AI Features and Data Usage
MuleSoft AI features use your data to generate responses. Understanding what data each feature uses helps you make informed decisions about privacy, compliance, and how you work with AI in MuleSoft.
By default, MuleSoft AI features access LLMs within the Salesforce Trust Boundary, a set of security and data governance controls that keeps your data within Salesforce-managed infrastructure. MuleSoft AI features process data you submit securely within Salesforce-managed boundaries and route it through your connected Salesforce organization for tracking and billing. If you configure or bring your own LLM (BYO-LLM), data processing is governed by your LLM provider’s terms. For clarity and without limitation, the various third-party machine learning and generative AI model building platforms that are supported within the MuleSoft Services and allow you to BYO-LLM are Non-SFDC applications.
This table lists the Customer Data you submit to our services, as defined in our Main Services Agreement (MSA), and usage data about your interactions with MuleSoft AI features. Customer Data that you submit to MuleSoft AI features is not used to train AI models or shared across customers.
For more information about data usage, see Customer Data and Usage Data in the MSA. For billing information, see Agentforce and Generative AI Usage and Billing.
| Product | Feature | Data Used | Uses Generative AI |
|---|---|---|---|
Intelligent Document Processing (IDP) |
Document Analysis with Custom Schemas |
This feature uses documents that you upload in supported formats (images, PDFs, and TIFFs) as input, along with natural language prompts and questions that you provide. It also uses document schema metadata that you define, such as field names, descriptions, and data types, and sample data that conveys the expected output format. |
Yes
|
MuleSoft Vibes |
Agent-Driven Development (flow generation, API specification generation, asset publishing, and more) |
This feature uses natural language prompts that you type and, with your permission, project files such as code, requirements, and configuration, along with context that you attach such as logs, terminal output, and URLs. It grounds responses with MuleSoft metadata, such as connector and component descriptions, API design documentation, and public Anypoint Exchange metadata. |
Yes Anthropic Claude Sonnet 3.5 with OpenAI GPT-4o Mini for planning (via Amazon Bedrock) |
DataWeave Generation |
This feature uses input and output sample data that you provide, such as JSON, XML, and CSV, along with optional structure metadata and natural language instructions. It grounds responses with MuleSoft DataWeave function and operator documentation. |
Yes OpenAI GPT-4o |
|
Anypoint Exchange |
API Documentation Generation |
This feature uses API specifications that you upload, such as RAML, OAS, SOAP, WSDL, AsyncAPI, GraphQL, and gRPC, along with asset metadata such as name, description, tags, and version. |
Yes OpenAI GPT-5 |
Connector Builder |
AI-assisted connector generation |
This feature uses OAS and API specifications that you provide, along with your natural language prompts and connector configuration metadata such as connections, operations, pagination, triggers, and value providers. It uses this data during generation, testing, packaging, and release phases of your connector project. |
Yes User’s configured model |
Enhanced MuleSoft Experience |
MuleSoft Agent |
This feature uses natural language prompts that you type, along with the current page context within the enhanced experience and metadata accessible under your Access Management permissions, such as portfolio, governance, monitoring, cost, organization, and environment information. |
Yes OpenAI GPT-5 mini |
Partner Manager |
View Message Summaries |
This feature uses the source or target payload that you open in Activity and summarize, including pre-translated and post-translated message content such as EDI and mapped JSON or XML. It also uses message context metadata, such as message type, sender, and receiver, along with extracted business data in the payload, such as order number, customer name, order details, and configured custom attributes like PO number or invoice number. |
Yes Einstein Generative AI |
Runtime Manager |
Diagnostics Agent |
This feature uses Java thread dumps from your deployed applications and the Diagnostic Information Analysis File (DIAF), which includes application metadata, performance statistics, error messages and stack traces, and recent log entries. It also uses CloudHub or CloudHub 2.0 instance information and deployment configuration. |
Yes Google Gemini 2.5 Flash |