Choose which Large Language Model (LLM) providers to allow or not allow in your organization. When access to a model provider is turned on, you can use its language learning models (LLMs) in agents, prompt templates, APIs, and other features in generative AI solutions. Turn off a model provider to block access to its models in your org.
Required Editions
| Data Cloud in Developer, Enterprise, Performance, and Unlimited editions. |
| Einstein generative AI is available in Lightning Experience. Setup for Einstein generative AI is available in Lightning Experience. |
| Available in: Enterprise, Performance, and Unlimited Editions with an Einstein for Sales, Einstein for Platform, Einstein for Service, Einstein 1 Service, or Einstein GPT Service add-on. To purchase add-ons, contact your Salesforce account executive. |
| User Permissions Needed | |
|---|---|
| To change model provider access: | Admin permissions |
Warning Turning off a model provider can stop Agentforce agents, prompt templates,
and other generative AI features that use its models from running. Be sure to test and verify
AI-enabled functionality in a sandbox org before you deploy it into production. Affected
components can include (but aren’t limited to):
- Agentforce Agents
- Agent Actions
- Agentforce Testing Center
- Prompt templates
- APIs
- Email Summaries
- Knowledge Generation
- Service Replies
- Generative Campaigns
- Sales Summaries
- Agentforce API
- Agentforce Vibes Extension
- Salesforce features that access LLMs
To manage model provider access:
- From Setup, in the Quick Find box, enter Einstein Setup.
- Turn on Einstein if it’s not already turned on.
-
Scroll down to Configure Model Providers.
-
Selectively turn on or off model providers.
Usually, model providers are enabled by default. For Amazon Bedrock, currently only Anthropic and Amazon models are affected.
-
If you turn off access for a model provider, you’re warned that disabling access stops affected components from running. Confirm your selection.
Important It can take up to ten (10) minutes for model provider disablement to take effect.