AI provider integration

When you configure the Reasoning Service with AI providers, you need to obtain API keys and select appropriate models for your deployment.

The Reasoning Service supports integration with multiple AI providers, including Microsoft Azure AI Foundry and IBM watsonx.ai. Understanding model capabilities and limitations help you make informed decisions about your deployment.

API key acquisition

Before you configure the Reasoning Service with an AI provider, you need to obtain an API key from the respective platform. Refer to your AI provider official documentation for instructions on API key generation and management.

Microsoft Azure AI Foundry

To obtain your Microsoft Azure AI Foundry API key, refer to the Microsoft Foundry External link opens a new window or tab documentation.

Note: API key configuration steps can vary depending on your Microsoft Azure subscription tier and region. Refer to the Microsoft Foundry External link opens a new window or tab documentation for instructions.
IBM watsonx.ai

To obtain your watsonx.ai API key, refer to the IBM watsonx External link opens a new window or tab documentation.

Note: watsonx.ai API key setup requires an active IBM Cloud account. Refer to the IBM watsonx External link opens a new window or tab documentation for your deployment environment.

Recommended models for Content Cortex AI Services

The following models are available for use with Content Cortex AI Services.

Azure AI Foundry models

The following models are available for use with Content Cortex AI Services on Microsoft Azure AI Foundry.

Table 1. Azure AI Foundry supported models
Model Testing status
GPT-5.4 Tested and validated
GPT-5.5 Expected to perform well (not yet tested)
Claude Sonnet 4.6 Expected to perform well (not yet tested)
Important: GPT-5.5 and Claude Sonnet 4.6 are not tested with Content Cortex AI Services. Use these models at your own discretion and validate results in your environment before production deployment.

IBM watsonx.ai and lightweight engine models

The following model is supported in both watsonx.ai and the watsonx.ai lightweight engine:

Table 2. IBM watsonx.ai supported models
Model Support
gpt-oss-120b Supported in both watsonx.ai and the watsonx.ai lightweight engine

The gpt-oss-120b model has the following limitations when used with Content Cortex AI Services in a watsonx.ai environment:

Single-object operations only

Do not attempt to use the Core Agent to perform operations on more than one object at a time. All operations must be limited to a single object.

For example, you cannot reference more than one document when performing classification and property updates. Each document must be processed individually.

Use explicit, precise prompt language

Prompts must be explicit and exact. Vague or paraphrased phrasing produces inconsistent or incorrect results.

Recommended: "Find documents that use this exact phrase 'Indemnification clause'"

Not recommended: "Find documents that contain 'Indemnification clause'"

Avoid multiple actions in a single prompt

Do not chain or combine multiple actions into a single prompt. Break compound operations into separate, sequential prompts.

Recommended: Use two prompts:

  1. "Create the following folder..."
  2. "File these documents in [folder name]"

Not recommended: "Create the following folder and file these documents in that folder"