AI Engine works with OpenAI, Anthropic (Claude) and Azure OpenAI. The features are the same with all three, apart from how web search works with Azure OpenAI (see below). What differs is who processes your data, under which agreement and where. This article helps admins and compliance teams choose. It is not legal advice; check the final choice against your own policies and your provider's terms.
What is the same for every provider#
- You hold the account and the key. AI Engine calls the provider through a Named Credential in your org. Cloud Connected is not a party to the data flow and does not receive your data.
- AI Engine applies the same controls before anything is sent: the user's Salesforce permissions, the Agent Objects scope and PII masking. See What is sent to the AI provider.
- AI Engine does not store full prompts or answers in Salesforce, whichever provider you use.
How the providers differ#
| OpenAI | Anthropic | Azure OpenAI | |
|---|---|---|---|
| Who processes requests | OpenAI, under your OpenAI API account | Anthropic, under your Anthropic API account | Microsoft, in the Azure OpenAI resource you create in your own Azure subscription |
| Where | OpenAI's infrastructure | Anthropic's infrastructure | The Azure region you chose for the resource |
| Agreement | Your agreement with OpenAI | Your agreement with Anthropic | Your existing agreement with Microsoft |
| Models | Chosen from a list in the Org Profile | Chosen from a list in the Org Profile | Your own deployments, entered by deployment name |
| Web search | OpenAI's web search tool | Anthropic's web search tool | Not available through Azure OpenAI; see below |
Retention periods, use of API data for training, and data processing terms are set by each provider and can change. Review them in your provider agreement before go-live.
When Azure OpenAI is the better fit#
If your security or data protection team requires AI processing to stay inside your own cloud tenant and a region you choose, Azure OpenAI is usually the most direct answer. Requests go to your Azure OpenAI resource, under your Microsoft agreement. Note the exception for web search below.
Web search on Azure OpenAI#
Azure OpenAI does not provide the web search tool that AI Engine uses. In an org configured for Azure OpenAI, web-search features call the OpenAI API instead, through a Named Credential named OpenAI and with OpenAI's gpt-4.1-mini model. These features are:
- Web Check and Company News
- research questions in the chat, such as "research company X"
- Meet Prep: include web search
- Verify Company On The Web in AI Scoring
- Research this company and Find companies in Magic Snap
If all AI processing must stay in Azure, do not create a Named Credential named OpenAI, keep Meet Prep: include web search and Verify Company On The Web off, and ask users not to use the web-search features. Without that credential, these web searches fail and nothing is sent to OpenAI. If you do create it, the searches and their subjects (typically company names) go to OpenAI under your OpenAI account.
Switching providers#
Changing provider is a configuration change; no reinstall is needed.
- Set up and test a Named Credential for the new provider; see API keys and Named Credentials.
- Open the Ai Agent Admin Setup tab → Org Profile and open your profile.
- Choose the AI Provider, then the AI Model and the Fast model (internal steps). The model options depend on the provider; for Azure OpenAI, enter your deployment names.
- Enter the new provider's Named Credential name and click Save.
Keeping a second provider's Named Credential set up and tested lets you switch quickly during an outage or a policy change.
After any change of provider or model, test the features that return structured results before users rely on them: Meet Prep, AI Score and answers shown as cards.