Message

Message class for conversation management.

Overview

The Message class represents a single message in a conversation with support for multiple attachments.

import { Message } from 'multi-llm-ts'

const message = new Message('user', 'Hello, world!')

See the Messages Guide for detailed usage.

Constructor

new Message(role: string, content: string)

Parameters:

• role: Message role ('system', 'user', 'assistant')
• content: Message text content

Example:

const system = new Message('system', 'You are helpful')
const user = new Message('user', 'What is TypeScript?')
const assistant = new Message('assistant', 'TypeScript is...')

Properties

role

Message role:

message.role: 'system' | 'user' | 'assistant'

content

Message text content:

message.content: string

attachments

Array of attachments:

message.attachments: Attachment[]

transient

Exclude from conversation history:

message.transient?: boolean

Methods

attach()

Add an attachment:

attach(attachment: Attachment): void

Example:

message.attach({
  url: '/path/to/image.jpg',
  mimeType: 'image/jpeg'
})

detach()

Remove an attachment:

detach(attachment: Attachment): void

Example:

message.detach(attachment)

setContent()

Update message content:

setContent(content: string): void

Example:

message.setContent('Updated text')

appendText()

Append text to content:

appendText(text: string): void

Example:

message.appendText(' Additional text.')

Attachment

Attachment object structure:

interface Attachment {
  url: string           // File path or URL
  mimeType: string      // MIME type
  downloaded?: boolean  // Auto-populated
  content?: string      // Base64 content (auto-populated)
}

Supported MIME Types

Images:

• image/jpeg
• image/png
• image/webp
• image/gif

Text:

• text/plain
• text/html
• text/markdown

Other:

• application/pdf
• Provider-specific types

Examples

Basic Message

const message = new Message('user', 'Hello!')

System Message

const system = new Message('system', 'You are a helpful coding assistant')

Message with Image

const message = new Message('user', 'Describe this image')
message.attach({
  url: 'photo.jpg',
  mimeType: 'image/jpeg'
})

Multiple Attachments

const message = new Message('user', 'Compare these images')
message.attach({ url: 'image1.jpg', mimeType: 'image/jpeg' })
message.attach({ url: 'image2.png', mimeType: 'image/png' })

Transient Message

const message = new Message('user', 'One-time instruction')
message.transient = true

Modifying Content

const message = new Message('user', 'Hello')
message.appendText(' World')
message.setContent('Hello Universe')

Conversation Building

Single Turn

const messages = [
  new Message('system', 'You are helpful'),
  new Message('user', 'What is JavaScript?')
]

const response = await model.complete(messages)

Multi-Turn

const conversation = [
  new Message('system', 'You are helpful')
]

// Turn 1
conversation.push(new Message('user', 'Tell me about TypeScript'))
const response1 = await model.complete(conversation)
conversation.push(new Message('assistant', response1.content))

// Turn 2
conversation.push(new Message('user', 'How do I use interfaces?'))
const response2 = await model.complete(conversation)

With Vision

const message = new Message('user', 'Analyze this code screenshot')
message.attach({
  url: 'code_screenshot.png',
  mimeType: 'image/png'
})

const response = await model.complete([message])

Role Types

System

Instructions and context for the model:

new Message('system', 'You are a helpful assistant specializing in...')

Best practices:

• Place first in conversation
• Keep concise
• Define personality and constraints

User

User input and prompts:

new Message('user', 'What is the capital of France?')

Assistant

Model responses in multi-turn conversations:

new Message('assistant', 'The capital of France is Paris.')

Attachment Loading

Attachments are loaded automatically:

const message = new Message('user', 'Read this file')
message.attach({
  url: '/local/document.txt',
  mimeType: 'text/plain'
})

// File is read and base64-encoded during generation
await model.complete([message])

Transient Messages

Messages marked as transient are excluded from history:

const instruction = new Message('user', 'Use simple language')
instruction.transient = true

const messages = [
  new Message('system', 'You are helpful'),
  instruction,  // Won't persist in conversation
  new Message('user', 'Explain quantum physics')
]

Next Steps

• Learn about Messages usage
• See Vision for image attachments
• Review Completion for generation