Skip to main content

Assistant List

This document explains the assistant features of AI-Starter. Assistants help users create content tailored to their needs by combining various generative AI models and options.

Supported Models​

AI-Starter supports a variety of generative AI models, and the available models may vary depending on the environment.

Characteristics of Each Model​

While this is an evaluation by the development team, each model has different strengths and characteristics:

  • Amazon Nova series: Models that can be used at a low cost.
  • Claude series (AWS): Models that excel in coding and Japanese language processing.
  • Gemini series (Google): Models with excellent cost performance and long text comprehension. As of February 2025, these are the most notable models.
  • GPT-4 series (Azure & OpenAI): Models that excel in image recognition.
  • o series (Azure & OpenAI): Models with high logical thinking abilities.
  • DALL·E 3 (OpenAI): An AI model that generates images from text.

List of Supported Models​

  • AWS
    • Claude 3.7 Sonnet
    • Claude 3.7 Sonnet Extended thinking Mode (Beta)
    • Claude 3.5 Sonnet v2
    • Claude 3.5 Haiku
    • Claude 3.5 Sonnet
    • Claude 3 Haiku
    • Nova Micro
    • Nova Lite
    • Nova Pro
  • Azure
    • GPT-4o mini
    • GPT-4o
    • o3 mini
    • o1 mini
    • o1
  • Google
    • Gemini 2.0 Flash
    • Gemini 2.0 Flash-Lite Preview
    • Gemini 2.0 Pro Experimental
    • Gemini 2.0 Flash Thinking Experimental
    • Gemini 1.5 Pro
  • OpenAI
    • GPT-4.5 Preview
      • GPT-4o mini
      • GPT-4o
      • o3 mini
      • o1 mini
      • o1
      • ChatGPT-4o latest (experimental model)
      • DALL·E 3

Notes:

  • Some models support video input, but this is not currently supported in AI-Starter.
  • Some Gemini series models support PDF file input, but this is not currently supported in AI-Starter.
  • GPT-4.5 Preview is a high-cost model, so by default, conversations are limited to one exchange. This limitation can be changed in the admin settings.
  • Claude 3.7 Sonnet Extended thinking Mode (Beta) is currently in beta. Due to AI-Starter specifications, conversations are limited to one exchange, but this restriction is expected to be relaxed in the future.

Related Information

Options​

Assistants can use various options in combination with generative AI models. By utilizing these options, more advanced content creation is possible. For enabling settings, please contact the support desk through your internal administrator.

Option NameDescription
KendraRetrievalRAG (Retrieval Augmented Generation) option using Amazon Kendra.
GoogleCustomSearchRetrievalWeb search option using Google Custom Search.
BingSearchRetrievalWeb search option using Bing Search API.
WebPagesOption to retrieve information from web pages using scraping techniques.
StorageFilesOption to use locally uploaded files as input.

KendraRetrieval​

By using the KendraRetrieval option, you can perform RAG (Retrieval Augmented Generation) using Amazon Kendra.

In RAG, when a user asks a question, documents containing the necessary information for the answer are searched, and that document is added to the input (prompt) for the generative AI to generate an answer based on the information the user is seeking. This enables answers based on specific internal information such as internal system manuals and FAQs.

To use this, you need to create an Amazon Kendra index and specify the index ID for the assistant.

GoogleCustomSearchRetrieval​

Using the GoogleCustomSearchRetrieval option allows you to get answers based on Google Custom Search results. To use this, you need to set the following information in environment variables:

Details

Precautions when using Custom Search Google Custom Search collects data during requests and responses. When using it, it is recommended to advise users not to input sensitive information. Custom Search Engine Terms of Service

Google and its wholly-owned subsidiaries may retain and use, subject to the terms of its privacy policy (http://www.google.com/privacy.html or such other URL as Google may provide from time to time), information collected in your use of the Service.

BingSearchRetrieval​

Using the BingSearchRetrieval option allows you to get answers utilizing Bing Search API search results. To use this, you need to set the Bing Search API v7 API key in environment variables.

Details

Precautions when using Bing Search API Bing Search API collects data during requests and responses. When using it, it is recommended to advise users not to input sensitive information. Bing Search API Legal Information

Microsoft may collect and use information related to transaction requests to the Service, including but not limited to End User IP address, request, time of request, and results returned.

WebPages​

Using the WebPages option allows you to retrieve information through web page scraping. This is a technique that analyzes the HTML structure of web pages and extracts necessary information.

With the WebPages option, you can retrieve information in the following ways:

  • Collect information from pre-specified web page URLs
  • Collect information from URLs input by users

StorageFiles​

Using the StorageFiles option allows you to use files uploaded locally as input for the assistant. For example, you can upload a PDF file and ask questions about its contents. You can choose which settings to enable for each assistant.

For details, please refer to Supported File Formats for Attachments.

ContentRecommended Model
General Q&AClaude 3.7 Sonnet
Business document creation supportGemini 2.0 Flash
Coding support for system developmentClaude 3.5 Sonnet
Translation supportGemini 2.0 Flash
Long text comprehensionGemini 2.0 Flash
Questions about imagesGPT-4o
Complex questionso1
Questions about text filesGemini 2.0 Flash (+ StorageFiles)
Questions about PDF filesClaude 3.5 Sonnet (+ StorageFiles)
Questions about spreadsheet filesClaude 3.5 Sonnet (+ StorageFiles)
For initial testingGemini 2.0 Flash

Additional Information​

Comparison of Generative AI Models​

The following external sites are useful for comparing generative AI models. However, as information changes frequently, always check for the latest information and reliability: