this Excel file contain the SQL Query on the source_query worksheet and the mapping of table or view names to worksheet names on the Sheet Mapping worksheet.
   2. A README.txt file with the same name as the zip file – this file contains notes about Excel limitations regarding maximum number of rows and maximum number of characters for worksheet names and how these are handled.
   3. A subfolder with the same name as the zip file which contains 3 more files. These are the same files that will be downloaded when user chooses to Export to CSV (see next bullet, #2):
      1. metadata.txt – contains the log messages of the export.
      2. query_results.csv – contains all results of the query.
      3. source_query.sql – contains the SQL Statement that was used to generate the data in the query_results.csv, the same SQL SELECT statement seen in Leapfrog’s response.

2. Clicking on the icon with the comma exports the data grid to a .csv file. The same 2 messages as shown above for exporting Excel files will appear. Once the file is created and user has clicked on the Download File button, a zip file with the name of the active Cosmic Frog model appended with an ID, is downloaded to the user’s Downloads folder. This folder contains 3 files, which are the same as those listed under bullet 1.c above.
3. Clicking on the disk icon allows user to save the data grid as a new table or view within the active Cosmic Frog model’s database. It brings up the following form on the right-hand side where user can choose to Save as either a Table or View, give the new table/view a Name and then click on Save to save the table or view.

After clicking on Save, following message appears beneath the Data Grid in Leapfrog’s response:

Looking in the Custom Tables section (#2 in screenshot below) of the Data module (#1 in screenshot below), we indeed see this newly created table named top3products (#3 in screenshot below) with the same contents as the Data Grid of the Leapfrog response:

If we choose to save the Data Grid as a view instead of a table, it goes as follows:

We choose Save as View and give it the name of Top3Products_View. The message that comes up once the view is created reads as follows:

Going to the Analytics module in Cosmic Frog, choosing to add a new dashboard and in this new dashboard a new visualization, we can find the top3products_view in the Views section:

We will go back to the original Data Grid in Leapfrog’s response to explore a few more options user has here:

1. The Data Grid previews in Leapfrog responses have some of the same options as the input, output, and custom tables in the Data module of Cosmic Frog. When clicking on the 3 vertical dots, the available options are shown:

1. 1. Pin column: option to fix the location of the column, either left or right. By default, columns are not pinned (the No Pin option).
   2. Autosize This Column: resizes the width of the selected column to fit the column heading and column values.
   3. Autosize All Columns: resizes the width of all columns to fit the column heading and column values.
   4. Choose Columns: lets user select which columns to show and which to hide.
   5. Reset Columns: undo any resizing, reordering and custom selection of columns to revert to the original grid layout and contents.

2. Users can manually resize columns by clicking on the vertical bar to the right of the icon with 3 vertical dots and then dragging left or right.

Please note:

• The Data Grid preview within Leapfrog responses is limited to 50 rows. When exporting to Excel/CSV or saving as a table/view, all records that result from the SQL SELECT statement will be included.
• A Custom Table created by saving a Data Grid as a table is a static snapshot of the data represented by the SELECT at the time the table is created and will not change if data changes in the model. Conversely, a View saves the SQL statement so any time you access that View it will pull a live, up-to-date picture of the data represented by the SELECT statement.

Leapfrog Capabilities

In this section we will list out what Leapfrog is capable of and give examples of each capability. These capabilities include (the LLM each capability applies to is listed in parentheses):

1. Querying data in the input and output tables (Text2SQL)
2. Changing data in the input tables directly (Text2SQL)
3. Creating scenarios and scenario items (Text2SQL)
4. Creating groups, and adding group members (Text2SQL)
5. Creating new models (Text2SQL)
6. Running models (Text2SQL)
7. Geocoding locations (Text2SQL)
8. Expert guidance on Anura Schema (Anura Help)
9. Systems integration knowledge (Anura Help)
10. Leapfrog is self-aware (Text2SQL, Anura Help)
11. Multi-language support (Text2SQL, Anura Help)

Each of these capabilities will be discussed in the following sections, where a brief description of each capability is given, several example prompts illustrating the capability are listed, and a few screenshots showing the capability are included as well. Please remember that many more example prompts can be found in the Prompt Library on the Frogger Pond community.

Querying Data (Text2SQL)

Interrogate input and output data using natural language. Use it to check completeness of input data, and to summarize input and/or output data. Leapfrog responds with SELECT Statements and shows a Data Grid preview as we have seen above. Export the data grid or save it as a table or view for further use, which has been covered above already too.

Example prompts:

• Are there any facilities, customers or suppliers that are not geocoded?
• Show me any product names used in the Bills of Materials table that are not specified as product names in the Products table.
• Show me the lowest cost scenario and any other scenarios within 5% of that lowest cost.
• Which largest customers account for 80% of the demand quantity?
• What is the average distance for flows from Assembly locations to DCs?
• Show me work centers that have utilization below 30% during all periods of the model.
• Which facilities have throughput utilization above 90% during any period of the model?
• How many routes does each scenario have?
• What is the lowest cost Greenfield scenario?

The following 3 screenshots show examples of checking input data (first screenshot), and interrogating output data (second and third screenshot):

Changing, Adding, and Removing Data (Text2SQL)

Tell Leapfrog what you want to edit in the input data of your Cosmic Frog model, and it will respond with UPDATE, INSERT, and DELETE SQL Statements. User can opt to run these SQL Queries to permanently make the change in the master input data. For UPDATE SQL Queries, Leapfrog’s response will also include the option to create a scenario and scenario item that will make the change, which we will focus on in the next section.

Example prompts:

• Increase demand by 15%.
• Exclude all Suppliers that are based in Europe.
• Remove all Facilities that are located in the USA.
• Add a transportation policy from the group DCs to the group Customers.
• On the DCs to Customers transportation policy set unit cost to 0.8.

The following 3 screenshots show examples changing values in the input data (first screenshot), adding records to an input table (second screenshot), and deleting records from an input table (third screenshot):

Creating Scenarios and Scenario Items (Text2SQL)

Make changes to input data, but through scenarios rather than updating the master tables directly. Prompts that result in UPDATE SQL Queries will have a Scenarios part in their responses and users can easily create a new scenario that will make the input data change by one click of a button.

Example prompts:

• Change the inventory carrying cost percentage in the Model Settings table to 12.
• Include processes that have “Include Expansion 2027” in the Notes field.
• Exclude all inventory policies.
• Decrease demand by 20% and increase sourcing costs by 5%.

The following 3 screenshots show example prompts with responses from which scenarios can be created: create a scenario which makes a change to all records in 1 input table (first screenshot), create a scenario which makes a change to records in 1 input table that match a condition (second screenshot), and create a scenario that makes changes in 2 input tables (third screenshot):

The above screenshots show examples of Leapfrog responses that contain a Scenarios section and from which new scenarios and scenario items can be created by clicking on the Create Scenario button. In addition to the above, users can also use Leapfrog to manage scenarios by using prompts that specifically create scenarios and/or items and assigning specific scenario items to specific scenarios. These result in INSERT INTO SQL Statements which can then be implemented by using the Run SQL button. See the following 2 screenshots for examples of this, where 1) a new scenario is created and an existing scenario item is then assigned to it, and 2) a new scenario item is created which is then assigned to an already existing scenario:

Creating Groups and Adding Group Members (Text2SQL)

Leapfrog can create new groups and add group members to new and existing groups. Just specify the group name and which members it needs to have in the prompt and Leapfrog’s response will be one or multiple INSERT INTO SQL Statements.

Example prompts:

• Add all Facilities with FacilityStatus = Consider to a new group named Candidate_Facilities.
• Create a new Group named TruckModes and add Modes that contain Truck in their names to it.
• Add RM10 to the RAW_MATERIALS group.
• Create a group “TEXAS_Customers” and add customers located in Texas to it.

The following 4 screenshots show example prompts of creating groups and group members: 1) creates a new products group and adds products that have names with a certain prefix (FG_) to it, 2) creates a new periods group and adds 3 specific periods to it, 3) creates a new suppliers group and adds all suppliers that are located in China to it, and 4) adds a new member to an existing facilities group, and in addition explicitly sets the Status and Notes field of this new record in the Groups table:

Creating New Models (Text2SQL)

Leapfrog can create a new, blank model. Leapfrog’s response will ask user to confirm if they want to create the new model before creating it. If confirmed, the response will update to contain a link which takes user to the Leapfrog module in this newly created model in a new tab of the browser.

Example prompts:

• Create a new model named “Simulation Test 1”
• Start a new blank model
  • In this case Leapfrog will give it a default name like NewModel

Following 2 screenshots show an example where a new model named “FrogsLeaping” is created:

1. The user’s prompt asks Leapfrog to create a new model named “FrogsLeaping”.
2. Leapfrog’s response asks user to confirm by clicking on the Create button. After the Create button is clicked, the response changes to say the model is being created and that it can be accessed by clicking on the word “here”:

Running Models (Text2SQL)

You can ask Leapfrog to kick off any model runs for you. Optionally you can specify the scenario(s) you want to be run, which engine to use, and what resource size to use. For Neo (network optimization) runs, user can additionally indicate if the infeasibility check should be turned on. If no scenario(s) are specified, all scenarios present in the model will be run. If no engine is specified, the Neo engine (network optimization) will be used. If no resource size is specified, S will be used. If for Neo runs it is not specified if the infeasibility check should be turned on or off it will be off by default.

Leapfrog’s response will summarize the scenario(s) that are to be run, the engine that will be used, the resource size that will be used, and for Neo runs if the infeasibility check will be on or off. If user indeed wants to run the scenario(s) with these settings, they can confirm by clicking on the Run button. If so, the response will change to contain a link to the Run Manager application on Optilogic’s platform, which will be opened in a new tab of the browser when clicked. In the Run Manager, users can monitor the progress of any model runs.

The engines available in Cosmic Frog are:

• Neo – Network Optimization
• Infeasibility – infeasibility diagnosis for Neo
• Throg – Simulation
• Triad – Greenfield
• Hopper – Transportation Optimization
• Dendro – Inventory Optimization

The resource sizes available are as follows, from smallest to largest: Mini, 4XS, 3XS, 2XS, XS, S, M, L, XL, 2XL, 3XL, 4XL, Overkill. Guidance on choosing a resource size can be found here.

Example prompts:

• Run my model.
  • Please note that this will run all scenarios present in the model.
• Please run the Baseline scenario using the Hopper engine and a resource size of XS.
• Please run Greenfield on the 7 Optimal Facilities scenario.

The following 2 screenshots show an example prompt where the response is to run the model: only the scenario name is specified in which case a network optimization (Neo) is run using resource size S with the infeasibility check turned off (False):

1. The prompt where user ask to run 1 specific scenario.
2. Leapfrog’s response summarizes the scenario(s) to be run, the engine to be used, and the resource size to be used.
3. If user wants to go ahead with the run, they can click on the Run button. If clicked, the response changes as follows:

Leapfrog’s response now indicates the run has been kicked off and provides a link (click on the word “here”) to check the progress of the scenario run(s) in the Run Manager.

The next screenshot shows a prompt asking to specifically run Greenfield (Triad) on 2 scenarios, where the resource size to be used is specified in the prompt too:

The last screenshot in this section shows a prompt to run a specific scenario with the infeasibility check turned on:

Geocoding Locations (Text2SQL)

Leapfrog can find latitude & longitude pairs for locations (customers, facilities, and suppliers) based on the location information specified in these input tables (e.g. Address, City, Region, Country). Leapfrog’s response will ask user to confirm they want to geocode the specified table(s). If so, the response will change to contain a link which will open a Cosmic Frog map showing the locations that have been geocoded in a new tab of the browser.

Example prompts:

• Geocode all locations.
• Geocode my customers.
• Geocode Facilities and Suppliers.

Notes on using Leapfrog for geocoding locations:

• Only locations where either the latitude value or the longitude value or both are blank will be geocoded. Locations where latitude = longitude = 0 will not be geocoded and neither will locations which already have values for both latitude and longitude.
• Currently, it is not possible to specify that only a subset of locations in a table should be geocoded.

In the following screenshot, user asks Leapfrog to geocode customers:

1. The prompt asking to geocode the customers in the model.
2. Leapfrog summarizes the table(s) to be geocoded and asks user to confirm by clicking on the Geocode button. When clicked, Leapfrog’s response changes to say that the geocoding is in progress and will now include a link (click on the word “here”) which will open a new tab in the browser to show the geocoded locations.

As geocoding a larger set of locations can take some time, it may look like the geocoding was not done or done incompletely if looking at the map or in the Customers / Facilities / Suppliers input tables shortly after kicking off the geocoding. A helpful tool which shows the progress of the geocoding (and other tools / utilities within Cosmic Frog) is the Model Activity list:

1. Open the Model Activity list by clicking on the dark blue speech bubble icon at the top of the screen, towards the right.
2. After the “Geocode my customers” prompt was submitted and confirmed, the “Geocoding customers table” activity appeared in the list of model activities. It shows the progress of the geocoding, and this status is refreshed frequently.
3. Once the geocoding completes, click on the link in Leapfrog’s response to view the geocoded data. The link opens a new tab in the browser and shows Cosmic Frog’s Maps module with the Customers (the only table that was geocoded in this example) layer on the Supply Chain map enabled:

Expert Guidance on Anura Schema (Anura Help)

Leapfrog can teach users all about the Anura schema that underlies Cosmic Frog models, including:

• Table details: provides descriptions, purposes, and complete column listings for all tables.
• Column details: explains the usage, data types, and validation rules for individual columns.
• Relationships: maps connections between tables and their dependencies.
• Technical requirements: details primary keys, required fields, and default values.
• SCG model conversion: maps fields from SCG models to the Anura schema.
• Anura versioning: provides schema version and change details.

Example prompts:

• Can you teach me how to use the Unit Value field on the Products table?
• Where can I set the inventory carrying cost percentage?
• What is the SCG equivalent of Cosmic Frog’s minimum order quantity field on the Customer Fulfillment Policies table?
• What can I use the Customer Orders table for?
• Which fields on the Warehousing Policies table are required?
• What is the default value for the circuity factor in the Model Settings table?
• What are valid values for the latitude field on the Facilities table?
• Were there breaking changes in the latest update of the Anura schema?

The following 4 screenshots show examples of these types of prompts & Leapfrog’s responses: 1) ask Leapfrog to teach us about a specific field on a specific table, 2) find out which table to use for a specific modelling construct, 3) understand the SCG to Cosmic Frog’s Anura mapping for a specific field on a specific table, and 4) ask about breaking changes in the latest Anura schema update:

Systems Integration Knowledge (Anura Help)

Anura Help provides information around system integration, which includes:

• Engine awareness: identifies which tables and columns are used by different Cosmic Frog engines.
• Template models: provides information about various Cosmic Frog template models.

Example prompts:

• Which tables are required to run Throg?
• Which engines use the Bills of Materials table?
• What is the name of the main output summary table for Greenfield?
• Are there any template models available for network optimization?

The following 4 screenshots show examples of these types of prompts & Leapfrog’s responses: 1) ask which tables are required to run a specific engine, 2) find out which engines use a specific table, 3) learn which table contains a certain type of outputs, and 4) ask about availability of template models for a specific purpose:

Leapfrog is Self-aware (Text2SQL, Anura Help)

Leapfrog knows about itself, Optilogic, Cosmic Frog, the Anura database schema, LLM’s, and more. Ask Leapfrog questions so it can share its knowledge with you. For most general questions both LLMs will generate the same or a very similar answer, whereas for questions that are around capabilities, each may only answer what is relevant to it.

Example prompts:

• Can you show me the latest release notes?
• What version are you?
• What are your strengths and weaknesses?
• How do I create a Scenario?
• What does group name behavior aggregate do and what about enumerate?

The following 5 screenshots show examples of these types of prompts & Leapfrog’s responses: 1) ask both LLMs about their version, 2) ask a general question about how to do something in Cosmic Frog (Text2SQL), 3) ask Anura Help for the Release Notes, and 4 & 5) ask both LLMs about what they are good at and what they are not good at:

Multi-Language Support (Text2SQL, Anura Help)

Even though this documentation and Leapfrog example prompts are predominantly in English, Leapfrog supports many languages so users can ask questions in their most natural language. Where the Leapfrog response is in text form, it can respond in the language the question was asked in. Other response types like a standard message with a link or names of scenarios and scenario items will be in English.

The following 3 screenshots show: 1) a list of languages Leapfrog supports, 2) a French prompt to increase demand by 20%, and 3) a Spanish prompt asking Leapfrog to explain the Primary Quantity UOM field on the Model Settings table:

Tips & Tricks

To get the most out of Leapfrog, please take note of these tips & tricks:

1. Each Conversation thread is related to the active Cosmic Frog model. So, if you ask Leapfrog to “show me [this]” it will only pull that information from your active model.
2. Conversation history is stored on the platform at the user level – not in the model database – so it does not get shared when a model is shared. Note that if you are working in a Team rather than in your My Account (see documentation on Teams on the Optilogic platform here), the Leapfrog conversations you are creating will be available to the other team members when they are working with the same model.
3. Asking Leapfrog a question does not change your model: the response is not executed unless user decides to go ahead with it by clicking on an additional button (e.g. Run SQL, Create Scenario, etc.).
4. Optilogic is confident Leapfrog will help many users use Cosmic Frog more efficiently and we are very proud of it. It is not perfect, however. Therefore, users are encouraged to review Leapfrog’s responses carefully before making changes to model data.
5. Leapfrog (and LLMs generally) are non-deterministic. This means that you can expect to see slight variations in responses – in particular with text responses – for the exact same input prompt.
6. Leapfrog has a ‘memory’ within each unique Conversation. This means you can reference a previous prompt or response in a subsequent request. This is helpful for building upon, clarifying, or making adjustments to get the exact response you are looking for. Think for example of follow-up prompts like “I meant …”, “Like that, but with X changed”, “And how about Y?”, etc. Some examples of these are shown in screenshots at the end of this section.
7. Leapfrog is an eager conversationalist and will always do its best to answer your request. It is still learning though, and sometimes it gets off track. Some tactics to resolve this are:
   1. Try achieving what you desire in multiple steps. Some examples are shown in screenshots at the end of this section.
   2. If possible, use more explicit prompts which use table and field names (if you do not know the table / field names, ask Anura Help first!). Some examples are shown in screenshots at the end of this section.
   3. Start a new conversation when your line of questioning changes or when the responses you are receiving are clearly off track.
8. Within prompts, users can create new lines using Shift + Enter. This is convenient when for example entering a list, as shown in the following screenshot:

10. You can personalize the avatar you see when conversing with Leapfrog. By default, this avatar just uses the first letter of your Optilogic account name. Your Leapfrog user avatar is linked to your Frogger Pond Community profile: simply go to Frogger Pond Community > Profile icon > Preferences > Preferences > Profile Picture to update it.
11. Instead of typing prompts, you can also use voice to text if your computer supports this. “Online speech recognition” will need to be turned on when using Windows. You can find this by going to Start > Settings > Privacy (Windows 10) / Privacy & security (Windows 11) > Speech. Once there, you will see a toggle to turn online speech recognition on and off:

After this is turned on, you can start using it by pressing the keyboard’s Windows key + H. A bar with a microphone which shows messages like “initializing”, “listening”, “thinking” will show up at the top of your active monitor:

Now you can speak into your computer’s microphone, and your spoken words will be turned into text. If you put your cursor in Leapfrog’s question / prompt area, click on the microphone in the bar at the top so your computer starts listening, and then say what you want to ask Leapfrog, it will appear in the prompt area. You can then click on the send icon to submit your prompt / question to Leapfrog.

The following screenshots show several examples of how one can build on previous prompts and responses and try to re-direct Leapfrog as described in bullets 6 and 7 of the Tips & Tricks above. In the first example user wants to delete records from an input table whereas Leapfrog’s initial response is to change the Status of these records to Exclude. The follow-up prompt clarifies that user wants to remove them. Note that it is not needed to repeat that it is about facilities based in the USA, which Leapfrog still knows from the previous prompt:

In the following example shown in the next 3 screenshots, user starts by asking Leapfrog to show the 2 DCs with the highest throughput. The SQL query response only looks at Replenishment flows, but user wants to include Customer Fulfillment flows too. Also, the SQL Query does not limit the list to the top 2 DCs. In the follow-up prompt the user clarifies this (“Like that, but…”) without needing to repeat the whole question. However, Leapfrog only picks up on the first request (adding the Customer Fulfillment flows), so in the third prompt user clarifies further (again: “Like that, but…”), and achieves what they set out to do:

In the next 2 screenshots we see an example of first asking Leapfrog to show outputs that meet certain criteria (within 3%), and then essentially wanting to ask the same question but with the criteria changed (within 6%). There is no need to repeat the first prompt, it suffices to say something like “How about with [changed criteria]?”:

When Leapfrog only does part of what a user intends to do, it can often still be achieved in multiple steps. See the following screenshots where user intended to change 2 fields on the Production Count Constraints table and initially Leapfrog only changes one. The follow-up prompt simply consists of “And [change 2]”, building on the previous prompt. In the third prompt user was more explicit in describing the 2 changes and then Leapfrog’s response is what user intended to achieve:

Build, Run, and Analyze a Model using just Leapfrog!

Here we will step through the process of building a complete Cosmic Frog demo model, creating an additional scenario, running this new scenario and the Baseline scenario, and interrogating some of the scenarios’ outputs, all by using only Leapfrog.

Please note that if you are trying the same steps using Leapfrog in your Cosmic Frog:

• You may notice some small differences in data in the input and output tables as compared to the below, as the same prompt can result in different responses. For example, names of locations can be different if not explicitly specified in the prompt, and generating random numbers can result in different sets of numbers on different computers.
• In the screenshots below the order of the columns has sometimes been changed and records may appear in a different order due to sorting.
• In the screenshots below, the thumbs up/down for submitting feedback are shown at the top right of responses whereas they will be at the bottom left for you. These screenshots also do not show the icon with 3 dots to expand/collapse the Response Metadata section, but you will see them at the bottom right of your responses.

We will first list the prompts that were used to build, run, and analyze the model, and then review the whole process step-by-step through (lots of!) screenshots. Here is the list of prompts that were submitted to Leapfrog (all of them used the Text2SQL LLM):

1. Create a new blank model called US Distribution
2. Add 4 items to the Products table. Name them: Bullfrog_Train, Amphibian_Slime, Toady_Squishy, and Polliwog_Table with status = Include
3. Help me add locations. Set status for all of these to Include:
   1. add 3 DCs into Facilities table at the following locations: Memphis, TN Reno, NV Jacksonville, FL
   2. Add into Facilities table 2 MFG locations: one in Detroit, MI and the other in Dallas, TX
   3. add 50 customers into Customers table located at the capital city of each state of USA
4. Geocode all of my sites
5. Generate demand for all customers for all products with random quantity between 10 and 1000. Insert this into CustomerDemand table with status = include
6. Set model start date = 1/1/2025 and end date = 12/31/2025
7. Insert the following lanes into Transportation Policies with status = Include.
   1. From all facilities starting with MFG to all facilities starting with DC, these all come from the Facilities table
   2. From all facilities starting with DC to all customers
8. In transportation policies, put in transportation unit cost of $0.02 for lanes from DCs and $0.01 for MFG-DC lanes. Use EA-MI as uom for the cost
9. Add Production Policies for each facility starting with “MFG” for all products
10. Populate throughput capacity for DCs = 50k for Reno and Memphis and 100k for Jacksonville
11. Create a scenario which increases DC Memphis throughput capacity to 100k
12. Run network optimization on both scenarios
13. Where do I look for customer flow comparison between scenarios?
14. From the optimizationflowsummary table, using the CustomerFulfillment flows, compare the source used by each destination and show me the destinations with different sources in baseline vs. increase Memphis dc capacity scenario.
15. Like that, but show me the distinct combinations
16. What’s the cost difference between both scenarios?
17. Show me outbound quantity comparison for DCs

And here is the step-by-step process shown through screenshots, starting with the first prompt given to Leapfrog to create a new empty Cosmic Frog model with the name “US Distribution”:

Clicking on the link in Leapfrog’s response will take user to the Leapfrog module in this newly created US Distribution model:

1. The name of the model shown at the top in Cosmic Frog: US Distribution.
2. Next, we have added the second prompt to add 4 products to the Products table, where we specify their names, and that the Status field needs to be set to Include.
3. User clicked on the Run SQL button to execute the SQL query to add the products, which we double-check ran OK by opening the Products table:

In the next prompt (the third one from the list), distribution center (DC) and manufacturing (MFG) locations are added to the Facilities table, and customer locations to the Customers table. Note the use of a numbered list to help Leapfrog break the response up into multiple INSERT INTO statements:

After running the SQL of that Leapfrog response, user has a look in the Facilities and Customers tables and notices that as expected all Latitude and Longitude values are blank:

Since all Facilities and Customers have blank Latitudes and Longitudes, our next (fourth) prompt is to geocode all sites:

Once the geocoding completes (which can be checked in the Model Activity list), user clicks on one of the links in the Leapfrog response. This opens the Supply Chain map of the model in a new tab in the browser, showing Facilities and Customers, which all look to be geocoded correctly:

We can also double-check this in the Customers and Facilities tables, see for example next screenshot of a subset of 5 customers which now have values in their Latitude and Longitude fields:

For a (network optimization – Neo) model to work, we will also need to add demand. As this is an example/demo model, we can use Leapfrog to generate random demand quantities for us, see this next (fifth) prompt and response:

After clicking the Run SQL button, we can have a look in the Customer Demand input table, where we find the expected 200 records (50 customers which each have demand for 4 products) and eyeballing the values in the Quantity field we see the numbers are as expected between 10 and 1000:

Our sixth prompt sets the model end and start dates, so the model horizon is all of 2025:

Again, we can double-check this after running the SQL response by having a look in the Model Settings input table:

We also need Transportation Policies, the following prompt (the seventh from our list) takes care of this and creates lanes from all MFGs to all DCs and from all DCs to all customers:

We see the 6 enumerated MFG (2 locations) to DC (3 locations) lanes when opening and sorting the Transportation policies table, plus the first few records of the 150 enumerated DC to customer lanes. No Unit Costs are set so far (blank values):

Our eighth prompt sets the transportation unit costs on the transportation policies created in the previous step. All use a unit of measure of EA-MI which means the costs entered are per unit per mile, and the cost itself is 1 cent on MFG to DC lanes and 2 cents on DC to customer lanes:

Clicking the Run SQL button will run the 4 UPDATE statements, and we can see the changes in the Transportation Policies input table:

In order to run, the model also needs Production Policies, which the next (ninth) prompt takes care of: both MFG locations can produce all 4 products:

Again, double-checking after running the SQL from the response, we see the 8 expected records in the Production Policies input table:

Our 3 DCs have an upper limit as to how much throughput they can handle over the year, this is 50,000 for the DCs in Reno and Memphis and 100,000 for the DC in Jacksonville. Prompt number 10 sets these:

We can see these numbers appear in the Throughput Capacity field on the Facilities input table after running the SQL of Leapfrog’s response:

We want to explore what happens if the maximum throughput of the DC in Memphis is increased to 100,000; this is what the eleventh prompt asks to do:

Leapfrog’s response has both a SQL UPDATE query, which would change the throughput at DC_Memphis in the Facilities input table, and a Scenarios section. We choose to click on the Create Scenario button so a new scenario is created (Increase Memphis DC Capacity) which will contain 1 scenario item (set_dc_memphis_capacity_to_100000) that sets the throughput capacity at DC_Memphis to 100,000:

Our small demo model is now complete, and we will use Leapfrog (using our twelfth prompt) to run network optimization (using the Neo engine) on the Baseline and Increase Memphis DC Capacity scenarios:

While the scenarios are running, we are thinking about what will be interesting outputs to review, and ask Leapfrog about how one can compare customers flows between scenarios (prompt number 13):

This information can come in handy in one of the next prompts to direct Leapfrog on where to look.

Using the link from the previous Leapfrog response where we started the optimization runs for both scenarios, we open the Run Manager in a new tab of the browser. Both scenarios have completed successfully as their State is set to Done:

Looking in the Optimization Network Summary output table, we also see there are results for both scenarios:

In the next few prompts Leapfrog is used to look at outputs of the 2 scenarios that have been run. The prompt (number 14 from our list) in the next screenshot aims to get Leapfrog to show us which customers have a different source in the Increase Memphis DC Capacity scenario as compared to the Baseline scenario:

Leapfrog’s response is almost what we want it to be, however it has duplicates in the Data Grid. Therefore, we follow our previous prompt up with the next one (number 15), where we ask to see only distinct combinations. Instead of “distinct” we could have also used the word “unique” in our prompt:

We see that the source for around 11-12 customers changed from the DC in Jacksonville in the Baseline to the DC in Memphis in the Increase Memphis DC Capacity scenario.

Cost comparisons between scenarios are usually interesting too, so that is what prompt number 16 asks about:

We notice that increasing the throughput capacity at DC_Memphis leads to a lower total supply chain cost by about 56.5k USD. Next, we want to see how much flow has shifted between the DCs in the Baseline scenario compared to the Increase Memphis DC Capacity scenario, which is what the last prompt (number 17) asks about:

This tells us that the throughput at DC_Reno is the same in both scenarios, but that increasing the DC_Memphis throughput capacity allows a shift of about 24k units from the DC in Jacksonville to the DC in Memphis (which was at its maximum 50k throughput in the Baseline scenario). This volume shift is what leads to the reduction in total supply chain cost.

We hope this gives you a good idea of what Leapfrog is capable of today. Stay tuned for more exciting features to be added in future releases!

Do you have any Leapfrog questions or feedback? Feel free to use the Frogger Pond Community to ask questions, share your experiences, and provide feedback. Or, shoot us an email at Leapfrog@Optilogic.com.

Happy Leapfrogging!

Appendix – Personas Illustrating Text2SQL vs Anura Help LLM Usage

Example 1 – Alex output analysis and scenario creation/run

PERSONA: Alex is an experienced supply chain modeler who knows exactly what to analyze but often spends too much time pulling and formatting outputs. They are looking to be more efficient in summarizing results and identifying key drivers across scenarios. While confident in their domain expertise, they want help extracting insights faster without losing control or accuracy. They see AI as a time-saving partner that helps them focus on decision-making, not data wrangling.

USE CASE: After running multiple scenarios in Cosmic Frog, Alex wants to quickly understand the key differences in cost and service across designs. Instead of manually exporting data or writing SQL queries, Alex uses Leapfrog to ask natural-language questions which saves Alex hours and lets them focus on insight generation and strategic decision-making.

Model to use: Global Supply Chain Strategy (available under Get Started Here in the Explorer).

#	LLM	Prompt	Response
1	Text2SQL	Show me total supply chain cost comparison between ‘Baseline’ and ‘No Flow Constriants’ scenarios. Also include delta from baseline cost	SQL Select statements and Data Grid showing the requested comparison and delta:
2	Text2SQL	Using the facility summary output table, for each DC give me the total outbound quantity for the Baseline and No Flow Constraints scenarios (make the scenario names columns in the Data Grid) and add a column with the delta	SQL Select statements and Data Grid showing the requested comparison and delta:
3	Text2SQL	Now create a new scenario to exclude Salt Lake City DC and exclude flow constraint.	2 UPDATE SQL queries to make the requested changes and an option to create a scenario containing 2 items to make those changes. Click on the Create Scenario button to create the scenario.
4	Text2SQL	Run Neo on the exclude_salt_lake_dc_and _flow_constraints scenario	“Please confirm…” message to check what to run listing: Scenario Name(s): exclude_salt_lake_dc_and_flow_constraints Engine Name: NEO Resource Size: S Check Infeasibility: False Click on the Run button to go ahead with the run.
5	Text2SQL	(optional) compare total supply chain cost for Baseline, No Detroit DC, and exclude_salt_lake_dc_and _flow_constraints scenarios	SQL Select statements and Data Grid showing the requested comparison:

Example 2 – Chris with questions about Anura versions

PERSONA: Chris is an experienced supply chain modeler with a well-established, repeatable workflow that pulls data from internal systems to rebuild models every quarter.  He relies on consistency in the model schema to keep his automation running smoothly.

USE CASE: With an upcoming schema change in Cosmic Frog, Chris is concerned about disruptions or errors in his process and wants Leapfrog to help provide info on the changes that may require him to update his workflow.

Model to use: any.

#	LLM	Prompt	Response
1	Anura Help	What is the current Anura version?	“The current Anura version is 2.8.15, which is the latest released version. The upcoming release is 2.8.16.” Context for the response is provided too
2	Anura Help	Are there any breaking changes in the last released version?	“Yes, in the last version release (2.8.15), there are 3 breaking changes. These include:” and then they are listed. Context for the response is provided too
3	Anura Help	What about the next one?	“In the upcoming release (2.8.16), there are 6 breaking changes. These include:” and then they are listed. Context for the response is provided too

Example 3 – Larry Loves LLMs

PERSONA: Larry Loves LLMs – he wants to use an LLM to find answers.

USE CASE: I need to understand the outputs of this model someone else built. I want to know how many products come from each supplier for each network configuration. Can Leapfrog help with that?

Yes, Leapfrog can help with that! Let’s use Anura Help to better understand which tables have that data and then ask Text2SQL to pull the data.

Model to use: any; Global Supply Chain Strategy (available under Get Started Here in the Explorer).

#	LLM	Prompt	Response
1	Anura Help	I would like to understand how many products come from each supplier for each of my network opitmization scenarios, which table has that data?	“The table that contains data on the quantity of products supplied by each supplier for network optimization scenarios is the OptimizationSupplySummary table. This table provides a summary of the supply data, including the supplied quantity for different suppliers and products within a specific period.” Context for the response is provided too
2	Text2SQL	Show me the total number of products that come from each supplier for each scenario, sort by scenario then supplier. Pull data from the Optimization Supply Summary	SQL Select statement and Data Grid showing the requested data: