Helper components in LLM Controls

Helper components provide utility functions to help manage data, tasks, and other components in your flow.

​

Use a helper component in a flow​

Chat memory in LLM Controls is stored either in local LLM Controls tables with LCBufferMemory or connected to an external database. The Store Message helper component stores chat memories as Data objects, and the Message History helper component retrieves chat messages as data objects or strings. This example flow stores and retrieves chat history from an AstraDBChatMemory component with Store Message and Chat Memory components.

​

Batch Run​

The Batch Run component runs a language model over each row of a DataFrame text column and returns a new DataFrame with the original text and an LLM response. The response contains the following columns:

• text_input: The original text from the input DataFrame.
• model_response: The model’s response for each input.
• batch_index: The processing order, with a 0-based index.
• metadata (optional): Additional information about the processing.

These columns, when connected to a Parser component, can be used as variables within curly braces. To use the Batch Run component with a Parser component, do the following:

1. Connect a Model component to the Batch Run component’s Language model port.
2. Connect a component that outputs a DataFrame, like the File component, to the Batch Run component’s DataFrame input.
3. Connect the Batch Run component’s Batch Results output to a Parser component’s DataFrame input. The flow looks like this:

4. In the Column Name field of the Batch Run component, enter a column name based on the data you’re loading from the File loader. For example, to process a column of names, enter name.
5. Optionally, in the System Message field of the Batch Run component, enter a System Message to instruct the connected LLM on how to process your file. For example, create a business card for each name.
6. In the Template field of the Parser component, enter a template for using the Batch Run component’s new DataFrame columns. To use all three columns from the Batch Run component, include them like this:
7. To run the flow, in the Parser component, click Run component.
8. To view your created DataFrame, in the Parser component, click.
9. Optionally, connect a Chat Output component, and open the Playground to see the output.

Parameters

Inputs

Name	Type	Description
model	HandleInput	Connect the ‘Language Model’ output from your LLM component here. Required.
system_message	MultilineInput	A multi-line system instruction for all rows in the DataFrame.
df	DataFrameInput	The DataFrame whose column is treated as text messages, as specified by ‘column_name’. Required.
column_name	MessageTextInput	The name of the DataFrame column to treat as text messages. If empty, all columns are formatted in TOML.
output_column_name	MessageTextInput	Name of the column where the model’s response is stored. Default=model_response.
enable_metadata	BoolInput	If True, add metadata to the output DataFrame.

Outputs

Name	Type	Description
batch_results	DataFrame	A DataFrame with all original columns plus the model’s response column.

​

Current date

The Current Date component returns the current date and time in a selected timezone. This component provides a flexible way to obtain timezone-specific date and time information within an LLM Controls pipeline.

Parameters

Inputs

Name	Type	Description
timezone	String	The timezone for the current date and time.

Outputs

Name	Type	Description
current_date	String	The resulting current date and time in the selected timezone.

​

ID Generator

This component generates a unique ID.

Parameters

Inputs

Name	Type	Description
unique_id	String	The generated unique ID.

Outputs

Name	Type	Description
id	String	The generated unique ID.

​

Message history

This component retrieves chat messages from LLM Controls tables or external memory. In this example, the Message Store component stores the complete chat history in a local LLM Controls table, which the Message History component retrieves as context for the LLM to answer each question. For more information on configuring memory in LLM Controls.

Parameters

Inputs

Name	Type	Description
memory	Memory	Retrieve messages from an external memory. If empty, the LLM Controls tables are used.
sender	String	Filter by sender type.
sender_name	String	Filter by sender name.
n_messages	Integer	The number of messages to retrieve.
session_id	String	The session ID of the chat. If empty, the current session ID parameter is used.
order	String	The order of the messages.
template	String	The template to use for formatting the data. It can contain the keys {text}, {sender} or any other key in the message data.

Outputs

Name	Type	Description
messages	Data	The retrieved messages as Data objects.
messages_text	Message	The retrieved messages formatted as text.
dataframe	DataFrame	A DataFrame containing the message data.

​

Message store

This component stores chat messages or text in LLM Controls tables or external memory. In this example, the Message Store component stores the complete chat history in a local LLM Controls table, which the Message History component retrieves as context for the LLM to answer each question. For more information on configuring memory in LLM Controls.

Parameters

Inputs

Name	Type	Description
message	String	The chat message to be stored. (Required)
memory	Memory	The external memory to store the message. If empty, the LLM Controls tables are used.
sender	String	The sender of the message. Can be Machine or User. If empty, the current sender parameter is used.
sender_name	String	The name of the sender. Can be AI or User. If empty, the current sender parameter is used.
session_id	String	The session ID of the chat. If empty, the current session ID parameter is used.

Outputs

Name	Type	Description
stored_messages	List[Data]	The list of stored messages after the current message has been added.

​

Structured output

This component transforms LLM responses into structured data formats. In this example from the Financial Report Parser template, the Structured Output component transforms unstructured financial reports into structured data. The connected LLM model is prompted by the Structured Output component’s Format Instructions parameter to extract structured output from the unstructured text. Format Instructions is utilized as the system prompt for the Structured Output component. In the Structured Output component, click the Open table button to view the Output Schema table. The Output Schema parameter defines the structure and data types for the model’s output using a table with the following fields:

• Name: The name of the output field.
• Description: The purpose of the output field.
• Type: The data type of the output field. The available types are str, int, float, bool, list, or dict. The default is text.
• Multiple: This feature is deprecated. Currently, it is set to True by default if you expect multiple values for a single field. For example, a list of features is set to True to contain multiple values, such as ["waterproof", "durable", "lightweight"]. Default: True.

The Parser component parses the structured output into a template for orderly presentation in chat output. The template receives the values from the output_schema table with curly braces. For example, the template EBITDA: {EBITDA}, Net Income: {NET_INCOME}, GROSS_PROFIT: {GROSS_PROFIT} presents the extracted values in the Playground as EBITDA: 900 million, Net Income: 500 million, GROSS_PROFIT: 1.2 billion.

Parameters

Inputs

Name	Type	Description
llm	LanguageModel	The language model to use to generate the structured output.
input_value	String	The input message to the language model.
system_prompt	String	The instructions to the language model for formatting the output.
schema_name	String	The name for the output data schema.
output_schema	Table	The structure and data types for the model’s output.
multiple	Boolean	[Deprecated] Always set to True.

Outputs

Name	Type	Description
structured_output	Data	The structured output is a Data object based on the defined schema.