Large Language Model
A Large Language Model (LLM) is a deep-learning model trained on massive amounts of text that can understand and generate natural language at scale.
Definition
A Large Language Model (LLM) is a deep-learning model trained on massive amounts of text that can understand and generate natural language at scale. In the Salesforce ecosystem, LLMs are the underlying engine powering features like Agentforce agents, Einstein Copilot, and Prompt Builder across Sales, Service, Marketing, and Commerce Cloud. Salesforce exposes LLMs through the Einstein Trust Layer (the privacy and safety architecture formerly marketed alongside the Einstein GPT brand), which supports Salesforce-hosted foundation models as well as bring-your-own-model integrations from providers like OpenAI, Azure OpenAI, Anthropic, Google Vertex AI, and Amazon Bedrock. The Trust Layer adds data masking, prompt grounding, toxicity detection, and audit logging around every LLM call.
In plain English
“A Large Language Model, or LLM, is an AI model trained on huge amounts of text that can read and write human language. Salesforce uses LLMs to power Agentforce, Einstein Copilot, and Prompt Builder. You don't call the raw model directly - every request flows through the Einstein Trust Layer, which hides sensitive data, grounds answers in real CRM records, and logs each call for audit.”
Worked example
At Nexus Innovations, an AI specialist builds a Service Cloud assistant that summarizes long case threads for agents. Rather than calling OpenAI directly, they configure a Prompt Builder template that routes through the Einstein Trust Layer - the Trust Layer masks PII in the case content, grounds the prompt with the current Case record and related Contacts, sends the scrubbed prompt to the selected LLM (Salesforce-hosted or bring-your-own), and returns a summary with citations back to the underlying records. The summary renders inside the Case page, with nothing sensitive having left the secure boundary.
Why Large Language Model matters
A Large Language Model (LLM) is a type of deep-learning model trained on massive amounts of text that can understand and generate natural language at scale. In the Salesforce ecosystem, LLMs are the underlying AI technology powering features like Agentforce agents, Einstein Copilot, and Prompt Builder across Sales, Service, Marketing, and Commerce clouds. They're the engine behind the generative AI features Salesforce offers under the Einstein 1 Platform banner (the broader platform identity that succeeded the Einstein GPT brand used in 2023).
Salesforce exposes LLMs through the Einstein Trust Layer, which supports both Salesforce-hosted foundation models and bring-your-own-model integrations from providers like OpenAI, Azure OpenAI, Anthropic, Google Vertex AI, and Amazon Bedrock. The Trust Layer adds the safeguards that make LLM use viable in enterprise CRM: data masking (sensitive fields hidden before reaching the model), prompt grounding (responses anchored in real CRM records via Prompt Builder), toxicity detection (harmful outputs filtered), and audit logging (every call tracked). Without these, unprotected LLM access would be a regulatory and security non-starter for most Salesforce customers.
How organizations use Large Language Model
Uses Einstein Copilot powered by LLMs through the Einstein Trust Layer, with sensitive data automatically masked before reaching the underlying model.
Integrates a third-party LLM via the Trust Layer's bring-your-own-model support, gaining access to the model's capabilities while keeping Salesforce's safety controls in place.
Relies on the Einstein Trust Layer to handle LLM interactions safely for clinical use cases where data leakage would be unacceptable.
Trust & references
Straight from the source - Salesforce's reference material on Large Language Model.
- Agentforce and Einstein Generative AISalesforce Help
Test your knowledge
Q1. What is a Large Language Model?
Q2. How does Salesforce wrap LLM interactions?
Q3. What LLM providers does Salesforce support?
Discussion
Loading discussion…