Retrieval-Augmented Generation Models in NetSuite 2026.1

TL;DR Opening

Starting in the Weekly Update on February 16, 2026, NetSuite introduces support for Retrieval-Augmented Generation (RAG) models within the context of its SuiteScript. This feature allows users to enhance LLM responses with additional document context, improving the accuracy and relevance of generated content.

What's New

The update expands the capabilities of the llm.generateText(options) and llm.generateTextStreamed(options) functions, providing models that support RAG and preambles. These new features allow developers to give context to language models, leading to more accurate outputs relevant to their needs.

Supported Models and Capabilities

The table below outlines the models that are enabled for use with llm.generateText and llm.generateTextStreamed, highlighting their supported capabilities:

Model Family	Model Name	RAG Supported	Preambles Supported
ModelFamily.COHERE_COMMAND	cohere.command-a-03-2025	Yes	Yes
ModelFamily.COHERE_COMMAND_LATEST	cohere.command-a-03-2025	Yes	Yes
ModelFamily.GPT_OSS	openai.gpt-oss-120b	No	Yes
ModelFamily.GPT_OSS_LATEST	openai.gpt-oss-120b	No	Yes

Key Parameters

When utilizing these functions, here are the primary parameters:

options.prompt (required): The input prompt for the LLM.
options.chatHistory (optional): Previously exchanged messages to maintain context.
options.documents (optional): A list of llm.Document objects for additional context with Cohere models.
options.modelFamily (optional): Specifies which LLM to utilize based on the provided enum values.

Real-World Application

Using RAG effectively can significantly enhance the quality of responses generated by the LLM. By including relevant documents as context, the LLM can produce tailored outputs that feel more informed and relevant to specific business scenarios, thereby improving user experience.

Developers can leverage the preambles functionality to set the tone or context for the LLM, which can be particularly useful in customer-facing applications or chatbots that require a specific style or framing of information.

Who This Affects

Developers: Create advanced scripts utilizing the new LLM features.
Administrators: Set up parameters and manage settings for AI features.
Business Analysts: Implement insights derived from enhanced LLM outputs.

Key Takeaways

NetSuite's Weekly Update on February 16, 2026, rolls out support for RAG models and preambles.
Improved accuracy and contextual relevance for LLM responses through additional document support.
Key parameters are introduced, enabling advanced configurations in SuiteScript.
Enhanced user experiences in applications utilizing LLMs through tailored responses.
The functionality is incorporated into the SuiteScript 2.1 environment for developers to utilize effectively.

Frequently Asked Questions (4)

What permissions are required to use Retrieval-Augmented Generation (RAG) models in SuiteScript?

The article does not specify the permissions required. It is likely that standard SuiteScript permissions would apply, but please refer to NetSuite's documentation or your account administrator for precise permission needs.

Do I need to enable a feature flag to use the RAG models in the updated `llm.generateText` functions?

The article does not mention the need for a feature flag to access the RAG models introduced in SuiteScript. It seems they are available as part of the update, but checking with NetSuite support or documentation for any specific settings would be prudent.

Is the Retrieval-Augmented Generation (RAG) feature available for all LLM models support in NetSuite 2026.1?

No, not all models support Retrieval-Augmented Generation. The ‘cohere.command-a-03-2025’ model supports RAG, while the ‘openai.gpt-oss-120b’ model does not.

How does the new RAG capability affect existing workflows in SuiteScript?

The article notes that RAG can improve LLM output accuracy and relevance, potentially enhancing workflows that rely on these outputs. However, specific effects on existing workflows are not detailed and would depend on how they are currently implemented and what outputs they rely on.