LLMMessage

A single message in an LLM conversation.

Defined in: src/core/types/providers.ts:651

A single message in an LLM conversation.

Remarks

Represents one turn in a multi-turn conversation. The SDK accumulates these messages when ConversationHistoryConfig is enabled and passes them to generateFromMessages.

Example

const messages: LLMMessage[] = [
  { role: 'system', content: 'You are a helpful assistant.' },
  { role: 'user', content: 'What is the weather today?' },
  { role: 'assistant', content: 'I cannot check the weather, but I can help with other questions!' },
  { role: 'user', content: 'Tell me a joke instead.' },
];

See

LLMProvider.generateFromMessages for using message arrays
ConversationHistoryConfig for enabling multi-turn history

Properties

Property	Type	Description	Defined in
`content`	`string`	The text content of the message.	src/core/types/providers.ts:664
`modality?`	`"voice"` \| `"text"`	How the message was produced. Remarks - `'voice'` - Transcribed from speech via STT (default for user messages) - `'text'` - Typed by the user via `sendMessage()` This is included in the conversation history so the LLM can adapt its response style (e.g., shorter for voice, richer formatting for text). Only meaningful for `role: 'user'` messages.	src/core/types/providers.ts:677
`role`	`"system"` \| `"user"` \| `"assistant"` \| `"tool"`	The role of the message author. Remarks - `'system'` - System instructions (typically the first message) - `'user'` - User input (transcribed speech or typed text) - `'assistant'` - Assistant response (LLM output) - `'tool'` - Tool execution result (sent back after a tool call)	src/core/types/providers.ts:661
`toolCallId?`	`string`	The tool call ID this result belongs to (on tool messages).	src/core/types/providers.ts:683
`toolCalls?`	`LLMToolCall`[]	Tool calls requested by the assistant (on assistant messages).	src/core/types/providers.ts:680