Two processing paths: the CAB app and the optional CAB AI Agent
The CAB app (managed package) builds and evaluates audiences in Salesforce. The CAB AI Agent is optional and has a separate, administrator-controlled processing boundary.
The AI Agent does not change where the core audience engine runs. It adds an optional service with its own activation, permissions, and data controls.
The CAB app runs inside your Salesforce org
The CAB app builds and evaluates audiences in your Salesforce org. It does not copy Salesforce records into a separate segmentation database.
- Built and managed inside Salesforce.
- Audience logic is evaluated by the CAB app.
The CAB AI Agent runs on Google Cloud, only when activated
When an administrator enables the AI Agent, approved context is processed by CAB services on Google Cloud and Google Vertex AI under the selected AI data processing mode.
The AI Agent is separately activated. An administrator must choose the AI data processing mode and accept the current terms before activation; users also need the Agent permission set.
Processor and subprocessors
LightBlaze is the processor. Google Cloud services are subprocessors:
- Cloud Run for compute.
- Firestore for storage.
- Secret Manager for credentials.
- Google Vertex AI for Gemini and embeddings.
- Google Maps Platform, only when geolocation filtering is used.
Data the AI Agent can receive
The AI Agent uses the context required to turn a request into a relevant audience configuration. The current Agent terms list every data category.
- User prompts and conversation context.
- Salesforce organization identity and schema metadata.
- Audience configurations, including filter values.
- Standard and administrator-approved Campaign metadata.
- Aggregate schema statistics and usage patterns.
- Administrator-defined Agent instructions.
Automated schema collection does not send individual Salesforce record values. Personal data can still be included when a user or administrator enters it in prompts, filter values, names, custom instructions, or approved Campaign fields.
What administrators control
CAB makes the consequential data choices visible in Agent setup, before the AI Agent is available to users.
Outcome data sharing
Off by default. When enabled, CAB sends aggregate outcome counts and definition metadata. It does not send member lists, individual record IDs, individual names, or individual Salesforce records.
Geolocation filtering
When enabled, location strings entered by users are sent to the Google Maps Geocoding API. Salesforce record data is not included in those requests.
Campaign context
The AI Agent receives a defined set of standard Campaign fields. Administrators explicitly allow any additional custom Campaign fields through a Salesforce field set.
EU or global AI processing
Administrators choose EU processing for Google Vertex AI model and embedding requests or a global endpoint optimized for availability and performance. The global option may process requests outside the EU or EEA.
How Agent data is stored and protected
Selected Agent data is stored in LightBlaze-operated Google Cloud resources for session continuity and suggestions. This can include conversations, audience configurations, Campaign context, embeddings, interaction outcomes, and optional aggregate outcome statistics.
Data is encrypted in transit and at rest. Organizations use tenant-specific authentication.
Customer data is not used for model training. Users remain responsible for reviewing and testing AI-generated audience configurations before use.
How long Agent data is kept and how to erase it
The July 2026 in-app Agent terms specify 24 months from the organization's last Agent interaction.
Administrators can deactivate the AI Agent and request erasure of the organization's stored Google Cloud Agent data. Deactivation without erasure retains stored data for up to that 24-month period.
Agent terms and DPA
The current AI Agent terms, last updated in July 2026, are presented inside CAB setup and must be accepted for the selected processing mode before activation. They describe the data categories, processing roles, subprocessors, storage, retention commitments, and deletion rights. A public web copy is not currently published.
A DPA is available on request where applicable. A self-service download is not currently published on this site. Request it by email.