LLM Controls objects

​

Data object​

• Main Attributes:
  • text_key: Specifies the key to retrieve the primary text data.
  • data: A dictionary to store additional data.
  • default_value: default value when the text_key is not present in the data dictionary.

​

Create a Data Object​

1. Create a Data object by directly assigning key-value pairs to it.
2. The text_key specifies which key in the data dictionary should be considered the primary text content. The default_value provides a fallback if the text_key is not present.
3. The Data object is also convenient for visualization of outputs, since the output preview has visual elements to inspect data as a table and its cells as pop-ups for basic types. The idea is to create a unified way to work and visualize complex information in LLM Controls.
4. To receive Data objects in a component input, use the DataInput input type.

​

Message object​

1. Core message data:
   • text: The main text content of the message
   • sender: Identifier for the sender (“User” or “AI”)
   • sender_name: Name of the sender
   • session_id: Identifier for the chat session (string or UUID)
   • timestamp: Timestamp when the message was created (UTC)
   • flow_id: Identifier for the flow (string or UUID)
   • id: Unique identifier for the message
2. Content and files:
   • files: List of files or images associated with the message
   • content_blocks: List of structured content block objects
   • properties: Additional properties, including visual styling and source information
3. Message state:
   • error: Boolean indicating if there was an error
   • edit: Boolean indicating if the message was edited
   • category: Message category (“message”, “error”, “warning”, “info”)

​

Create a Message object​

​

ContentBlock object​

• TextContent has a text field for storing strings of text
• MediaContent has a urls field for storing media file URLs
• CodeContent has code and language fields for code snippets
• JSONContent has a data field for storing arbitrary JSON data
• ToolContent has a tool_input field for storing input parameters for the tool

​

Create a ContentBlock object​

from llmcontrols.schema.content_block import ContentBlock
from llmcontrols.schema.content_types import TextContent, MediaContent, CodeContent, JSONContent, ToolContent

# Create a ContentBlock with mixed content types
content_block = ContentBlock(
    title="My Content Block",
    contents=[
        TextContent(text="Hello, this is a text block."),
        MediaContent(urls=["https://example.com/image.png"], caption="An example image"),
        CodeContent(code="print('Hello World')", language="python"),
        JSONContent(data={"key": "value", "count": 42}),
    ]
)

​

Add ContentBlocks objects to a message

from llmcontrols.schema.message import Message
from llmcontrols.schema.content_block import ContentBlock
from llmcontrols.schema.content_types import TextContent, MediaContent

# Create content blocks
text_block = ContentBlock(
    title="Summary",
    contents=[TextContent(text="Here is the analysis result.")]
)

media_block = ContentBlock(
    title="Related Images",
    contents=[MediaContent(urls=["https://example.com/chart.png"], caption="Results chart")]
)

# Add to a message
message = Message(text="Analysis complete.")
message.content_blocks = [text_block, media_block]

​

DataFrame object​

​

Create a DataFrame object​

from llmcontrols.schema import Data
from llmcontrols.schema.data import DataFrame
# From a list of Data objects
data_list = [Data(data={"name": "John"}), Data(data={"name": "Jane"})]
df = DataFrame(data_list)
# From a list of dictionaries
dict_list = [{"name": "John"}, {"name": "Jane"}]
df = DataFrame(dict_list)
# From a dictionary of lists
data_dict = {"name": ["John", "Jane"], "age": [30, 25]}
df = DataFrame(data_dict)

​

Key Methods

• to_data_list(): Converts the DataFrame back to a list of Data objects.
• add_row(data): Adds a single row (either a Data object or a dictionary) to the DataFrame.
• add_rows(data): Adds multiple rows (list of Data objects or dictionaries) to the DataFrame.

​

Usage Example

# Create a DataFrame
df = DataFrame([Data(data={"name": "John"}), Data(data={"name": "Jane"})])
# Add a new row
df = df.add_row({"name": "Alice"})
# Convert back to a list of Data objects
data_list = df.to_data_list()
# Use pandas functionality
filtered_df = df[df["name"].str.startswith("J")]

​

To use DataFrame objects in a component input, use the DataFrameInput input type:

DataFrameInput(
    name="dataframe_input", 
    display_name="DataFrame Input", 
    info="Input for DataFrame objects.", 
    tool_mode=True