AI Overview

​

Available Modules

​

LLM Providers

• OpenAI: mo.ai.llm.openai
• Anthropic: mo.ai.llm.anthropic
• Google: mo.ai.llm.google
• Groq: mo.ai.llm.groq
• Bedrock: mo.ai.llm.bedrock
• Pydantic AI: mo.ai.llm.pydantic_ai

​

Core Types

​

ChatMessage

• role (Literal[“user”, “assistant”, “system”]): The role of the message
• content (Any): The content of the message (can be a rich Python object)
• id (str): The id of the message
• parts (list[ChatPart]): Parts from AI SDK Stream Protocol (must be serializable to JSON)
• attachments (Optional[list[ChatAttachment]]): Optional attachments to the message
• metadata (Any | None): Optional metadata

​

ChatAttachment

• url (str): The URL of the attachment (can be a hosted file URL or a Data URL)
• name (str): The name of the attachment, usually the file name (default: “attachment”)
• content_type (Optional[str]): A string indicating the media type (extracted from pathname extension by default)

​

ChatModelConfig

• max_tokens (Optional[int]): Maximum number of tokens
• temperature (Optional[float]): Temperature for the model (randomness)
• top_p (Optional[float]): Restriction on the cumulative probability of prediction candidates
• top_k (Optional[int]): Number of top prediction candidates to consider
• frequency_penalty (Optional[float]): Penalty for tokens which appear frequently
• presence_penalty (Optional[float]): Penalty for tokens which already appeared at least once

​

Usage Example

import marimo as mo

# Create a chat message
message = mo.ai.ChatMessage(
    role="user",
    content="Hello, how can I help you?",
    id="msg-1"
)

# Create an attachment
attachment = mo.ai.ChatAttachment(
    url="https://example.com/image.png",
    name="example.png"
)

# Configure a model
config = mo.ai.ChatModelConfig(
    max_tokens=1000,
    temperature=0.7
)