Skip to main content

​
Overview

The ModelDefinition type defines AI models (LLMs, embeddings, etc.) that can be used within the Atomemo platform. Models describe their capabilities, supported modalities, and limitations.

​
Type Definition

type ModelDefinition = {
  name: string
  display_name: Record<string, string>
  description: Record<string, string>
  icon: string
  model_type: "llm" | "embedding" | "image" | "audio" | "video"
  input_modalities: Modality[]
  output_modalities: Modality[]
  unsupported_parameters: string[]
  credentials?: string[]
}

type Modality = "text" | "image" | "audio" | "video"

​
Properties

​
name
string
required
The unique identifier for the model, typically in the format provider/model-name.Example: "openai/gpt-4" or "anthropic/claude-3-opus"
​
display_name
Record<string, string>
required
Localized display names for the model.Example:
{
  en_US: "GPT-4 Turbo",
  es_ES: "GPT-4 Turbo"
}
​
description
Record<string, string>
required
Localized descriptions explaining the model’s capabilities.Example:
{
  en_US: "Advanced language model with vision capabilities",
  es_ES: "Modelo de lenguaje avanzado con capacidades de visión"
}
​
icon
string
required
An emoji or icon representing the model.Example: "🤖" or "🧠"
​
model_type
string
required
The category or type of the model.Options:
  • "llm" - Large Language Model (text generation, chat)
  • "embedding" - Text embedding model
  • "image" - Image generation or processing model
  • "audio" - Audio processing model
  • "video" - Video processing model
​
input_modalities
Modality[]
required
Array of input types the model can accept.Options: "text", "image", "audio", "video"Example: ["text", "image"] for a multimodal LLM
​
output_modalities
Modality[]
required
Array of output types the model can generate.Options: "text", "image", "audio", "video"Example: ["text"] for a text-only model
​
unsupported_parameters
string[]
required
List of standard parameters that this model does not support.Common parameters:
  • "temperature" - Controls randomness
  • "top_p" - Nucleus sampling parameter
  • "max_tokens" - Maximum output length
  • "stop" - Stop sequences
  • "frequency_penalty" - Penalizes frequent tokens
  • "presence_penalty" - Penalizes repeated tokens
Example: ["frequency_penalty", "presence_penalty"]
​
credentials
string[]
Optional array of credential names required to use this model. The credentials must be registered with the plugin using addCredential().Example: ["api-key"]

​
Usage

Models are registered using the addModel() method on the plugin instance: 
import { createPlugin } from "@choiceopen/atomemo-plugin-sdk-js"

const plugin = await createPlugin({
  name: "openai-plugin",
  display_name: { en_US: "OpenAI Plugin" },
  description: { en_US: "Access OpenAI models" },
  icon: "🤖",
  locales: ["en_US"]
})

plugin.addModel({
  name: "openai/gpt-4-turbo",
  display_name: {
    en_US: "GPT-4 Turbo"
  },
  description: {
    en_US: "Advanced language model with vision capabilities"
  },
  icon: "🤖",
  model_type: "llm",
  input_modalities: ["text", "image"],
  output_modalities: ["text"],
  unsupported_parameters: [],
  credentials: ["openai-api-key"]
})

​
Example: LLM Model

plugin.addModel({
  name: "anthropic/claude-3-opus",
  display_name: {
    en_US: "Claude 3 Opus"
  },
  description: {
    en_US: "Anthropic's most powerful model for complex tasks"
  },
  icon: "🧠",
  model_type: "llm",
  input_modalities: ["text", "image"],
  output_modalities: ["text"],
  unsupported_parameters: ["frequency_penalty", "presence_penalty"],
  credentials: ["anthropic-api-key"]
})

​
Example: Embedding Model

plugin.addModel({
  name: "openai/text-embedding-3-large",
  display_name: {
    en_US: "Text Embedding 3 Large"
  },
  description: {
    en_US: "High-performance text embedding model"
  },
  icon: "📊",
  model_type: "embedding",
  input_modalities: ["text"],
  output_modalities: ["text"],
  unsupported_parameters: [
    "temperature",
    "top_p",
    "max_tokens",
    "frequency_penalty",
    "presence_penalty"
  ],
  credentials: ["openai-api-key"]
})

​
Example: Image Generation Model

plugin.addModel({
  name: "openai/dall-e-3",
  display_name: {
    en_US: "DALL-E 3"
  },
  description: {
    en_US: "Advanced image generation model"
  },
  icon: "🎨",
  model_type: "image",
  input_modalities: ["text"],
  output_modalities: ["image"],
  unsupported_parameters: [
    "temperature",
    "top_p",
    "frequency_penalty",
    "presence_penalty"
  ],
  credentials: ["openai-api-key"]
})

​
Example: Model Without Credentials

plugin.addModel({
  name: "local/llama-3-8b",
  display_name: {
    en_US: "Llama 3 8B (Local)"
  },
  description: {
    en_US: "Locally hosted Llama 3 model"
  },
  icon: "🦙",
  model_type: "llm",
  input_modalities: ["text"],
  output_modalities: ["text"],
  unsupported_parameters: ["stop"]
  // No credentials required for local models
})

​
Model Capabilities

The input_modalities and output_modalities properties define what the model can process:

​
Text-only LLM

input_modalities: ["text"]
output_modalities: ["text"]

​
Multimodal LLM (Vision)

input_modalities: ["text", "image"]
output_modalities: ["text"]

​
Image Generator

input_modalities: ["text"]
output_modalities: ["image"]

​
Audio Transcription

input_modalities: ["audio"]
output_modalities: ["text"]

​
Unsupported Parameters

Use unsupported_parameters to indicate which standard parameters the model doesn’t support. This helps the platform provide appropriate UI and prevent errors. 
unsupported_parameters: [
  "frequency_penalty",  // Model doesn't support frequency penalty
  "presence_penalty",   // Model doesn't support presence penalty
  "stop"                // Model doesn't support custom stop sequences
]

​
See Also

Build docs developers (and LLMs) love

Get started for freeTalk to us