Overview

The Data Bank in Agent Studio is a centralized repository that stores and provides access to data generated or collected during a plugin’s execution. It enables subsequent steps in a workflow to reference data from previous steps, such as user inputs (slots), action outputs, or metadata. The Data Bank ensures seamless data flow across plugins, actions, and conversation processes, making it a core concept for building dynamic, context-aware workflows.

📘

Important: While the Data Bank shares a common structure (meta_info and data keys), availability of specific keys and sub-keys varies by context (Compound Actions, HTTP Actions, Conversation Processes, or Slot Validation Policies). Always check the context-specific details below to avoid reference errors.

Key Use Cases

• Reusing User Input: Reference a user-provided slot (e.g., ticket_id) in multiple actions.
• Chaining Actions: Pass the output of one action (e.g., a device ID) as input to another.
• Accessing Metadata: Use user attributes (e.g., meta_info.user.email_addr) for personalized workflows.

Data Bank Structure

The Data Bank contains two primary keys: meta_info and data These keys store metadata and workflow-specific data, respectively. Below, each key is detailed with its availability across contexts.

meta_info

Provides built-in metadata about the current execution context, such as user attributes or unique identifiers.

Availability

• Compound Actions: N/A.
• HTTP Actions: meta_info.user, meta_info.action_instance_id.
• Conversation Processes: meta_info.user.
• Slot Validation Policies: meta_info.user

meta_info.user

Contains attributes of the user who triggered the plugin. Use the notation meta_info.user.<attribute> to access specific attributes. For a full list, see the User Attributes Reference.

meta_info.action_instance_id

A unique UUID provided during execution, usable as an idempotency key to deduplicate retry calls.

Example (HTTP Actions Only):

Referencing meta_info.user.email_addr and meta_info.action_instance_id in an HTTP Action

data

Stores workflow-specific data, including slot values and action outputs. Use the notation data.<key>.

Availability

• Compound Actions: data.input_argument, data.output_key.
• HTTP Actions: N/A.
• Conversation Processes: data.slot_name, data.output_key.
• Slot Validation Policies: N/A.

data.slot_name

References user-provided slot values defined in a conversation process. Use this to access input variables collected from the user or inferred by the Assistant.

data.input_arg

References system-provided input arguments defined in a compound action.

data.output_key

References the output of actions (e.g., Built-in, HTTP) or expressions (e.g., for, raise, action). Each action or expression saves its output to a specified output_key, which can be used in subsequent steps.

Example: Action Output (Compound Actions):

steps:
  - action:
      action_name: get_user_device
      output_key: device
      input_args:
        user_email: meta_info.user.email_addr
  - action:
      action_name: clean_recycle_bin_on_device
      output_key: remote_action_result
      input_args:
        device_id: data.device.asset_uuid  # References output from first action

response

Used specifically in an action’s Output Mapper to reference the raw response from the action. Use DSL syntax to extract specific fields.

Availability

• Conversation Process Only (in Output Mappers).

Example

value

Used exclusively in Slot Validation Policy to reference the current slot’s value during validation. Combine with DSL expressions to enforce rules.

Example:

For a due_date slot, ensure the date is in the future:

$PARSE_TIME(value) > $TIME()

Accessing the Data Bank by Context

The following table summarizes key availability. Reference errors occur if you attempt to use a key outside its supported context.