Einstein Trust Layer architecture
Key Takeaway
Every request passes through the Einstein Trust Layer twice: once as a prompt on its way to the LLM, and once as a response on its way back to the CRM app.
The Einstein Trust Layer is a secure AI architecture built into the Salesforce platform. It combines agreements, security technology, and data and privacy controls, so a company can use generative AI without exposing its own data or its customers' data. The Trust Layer sits between the large language model (LLM) and your employees and customers. Its capabilities apply only to generative AI and Agentforce features.
The prompt journey and the response journey
The Trust Layer works as a checkpoint that every request passes through twice. A request moves through three stages:
- Prompt journey: A prompt travels from a CRM app through the Trust Layer to the LLM. On the way, the prompt is grounded and secured. Then it goes through the LLM gateway.
- Response generation: The LLM writes a response.
- Response journey: The response travels back through the Trust Layer to the CRM app. On the way, the Trust Layer scans the response for toxicity and then demasks it.