Getting Started with Ada & Agentic AI

Overview

Ada is Optilogic’s next-generation agentic AI, enabling supply chain teams to work faster and with greater confidence across the full modeling lifecycle — from raw data preparation to optimization runs to executive reporting — all through natural language interactions.

Unlike traditional UI chat assistants, it deploys purpose-built agents that can pursue multi-step goals, use specialized skills, maintain conversational context, and coordinate with each other to complete workflows that previously required significant manual effort. This dramatically reduces the time required to move from raw data to recommendations.

As a core part of Optilogic’s Next Generation User Interface Platform, Ada provides a more intelligent and conversational approach to supply chain design work.

Ada is named after Ada Lovelace, widely regarded as the world’s first computer programmer and one of the earliest visionaries to recognize the potential of computational systems beyond pure calculation. The name reflects Optilogic’s goal of building intelligent systems that help people solve complex problems through collaboration between human expertise and advanced computing.

Quick Start

1. Log into the Next Generation UI Optilogic platform at https://ai.optilogic.app or click on the Ada icon in the navigation bar on the current Optilogic platform at https://optilogic.app
2. Ask Ada your question or to perform a task through chat in the top central part of the platform:
   1. Choose which database(s) the prompt applies to
   2. Choose which AI Agent to use
   3. Optionally, attach any file(s) the prompt needs to use
   4. Type your prompt in the chat textbox and click on submit, OR
   5. Click on an example prompt
3. Respond to any clarifications Ada needs
4. Authorize any actions if needed
5. Review responses & results

‍

Understanding Ada

What is Ada?

Ada is your AI-first supply chain modeling partner, designed specifically for the Optilogic platform. Through a conversational interface, Ada helps users build, validate, analyze, and improve supply chain models.

You can think of Ada as a chat agent like for example Claude and ChatGPT. But, unlike general-purpose AI chat tools, Ada is trained around supply chain modelling workflows and has access to Optilogic-specific tools, applications, databases, schemas, and platform capabilities.

Today, Ada includes four specialized AI Agents:

• Modeler Agent - a supply-chain modeling assistant focused on turning data into optimization-ready ANURA datasets and running/diagnosing solves in a disciplined, schema-driven way.
• Executive Agent (beta) - a read-only agent for decision-makers who want business findings and recommendations from solved models across the organization.
• Data Cleanser Agent - A data quality and transformation agent that can profile datasets, identify issues, standardize data, apply transformations, and validate results.
• Next Gen UI Agent (beta) – an interface-focused agent that helps users explore data, monitor work, and take action inside the Optilogic Next Gen UI platform.

‍

The Select the AI Agent part of the Create Your First Prompt section further below includes guidance on which agent to use for what type of question/task.

For a deeper technical explanation of how AI agents, tools, and skills work together, see the AI Agents: Architecture and Components help center article.

What Should I Use Ada for?

Teams commonly use Ada for:

• analyzing attached spreadsheets, PDFs, text files, and structured data files
• loading attached data into a Cosmic Frog model or DataStar project sandbox
• managing connected databases during a conversation
• building optimization-ready datasets
• validating model structure and assumptions
• profiling and cleaning raw supply chain data
• generating scenarios for comparison
• interpreting model outputs
• creating a shareable, interactive model-anatomy diagram
• summarizing findings for stakeholders
• identifying modeling gaps before solver runs
• exploring unfamiliar models or databases
• creating repeatable data cleansing and model building workflows
• analyzing and summarizing existing workflows

Ada works best for:

• multi-step analytical workflows
• iterative model refinement
• exploratory analysis
• structured operational tasks

Ada is less suited for:

• highly ambiguous business strategy discussions with no data/model context
• production-critical actions without human review
• tasks requiring unsupported integrations or external internet access

How to Think About Ada

Ada is best thought of as:

• a collaborative modeling partner
• capable of reasoning across multiple steps
• able to interact with connected Optilogic tools and databases
• but still dependent on the quality of the instructions and context provided

Ada does not automatically understand:

• your business goals
• model intent
• undocumented assumptions
• organizational conventions

The clearer the context you provide, the better the results typically become.

What Ada Can Connect to

Ada can connect to:

• Files - .csv, .xlsx, .xls, .pdf, .txt, .md, and .json files can be attached or pointed to (if already in the user's Optilogic account) for Ada to analyze, use as context for the prompt, or be loaded into a Cosmic Frog model or DataStar project sandbox.
• Cosmic Frog model databases — Postgres databases with a standard Anura schema that optimization and simulation engines read from and write to. The Modeler Agent builds and validates models here and can query them to perform various types of analysis.
• DataStar project databases — Postgres databases with a Starburst schema used by DataStar, the Optilogic platform application for data transformation work. The Modeler Agent can interact with DataStar macros and tasks for data workflows. Both the Modeler Agent and the Data Cleanser Agent can read from and write to DataStar Project Sandboxes.
• Postgres databases — any other Postgres databases on the Optilogic platform. The Data Cleanser Agent and the Modeler Agent can work with these alongside Cosmic Frog models and DataStar Project Sandboxes. ‍
• Optilogic engines — the Modeler Agent can launch solver runs for any of the Cosmic Frog engines directly. ‍
• Optilogic platform storage — both the Modeler Agent and Data Cleanser Agent can create artifacts that can be saved in the user’s Optilogic account (accessible through the Explorer application). These artifacts include for example Python scripts (.py files), markdown documents (.md), and web browser files (.html).
  

Ada operates entirely within the Optilogic platform and your connected databases. It does not access the internet or any data or systems outside of the Optilogic environment. It does not send your data to third parties beyond what is required by the underlying GPT family model API (see AI Data Security & Privacy).

‍

What to Expect When Working with Ada

Ada may:

• ask clarifying questions
• propose plans before execution
• require approvals before modifying data
• occasionally provide imperfect or incomplete answers
• improve when given more structured context

Long-running Tasks

When working on a response that takes a considerable amount of time/work, Ada will:

• complete scripts and data jobs without a time cut-off
• divide a large task among helpers that can work at the same time
• preserve completed progress if a later step fails
• recover when the AI provider connection drops during a response
• provide short updates between steps of a multi-step answer, so it is clear that the task is progressing

How to Use Ada

‍

To start using Ada, log into the next generation Optilogic platform at https://ai.optilogic.app or navigate there by clicking on the Ada icon in the navigation sidebar while on the current Optilogic platform (https://optilogic.app):

Besides this documentation, you can also get a guided tour on how to use Ada from within the platform itself. In the sidebar on the left, click on the Apps Launcher icon:

Then search for “Start Ada chat” and click on the Start Ada Chat Walkthrough item in the Actions list to start the tour:

How to Access Ada

When logged into the next-gen platform at https://ai.optilogic.app, there are 2 main ways to start using Ada:

1. Go to the Home page, if not yet there, by clicking on the icon in the sidebar along the left of the platform.
2. Ada will be available through chat at the top in the central part of the platform. Note other widgets are available below Ada.
3. If you want to just chat with Ada without other widgets and/or access historical conversations, click on the chat bubble icon in the sidebar. The whole central part of the platform will then be used by your chat with Ada:

Create Your First Prompt

In a new conversation, you can select your database(s), choose your agent, and enter your question/task ("prompt") for Ada. Optionally, you can attach any files and configure the chat style.

Note that for any further prompts within the same conversation, the chat style, database(s), and agent will remain as they were set for the first prompt. You can adjust them if required during a conversation.

Attach File(s)

If you have any .csv, .xlsx, .xls, .pdf, .txt, .md, or .json files on your computer or in your Optilogic workspace that you want Ada to have access to during the conversation, you can attach them:

1. To attach files to your Ada conversation, click on the plus icon at the left bottom of the prompt textbox.
2. Then hover over the "Attach a file" menu that comes up.
3. Three options for attaching files are available:
   
   1. Upload from computer - opens File Explorer on your computer which you can use to browse to the local file(s) to attach. Hold the Ctrl key down to select multiple files.
   2. Upload a folder - opens File Explorer on your computer which you can use to browse to the local folder to attach. All files in the folder and any of its subfolders will be attached to the conversation.
   3. Attach workspace file - opens a browser modal where you can navigate to files in your Optilogic workspace to find the one you want to attach.

Once files are attached, they are shown above the prompt. They can be removed by clicking on the cross icon to the right of the file name:

Set Interaction Style

Four different interaction styles are available for conversations with Ada. You can choose the one that suits your needs best:

1. To choose the style for your conversations with Ada, click on the settings icon - second from the left at the bottom of the prompt textbox.
2. Click on the Style drop-down list to view the options. The selected style has a darker background and a checkmark. Choose from:
   1. Scout (default)
      • Low autonomy - asks before significant actions
      • Explains plans before execution
      • Best for careful, high-impact modeling work
   2. Dash
      • High autonomy - makes reasonable assumptions and proceeds
      • Minimizes confirmation pauses
      • Best for rapid iteration and experienced users
   3. Mentor
      • Medium autonomy – proceeds with reasonable assumptions for routine steps; pauses to confirm key modeling choices/trade-offs
      • Explains reasoning while working
      • Highlights trade-offs and modeling decisions
      • Best for collaborative and educational workflows
   4. Caveman
      • Medium autonomy - proceeds on defaults for routine steps; only asks when a decision materially changes outcomes or is irreversible
      • Short, terse responses with minimal commentary
      • Best for experienced users who prefer compact interactions

‍

Select Database(s)

When a prompt needs to work on a specific Cosmic Frog model, DataStar project, or other Postgres database, it needs to be pointed to it:

1. Click on the Databases drop-down list to view the available databases in the account you are working in. These can be Cosmic Frog models, DataStar projects, and Postgres databases.
2. You can use the search box to quickly find the database(s) of interest.
3. At the top of the list, Cosmic Frog models will be listed. Here the one named Global Supply Chain Strategy has been selected.
4. Further down the list, DataStar projects are listed. Here the one named AI Agents is selected.

‍

Select the AI Agent

Four agents are accessible via Ada, the descriptions below the screenshot will help you choose the one most suited for your prompt:

1. Click on the Agent drop-down to view the available agents.
2. Choose the one most suitable for your question/task.

To guide you on choosing the best agent for the task, here is an overview of what each is good at.

Modeler Agent:

• Mapping cleaned / staged data into ANURA-structured inputs for all Cosmic Frog engines (Customers, Facilities, Products, Customer Demand, Inventory Policies, Shipments, etc.)
• Schema / enumerate / UoM compliance checks, referential integrity for model tables
• Pre-solve viability checks (connectivity, capacity vs demand, scenario alignment)
• Running any of the engines and diagnosing preprocessing / validation / empty-output issues
• Output analysis of solves, including reports comparing scenarios, summarizing KPI’s, etc.
• Analyzing, designing and creating DataStar projects and macros, for example to leave behind data transformation steps in a repeatable Macro

Executive Insights:

• From available solved models/databases in the account, choosing the one(s) most suitable for answering the question.
• Summarizes up to 3 main findings, including:
  
  • Why it is an important finding
  • Recommended next steps
  • Evidence to back up the finding
  • Confidence in the finding

Note the Executive Insights agent is a read-only agent: it does not change models or data and it leaves modeling work to the Modeler Agent.

Data Cleanser:

• Profiling data
• Fixing missing / invalid values, duplicates, outliers
• Standardizing names / IDs (customer / product / facility crosswalks), units, date formats
• Reshaping raw extracts into clean staging tables
• Reconciling keys across sources (referential integrity checks)

Next Gen UI Agent:

• Showing live dashboards (jobs, model previews) as interactive widgets
• Querying your databases (read-only SQL) and render results as tables, KPIs, and charts
• Exploring schemas (tables / columns / relationships) so we can write correct queries
• Navigating / opening apps (SQL Editor, Cosmic Frog, Run Manager, etc.) via buttons
• Helping manage your work (projects/goals/tasks) with reviewable drafts for multi-step changes

Enter Prompt

Enter your question/task and submit it to Ada:

1. In the prompt textbox, type your prompt. See the Best Practices and Tips & Tricks sections further below for advice on writing successful prompts.
2. Either hit Enter or click on the send icon to submit the prompt.
3. Instead of writing a prompt, you can use one of the example prompts located underneath the prompt textbox. Use the white arrows on the left and right to scroll through the available example prompts.

Ada’s Response

Once a prompt has been submitted, Ada will process it and formulate a response. Responses can have different formats, here we will see a mostly text-based result with a generated .md file, while other response types are covered in the next section.

1. The user submitted the prompt “Please examine this Cosmic Frog model and summarize the inputs, scenarios and outputs.”.
2. Larger tasks will be split into multiple subtasks and a description plus status of each is shown when this section is expanded. The section can be expanded/collapsed by using the caret icon on the left.
3. While Ada is processing the prompt, a status line appears which includes how long she has been processing the prompt for and information on the current step she is on. This section can be expanded, like it is here, by using the caret icon on the left.
4. Since the status section is expanded, we see all of Ada’s steps here. Note the scroll bar on the right can be used to scroll through all the steps and the individual subsections, e.g. Agent and Ada sections are shown here, can be expanded/collapsed too.
5. These icons at the top left of the detailed status section can be used to just filter the list for Status updates, Thinking steps, Tool calls, AI steps, and Knowledge retrieval steps (left to right; latter not shown here as it was not used).
6. Use these 2 icons to collapse and expand all steps in the status list.
7. If you want to stop Ada’s execution, you can click on the stop button.

When everything is collapsed, it will look similar to this next screenshot while Ada is processing:

Once Ada is done processing the prompt, the response will be displayed:

1. In the, now collapsed, status line we see that the response took 5 minutes.
2. On the right-hand side in the status line, the user clicked on this icon to open the list of Tool Calls.
3. Clicking on this icon opens the list of Knowledge Retrieval calls made while processing.

The first part of the response for this prompt, which gives an overview of the model and summarizes the inputs, is shown in this next screenshot:

Then follow summaries of the scenarios and outputs (not shown) before concluding with the limitations of the review:

1. Users can copy the response text by clicking on this copy icon.
2. To help Ada continuously improve, users can give a thumbs up or down if the response is considered good or poor.
3. An .md file that contains the summarized inputs is generated and attached to the response. It can be downloaded, saved to the user's workspace, and opened in the Lightning Editor app from here.
4. To the right of the prompt textbox, information has now appeared to indicate that there is 1 artifact associated with this conversation (the current-inputs.md file just generated) and 0 attachments. Clicking on here gives users options to access associated artifacts and attachments.
5. To continue the conversation, type your follow up question/task in the prompt textbox. Ada will remember the context of the conversation so far and will use it to respond to follow-up prompts.

Additional Response Types and Interactions

Besides responses that are purely text-based, you will come across other types too. For example, when your input is needed, Ada will pause the response and ask you for feedback. Here the user was asked if they wanted to go ahead with cleaning a dataset after profiling it, to which they said yes:

1. As a follow-up from a previous prompt profiling a dataset, the user has asked Ada to go ahead and clean the just profiled tables.
2. The current status indicates that Human in the Loop (HITL) feedback is required now.
3. The top part summarizes why HITL feedback is needed and states what will happen once the feedback is received.
4. Often, Ada will propose several ways forward and all the user needs to do is click on the desired option. A selected option is shown in a darker color.
5. If the user does not want to go forward with an Ada-proposed option, they can use the textbox to provide instructions on what to do for that specific feedback item. Additional context can be provided here too.
6. Once feedback options are selected, click on “Send response” for Ada to continue processing.
7. If it has become clear that the prompt will not achieve what you intended or you need more time to look into something Ada surfaced, you can click on “Stop agent” to halt execution.

Note that if a HITL request times out, you can still continue the conversation and tell Ada your preferences on how to proceed in the follow up prompt or ask her to repeat/rephrase the feedback she requires.

When Ada modifies a Cosmic Frog model or a DataStar project sandbox, users can verify these in the respective applications on the current platform (https://optilogic.app). Here, we are checking if the data cleaning step indeed created the clean_ tables in the sandbox of the connected DataStar project:

Responses can also contain files as we have already seen above briefly too:

1. The user has prompted Ada to create a scenario comparison report in HTML format.
2. The top part of the response summarizes what Ada has done.
3. Two files are available in the response, this first one is a Python script which a user could save for future use to create scenario comparison reports using the same logic.
4. This is the report itself. Use the 3 buttons to the right of the report name to download it, save it to your workspace, or to open it in the Lightning Editor application.
5. Responses that generated multiple artifacts have options for users to simultaneously download all, save all to workspace, or open all in Lightning Editor.
6. If a conversation has generated any artifacts and/or it has any files attached to it by the user, this icon with the number of artifacts and number of attachments will be showing. This way you can easily access all artifacts and attachments if you are further up or down in a conversation.
7. Clicking on the icon showing the number of artifacts and attachments brings up this "Session assets" modal, where all artifacts and attachments are listed and can be accessed.
8. Hovering over an artifact brings up the options to download, save to workspace, and open in Lightning Editor again.

When saving an artifact to your workspace, the following modal will come up where you can choose the location to save it and indicate if any pre-existing files with the same name at the chosen location should be overwritten or not:

When choosing to open the report in Lightning Editor, it does so in the new platform, to the right of the conversation with Ada so users do not need to change context:

The Next Gen UI Agent can help you for example with changing the look and feel of the platform’s UI. In this example it created a tour on how to customize the UI where the user can click on the buttons in the response to be taken directly to that part of the UI. Note that only part of Ada's response is shown in the screenshot:

Historical Conversations

Conversations with Ada are by default saved and users can return to them to review, audit, or continue the conversation:

1. Click on the chat icon in the task bar to open Ada and any historical conversations.
2. To just see your current conversation, click on the close icon at the right top to close the list of historical conversations.
3. To start a new conversation, click on the plus button.
4. To quickly find a conversation, use the search box.
5. Use these buttons to refresh the list, select 1 or multiple conversations for bulk delete, or to filter the conversations list either by date or based on tags (automatically assigned).
6. A card for a conversation in the list contains:
   1. The conversation’s name, by default this is the first prompt in the conversation.
   2. When this conversation was had.
   3. How many messages there are in the conversation, this includes user prompts and Ada responses.
   4. Any tags to facilitate filtering conversations.
   5. Use this button to surface a context menu with pin, rename, copy ID, send to Support, and delete options for the conversation.
   6. This delete button can also be used to remove a conversation.

Best Practices

Keep Conversations Focused

Ada performs best when conversations stay centered on a single task or workflow. Avoid mixing unrelated activities — such as model building, reporting, and data cleansing — in the same chat.

Focused conversations improve response quality, reduce confusion, and make it easier for Ada to maintain context.

Give Context Before the Task

Provide business context, objectives, constraints, and relevant background before asking Ada to perform work.

Better prompts typically include:

1. Business or modeling context
2. Constraints or requirements
3. The specific task

Example: Instead of: “Build scenarios for this model.”

Try: “This model evaluates manufacturing diversification risk across LATAM and EMEA. The goal is to reduce China dependency while minimizing transportation cost increases. Create several realistic diversification scenarios.”

Ask Ada to Explore Before Acting

For complex workflows, first ask Ada to explore, profile, summarize, or analyze the environment before making changes.

Examples:

• “Explore the current model and summarize its structure and assumptions.”
• “Profile the selected datasets and identify data quality issues.”

This gives both you and Ada better shared context before execution and reduces downstream errors.

Plan First for Multi-Step Workflows

For larger workflows, ask Ada to propose a plan before executing actions.

Example: “Before making changes, provide a step-by-step plan for how you would approach this workflow.”

This allows you to:

• validate assumptions early
• review sequencing
• catch issues before execution
• maintain better control over complex tasks

Be Explicit About Constraints

Clearly state any important rules or limitations.

Examples:

• “Do not overwrite existing tables.”
• “Do not modify baseline assumptions.”
• “Only use cleansed datasets.”
• “Do not generate synthetic data.”

Explicit constraints improve consistency and reduce unintended actions.

Ask Ada Clarifying Questions

If a workflow is complex or ambiguous, invite Ada to ask clarifying questions before proceeding.

Example: “Before executing, ask any clarifying questions needed to complete this task correctly.”

This often improves first-pass accuracy significantly.

Start a New Conversation When Switching Contexts

Create a new conversation when:

• switching between unrelated tasks
• moving from brainstorming to execution
• responses become inconsistent or degraded

Long conversations can dilute context and reduce response quality over time.

Tips & Tricks

Ask for Multiple Options

Instead of requesting a single recommendation, ask Ada for multiple approaches and trade-offs.

Example: “Provide three approaches for supplier diversification and explain the trade-offs of each.”

This helps surface alternatives and improves decision-making.

Re-State Important Constraints During Long Workflows

In long conversations, periodically remind Ada about key requirements.

Examples:

• “Reminder: do not modify schema structure.”
• “Reminder: baseline assumptions must remain unchanged.”

This helps reduce context drift.

Generate a Summary Before Starting a New Chat

If a conversation becomes long or complex, ask Ada to summarize:

• goals
• decisions
• assumptions
• constraints
• unresolved issues

You can paste this summary into a new conversation to preserve context without carrying forward unnecessary noise.

Helpful prompt: “Summarize this conversation into a clean handoff document including goals, technical decisions, constraints, and next steps.”

Avoid Overloading Prompts

More information is not always better.

Instead:

• provide only relevant context
• break large inputs into sections
• clearly label important information
• keep the most critical instructions near the top

Focused prompts generally produce better results than overly large or unstructured requests.

Ask Ada What It Can Do

If you are unsure how to approach a task, ask Ada directly.

Examples:

• “What can this agent help me with?”
• “What are best practices for this workflow?”
• “Show me example prompts for this task.”
• “Create a guided tour of this feature.”

Ada can often suggest workflows, prompts, and capabilities you may not know are available.

Current Limitations and Known Behaviors

Ada is evolving rapidly, and some platform capabilities are still in active development.

DataStar and Cosmic Frog Integration

The Next Gen platform does not yet provide fully seamless integration of the DataStar and Cosmic Frog applications. For some processes, users will need to open these in the current platform (https://optilogic.app).

UI Rendering Requires the Next Gen UI Agent

Advanced inline visualizations and UI rendering features currently require selecting the Next Gen UI Agent explicitly.

Long Conversations Can Degrade Performance

As conversations grow longer, Ada may lose context, become repetitive, or produce less reliable responses.

Starting a fresh conversation for new workflows or projects generally improves results.

AI Responses Should Always Be Reviewed

Like all AI systems, Ada can occasionally produce incorrect or misleading outputs.

Always validate:

• optimization assumptions
• generated SQL
• data transformations
• KPI interpretations
• scenario recommendations

before using outputs in production or customer-facing work.

Session Stability

The platform refreshes sessions automatically when you return to an idle or backgrounded browser tab. If a workflow still appears interrupted, refresh the page and review the conversation state before restarting the conversation.

Model Anatomy Example

Asking Ada to show a model's anatomy will allow you to see the model as a picture: you get an interactive diagram with every input, decision variable, cost and constraint shown by what it does, sized by how much data is in it, all flowing toward the resulting total cost.

The model anatomy comes out as a single .html file you can share and anatomies can be generated for any of the Cosmic Frog engines.

We will walk through creating and examining the anatomy of a model using Ada here. With a model named Capacity Expansion v2 selected, the user asks Ada to show the anatomy of the model. Since there are multiple scenarios with outputs in this model, Ada asks the user which scenario to use. The user chooses to use the "2a Consider Reno from 2027" scenario:

Ada's response gives a quick summary of the scenario and includes the .html file with the anatomy, which the user can download and then open in a browser, along with several other artifacts:

The next 2 screenshots show the anatomy of this scenario, first the left part and then the right part. Along the top, the model name, engine, and total cost are listed on the left, along with statistics around customers, periods, flow and demand amounts on the right-hand side. The diagram itself shows input data, decision variables, costs, and constraints from top to bottom color coded by type (legend towards the bottom), while costs accumulates from left to right.

Hovering over an element in the diagram isolates it and shows just the element plus the elements it is connected to. In addition, it pops up a tooltip with information on the element. Here the user is hovering over the Open Status decision variable:

Other Helpful Resources

• Understanding Ada Credits
• The Modeler AI Agent
• The Data Cleansing AI Agent
• An overview of the Next Generation Optilogic Platform
• An overview of Optilogic's AI Agents and Utilities‍
• AI Agents: Architecture and Components‍
• Learn more about Cosmic Frog‍
• Learn more about DataStar

We hope you are going to have many productive conversations with Ada! Please do not hesitate to contact our Support team via support@optilogic.com if you have any questions or concerns.

Appendix – AI Data Security and Privacy

Applies to all Optilogic AI systems.

No confidential client information will be used as inputs or part of model training and validation datasets. In addition:

• Data Ownership: You retain ownership of your data. The data you input is not shared with any outside parties.
• No External Sharing: None of the data used when interacting with Optilogic Agentic AI will be shared or allowed to be used to train any LLM or model.
• Access Control: Access to your conversation history is limited to you and authorized Optilogic personnel. Your conversation data is not accessible to other users.
• Data Security: Optilogic employs industry-standard security practices to protect your data from unauthorized access and breaches.

Data Minimization: The amount of data shared with the AI provider depends on the task being performed. Optilogic's agents are engineered to query and pass only the minimum data required for each specific operation — ranging from structural metadata (table and column names, data types, statistical summaries) for schema-level tasks, to slices of actual data values for operations that require it, such as data cleansing, outlier detection, or name matching.

Optilogic does not transmit your entire dataset to the AI provider in a single operation. However, over the course of a session, the AI provider may process portions of your data as needed to complete the tasks you request.

No Model Training: Optilogic does not use your data to train AI models. Optilogic’s current AI provider (OpenAI) does not use API-submitted data for model training under their enterprise API terms. Refer to OpenAI policies here: https://openai.com/policies/.

Built-in Agent Safety Instructions: Optilogic agents include standing safety instructions in their core configuration to guard against prompt injection — attempts to manipulate agent behavior through user messages or data the agent processes. These instructions:

• Maintain platform rules regardless of override attempts made through conversation or through data in your models
• Treat query results, external files, and retrieved knowledge as content to analyze, not as commands to execute
• Decline requests to reveal or reconstruct system-level instructions
• Resist common manipulation techniques — including urgency, false authority, and roleplay scenarios — used to bypass safeguards
• These protections are instruction-based: they shape how the agent is directed to behave. They do not constitute a technical barrier, and they do not replace careful review of agent actions before approving high-impact operations.

Best Practices: Users should avoid including sensitive information (PII, credentials, etc.) in table/column names or prompts, as these are shared with the AI provider.