Org-level model enablement and blocking
Key Takeaway
Turn off a model provider to block all of its models everywhere in the org. Hide an LLM configuration to stop one model from being picked in Prompt Builder without breaking the prompt templates that already use it.
Admins have two main ways to control which large language models (LLMs) people can use in a Salesforce org. Turning off a model provider in Einstein Setup blocks every model from that provider across the org. Hiding an LLM configuration only removes that model from the choices in Prompt Builder. The right choice depends on whether you need a hard block or just want to steer prompt template authors toward a different model.
Model provider access: an org-wide switch for each provider
Model provider access is an org-wide on/off switch for each model provider. When a provider is on, the org can use its LLMs in generative AI features, including agents, prompt templates and APIs. When a provider is off, its models are blocked in the org.
Model providers are usually turned on by default. Salesforce-managed models are also turned on by default to speed up configuration. As a result, an org can use these models until an admin turns a provider off.
Turning off a provider is disruptive, because the affected components stop running. Affected areas include:
- Agentforce Agents
- Agent Actions
- Agentforce Testing Center
- Prompt templates
- APIs and the Agentforce API
- The Agentforce Vibes Extension
- Features such as:
- Email Summaries
- Knowledge Generation
- Service Replies
- Generative Campaigns
- Sales Summaries
The block isn't instant. Turning off a model provider can take up to 10 minutes to take effect.
Amazon Bedrock is a special case. When you turn off Amazon Bedrock, the change currently affects only Anthropic and Amazon models.