Supported LLMs for Prompt Builder and Agentforce
Key Takeaway
To block a provider's models everywhere in the org, including in Agentforce agents, turn off that provider in Einstein Setup. Hiding an LLM configuration in AI Models only keeps that one model out of Prompt Builder.
Prompt Builder and Agentforce run on large language models (LLMs) from several providers. Salesforce-managed models are on by default. You can also bring your own model. Agentforce agents choose from their own, smaller set of model options. Admins control which models people can use in two ways: turn off a whole model provider in Einstein Setup, or hide a single LLM configuration in AI Models. Turning off a provider reaches across the org, while hiding a configuration reaches only Prompt Builder.
Salesforce-managed models and your own models
Salesforce-managed models are available out of the box and enabled by default for embedded features such as Prompt Builder. Their providers include:
- Amazon Bedrock, with Amazon, Anthropic, and NVIDIA models
- OpenAI and Azure OpenAI
- Vertex AI, with Google models
A common reason to use BYOLLM is that your company has an LLM fine-tuned to its own data. For models and providers that BYOLLM doesn't cover, use the LLM Open Connector.
Third-party LLMs don't retain your data. Salesforce partners with its external model providers, such as OpenAI, Anthropic, and Google, to enforce a zero-data retention policy.