About the Fabricate Data Agent

What is the Data Agent?

The Data Agent is an AI-driven entity that performs data generation, analysis, and configuration actions within Tonic Fabricate. It is at the heart of the Fabricate application.

The following diagram provides an overview of the Data Agent input, output, and associated resources.

Diagram of the Data Agent flow, including input, output, and associated resources

How do you interact with the Data Agent?

You use a chat interface to provide instructions to the Data Agent. A given chat is scoped to a single project within an account.

You provide an initial prompt, along with any other clarifying details, including attached files with information such as database schemas and sample data. For example, Use the attached schema to create a database of transaction records that are linked to customer and product records.

The Data Agent, working with an LLM, uses your input to create scripts that it then uses to perform the prompted task, optionally asking for clarification and confirmation as it works.

Based on the results, you can instruct the Data Agent to make adjustments.

You can use multiple chats within a project. Each chat can refer to any database or data connection in the project.

What does the Data Agent produce?

The Data Agent produces synthetic data.

In many cases, the output is a set of structured data that might be generated to:

• A local database, where the data is stored in Fabricate. You can then export it in a variety of formats or to an external database.

• An external database that you connect directly to from Fabricate.

The Data Agent can also use synthetic data to populate unstructured files. For example, you can generate a set of credit application forms or medical discharge notes that are populated with synthetic values.

What resources does the Data Agent use to perform its tasks?

The Data Agent always has access to a set of built-in skills that it uses to perform a wide variety of analysis and generation tasks within Fabricate.

Within a project, the Agent can create and use plans, profiles, workflows, and custom skills.

• A plan provides a set of steps to complete a more complex task. For example, when you ask the Data Agent to create a database with a very large schema, or to generate data that includes complex relationships between tables or a long list of specifications for the data values.

• A profile is a thorough analysis of the data and its structure. Each table has its own profile. The Data Agent consults the profile whenever it needs to understand the content or structure of a table. It can also use a profile as the basis to generate new data.

• A workflow is a higher-level repeatable task, such as adding more data or validating new data. Instead of repeating the same instructions, you prompt the Agent to run the workflow.

• A custom skill describes a specific action to perform or a rule to follow in a specific context. For example, whenever you generate data to a REST API, you want the Data Agent to use a specific type of credential and a specific endpoint.

What affects the Data Agent behavior and performance?

At the highest level, the Data Agent behavior is based on the behavior configuration.

The Data Agent chat also provides options to adjust the behavior as you interact with the Data Agent within a project.

Data Agent instructions

The account, workspace, and user settings include a setting for Agent instructions, which provide a set of context rules for the Agent to remember across sessions and databases.

For each database, the Data Agent also creates a set of rules that it used to generate the data. You can adjust these rules to change how the Agent generates future data.

Data Agent behavior configuration

The settings for individual users include configuration for the Data Agent behavior:

• Verbosity determines how much explanation the Data Agent provides as it completes its work.

• Autonomy determines how frequently the Data Agent asks you for confirmation before it completes tasks.

LLM model

The Data Agent behavior and performance is partially based on the LLM model that it uses.

The available models are based on the LLM provider.

From the Data Agent chat, you can select the specific model to use.

Data Agent mode

The Data Agent mode also affects how the Data Agent behaves.

Agent mode is the default mode.

Plan mode means that the Data Agent creates a plan for each action.

Autonomous mode means that the Data Agent performs the action without asking you for any confirmation at all.

Data Agent feedback

The Data Agent also adjusts its behavior based on the feedback that you provide.

For each response, you can indicate whether it was helpful and, for non-helpful responses, provide additional feedback as to why it was not helpful.

The Data Agent uses the feedback to improve future responses.