Type definitions

Language Model Type Definitions

Inspired by Vercels AI SDK.

Language Model Prompt

Conversations with LLMs consist of a series of prompts/messages. Each prompt can be of four types:

• system: Provides context or instructions to the model.
• user: Represents the user’s input or question.
• assistant: Represents the model’s response.
• tool: Represents a tool call made by the model.

Check the next section for the structure of each type.

// system
{
  "role": "system",
  "content": string
}
// user
{
  "role": "user",
  "content": [LanguageModelText | LanguageModelFile]
}
// assistant
{
  "role": "assistant",
  "content": [LanguageModelText | LanguageModelFile | LanguageModelReasoning | LanguageModelToolCall]
}
// tool
{
  "role": "tool",
  "content": [LanguageModelToolResult]
}

Language Model Content

Language Model Text

{
  "type": "text",
  "text": string,
  "providerOptions": object // Optional provider-specific options
}

Language Model File

{
  "type": "file",
  "mediaType": string, // IANA media type (e.g., 'image/png', 'audio/mp3')
  "data": string, // file content encoded in base64
  "filename?": string, // Optional filename for the file
  "providerOptions": object // Optional provider-specific options
}

Language Model Reasoning

{
  "type": "reasoning",
  "text": string,
  "providerOptions": object // Optional provider-specific options
}

Language Model Tool Call

{
  "type": "tool_call",
  "toolCallId": string,
  "toolName": string,
  "args": string, // JSON string of arguments passed to the tool
  "providerOptions" object // Optional provider-specific options
}

Language Model Tool Result

{
  "type": "tool_result",
  "toolCallId": string,
  "toolName": string,
  "result": string, // JSON string of the tool result
  "providerOptions" object // Optional provider-specific options
}