Cabby settings
Every setting that controls what Cabby receives and how it answers, with its default and effect.
All settings are under CAB Settings → Configure → AI & Integrations → AI Agent Setup. To connect Cabby first, see Set up Cabby.
AI data processing mode
Section titled “AI data processing mode”AI Data Processing Mode is part of accepting the Agent terms:
| Choice | Processing boundary |
|---|---|
| Enforce EU data residency and processing | Keeps AI model and embedding requests on Google Cloud Platform infrastructure in the EU. Capacity can be more constrained, so responses, feature availability, or access to the latest models can arrive later. |
| Optimize availability and performance | Uses Google Cloud Platform’s global AI endpoint to improve availability and reduce capacity errors. Google may process requests outside the EU/EEA. |
The selected mode is attached to Agent requests rather than inferred from the current user’s location.
Metadata statistics
Section titled “Metadata statistics”Enable Metadata Statistics schedules a nightly schema analysis and runs the first collection immediately. Cabby receives accessible and filterable schema facts such as record counts, field population rates, and which fields are indexed. It does not receive raw field values from this job.
Keeping the job active helps Cabby prefer fields that exist, contain data, and are suitable for filtering. Unschedule stops future collections; Start enables them again.
Cabby data access
Section titled “Cabby data access”Under Cabby data access, the page states Changes take effect when you click Save. Select Save to apply a change to these controls.
Cabby always uses organization metadata, audience configurations, Campaign metadata, and aggregate schema statistics. Which record values Cabby can look up while it works is in Trust and security.
Outcome data
Section titled “Outcome data”Give Cabby access to outcome data is off by default. Turn it on so Cabby can explain results in Cabby’s summary on outcome pages and rank its audience suggestions by what actually performed in your org. It can be switched on once the Agent is activated.
When it is on, CAB shares totals from outcome results: how many people were measured and how many converted, per audience, campaign and outcome, for each results update, and the outcome names. When someone who may see custom metric results asks Cabby for a summary, Cabby also receives those results by the month people were added, with the metric names. A metric’s smallest and largest value each come from a single record. Record IDs and lists of people are never sent. The data stays in your org’s own storage in the Cabby service, is never shared across organizations, and is not used for model training.
Turning it on may require accepting the current Agent terms. Turning it off stops new sharing and deletes the stored outcome statistics. If deletion fails, use Retry deletion; sharing stays off until deletion is done. See Outcomes settings.
Filter Library data
Section titled “Filter Library data”Turn on Give Cabby access to Filter Library entries so Cabby can find and reuse approved library entries while it builds audiences. CAB shares the entry names and their filter definitions: object and field API names, operators, relationships and the values set in each filter. Individual records and query results are never included. The entries stay in your org’s own storage in the Cabby service, and filter values are not used for search indexing or usage logs.
Turning it off stops new sharing and deletes the stored entries. If deletion fails, use Retry deletion; you can turn sharing on again only after the deletion is done. See Set defaults and enforce filters.
Geolocation filtering
Section titled “Geolocation filtering”Under Enable Geolocation Filtering, turn on Allow Cabby to resolve locations via Google Maps only when the organization uses populated Address or Geolocation fields.
When enabled, location names that users type in prompts—for example, “contacts near Berlin”—are sent to the Google Maps Geocoding API to resolve coordinates. No Salesforce record data is included in those requests. Users can then ask Cabby for proximity filters; manual coordinate-based filtering remains available as described in Filter by location.
Instructions and response mode
Section titled “Instructions and response mode”Under Agent Settings:
- Add organization terminology, data-model conventions, and durable preferences under Custom Agent Instructions in the Instructions field. The limit is 32,768 characters. These edits are staged until you select the page-level Save; Cancel discards them.
- When the edition exposes Response Mode, choose Fast or Thinking. Treat this as a managed response preference, not as a promise of a particular underlying model. CAB can use the same model for both choices and can change routing as models evolve.
Campaign context
Section titled “Campaign context”Cabby’s Create with Cabby flow always receives standard Campaign context: Name, Type, Description, Start Date, End Date, Status, Parent Campaign name, Number of Contacts, and Number of Leads.
Deactivation and erasure
Section titled “Deactivation and erasure”| Choice | What happens |
|---|---|
| Leave the checkbox clear and select Deactivate | Credentials are revoked, but the organization’s stored GCP Agent data is retained. |
| Select the checkbox and choose Deactivate & Erase Data | CAB also requests deletion of the organization’s conversations, configurations, Campaign context, cached Agent data, job and usage records, and stored outcome data from GCP. |

