Amazon Q in Connect now allows LLM configuration through a GUI

Amazon Q in Connect now allows LLM configuration through a GUI

2025.09.08

Introduction

Amazon Q in Connect now allows selection of large language models (LLMs) for AI prompts through the GUI. This update isn't mentioned in AWS update information such as What's New with AWS, but it can be confirmed through actual use.

In May 2025, the number of LLMs available for selection in Amazon Q in Connect increased, but as introduced in the article below, at that time it was not possible to change models through the GUI.

https://dev.classmethod.jp/articles/amazon-q-in-connect-llm-comparison-claude-3-5-sonnet/

In this article, we will check the details of this update.

How to configure models

For new creations

For new creations, after creating an AI prompt, I was able to specify a model through the GUI in the AI prompt builder.

cm-hirai-screenshot 2025-09-08 8.44.19

cm-hirai-screenshot 2025-09-08 8.45.13

For existing prompts

For already created custom AI prompts, I was also able to specify the model through the GUI in the AI prompt builder.

cm-hirai-screenshot 2025-09-08 9.04.00

After specifying the model for the custom AI prompt, save and publish it. (A new version will be automatically created.)
cm-hirai-screenshot 2025-09-08 10.02.14
In the custom AI agent, specify the version of the custom AI prompt you created, then save and publish. (A new version will be automatically created.)
cm-hirai-screenshot 2025-09-08 10.04.31

You can apply the changes by setting the latest version of the custom AI agent as the default.
cm-hirai-screenshot 2025-09-08 10.04.55## Default AI Prompts

At the time of writing, although not documented in AWS documentation, different model IDs were set for each prompt in the default AI prompts.

Answer Recommendations

  • QinConnectAnswerGenerationPrompt
    • Type: Answer Generation
    • Model ID: apac.anthropic.claude-3-5-sonnet-20241022-v2:0
  • QinConnectIntentLabelingGenerationPrompt
    • Type: Intent Labeling Generation
    • Model ID: apac.amazon.nova-lite-v1:0
  • QinConnectQueryReformulationPrompt
    • Type: Answer Generation
    • Model ID: apac.amazon.nova-lite-v1:0

Content Search

  • QinConnectAnswerGenerationPrompt
    • Type: Answer Generation
    • Model ID: apac.anthropic.claude-3-5-sonnet-20241022-v2:0

Self-service

  • QinConnectSelfServiceAnswerGenerationPrompt
    • Type: Self Service Answer Generation
    • Model ID: apac.amazon.nova-pro-v1:0
  • QinConnectSelfServicePreProcessingPrompt
    • Type: Self Service Pre Processing
    • Model ID: apac.amazon.nova-pro-v1:0

cm-hirai-screenshot 2025-09-08 9.06.47

Anthropic's model is set for the answer generation part, while Amazon's models are set for the others.

https://docs.aws.amazon.com/ja_jp/connect/latest/adminguide/create-ai-prompts.html

Notes on Custom AI Agents

For custom AI agents, the default AI prompts described above are applied to prompt types where you haven't specified AI prompts, so it's recommended to check these settings.

Since default AI prompts may change in content, it's generally recommended to specify all prompts as custom AI prompts.

For example, consider a self-service type custom AI agent where you've specified a custom AI prompt for "Self Service Pre Processing" but not for "Self Service Answer Generation." In this case, the default AI prompt described above will be applied to the "Self Service Answer Generation" prompt type.

cm-hirai-screenshot 2025-09-08 9.11.46### Checking Custom AI Prompts Before Update

The custom AI prompt created in the article "Amazon Q in Connect's AI prompt settings can now be customized from the management screen" around March 2025, when the only LLM available was anthropic.claude-3-haiku-20240307-v1:0, naturally used the LLM anthropic.claude-3-haiku-20240307-v1:0.

https://dev.classmethod.jp/articles/amazon-q-in-connect-customize-ai-prompt-gui/

cm-hirai-screenshot 2025-09-08 8.43.16### Default Models for Current Custom AI Prompts

Now that multiple LLMs can be selected, I've checked the default model when creating custom AI prompts.

For example, when the prompt type is ANSWER_GENERATION, Claude 3.5 Sonnet is set as the default, just like with the default AI prompt.

  • apac.anthropic.claude-3-5-sonnet-20241022-v2:0 (cross-region) (system default)
    cm-hirai-screenshot 2025-09-08 8.44.19
    cm-hirai-screenshot 2025-09-08 8.45.13

I also checked other AI prompt types and found that they use the same models as the default AI prompts.

Recommended Answers

  • QinConnectAnswerGenerationPrompt
    • Type: Answer Generation
    • Model ID: apac.anthropic.claude-3-5-sonnet-20241022-v2:0
  • QinConnectIntentLabelingGenerationPrompt
    • Type: Intent Labeling Generation
    • Model ID: apac.amazon.nova-lite-v1:0
  • QinConnectQueryReformulationPrompt
    • Type: Answer Generation
    • Model ID: apac.amazon.nova-lite-v1:0

Content Search

  • QinConnectAnswerGenerationPrompt
    • Type: Answer Generation
    • Model ID: apac.anthropic.claude-3-5-sonnet-20241022-v2:0

Self-Service

  • QinConnectSelfServiceAnswerGenerationPrompt
    • Type: Self Service Answer Generation
    • Model ID: apac.amazon.nova-pro-v1:0
  • QinConnectSelfServicePreProcessingPrompt
    • Type: Self Service Pre Processing
    • Model ID: apac.amazon.nova-pro-v1:0

As introduced in the article below, apac.anthropic.claude-sonnet-4-20250514-v1:0 (cross-region) is now available, so I recommend selecting this model instead of apac.anthropic.claude-3-5-sonnet-20241022-v2:0.

https://dev.classmethod.jp/articles/amazon-q-connect-claude-4-sonnet-ap-northeast-1/

Finally

The ability to select LLM models via GUI has made model comparison during testing easier, improving development efficiency.

When using custom AI agents, please make sure to check the settings, as default AI prompts will be applied to prompt types where an AI prompt is not specified.

Since default AI prompts may have their content changed, it's generally recommended to specify all prompts with custom prompts.

Also, it's generally better to select the next-generation version of a model (such as upgrading from Claude 3.5 to 4 for Claude Sonnet) to expect higher performance.

Share this article

FacebookHatena blogX

Related articles