SoftTech AI Registry
HubSpot · Breeze

Data Agent Workflow Actions

HubSpot workflow actions (Data Agent: Custom prompt, Data Agent: Research, Data Agent: Fill Smart Property) that apply Breeze AI to enrolled record data inside HubSpot workflows.

Recorded characteristics

Function
HubSpot documents three AI workflow actions under the AI section of the workflow editor, each marked BETA in HubSpot's workflow action reference. "Data Agent: Custom prompt" runs a user-authored prompt (limited to 3,000 characters) against data explicitly added to the prompt, and can define structured action outputs including enumerated values. "Data Agent: Research" answers a prompt using either property data on the enrolled record or the five most recent associated call transcripts, returning a String, Number or Boolean. "Data Agent: Fill Smart Property" runs an existing smart property for a selected object and fill option, described by HubSpot as using AI to populate that smart property. Outputs of the custom prompt and research actions can be consumed by separately configured workflow actions, such as Branch or the CRM "Edit record" action.
Data access
Documentation establishes that the custom prompt action only receives the data explicitly added to the prompt, including properties of the enrolled record and associated records or activities added as data variables; HubSpot states it does not otherwise have access to the enrolled record's other properties, timeline activity or email contents. The research action draws on either an existing property on the enrolled record or the five most recent associated call transcripts. The Fill Smart Property action operates on a selected object and an existing smart property. HubSpot states the CRM data setting must be turned on for these actions to analyse record data.
Actions
Can take actions
External actions
Unknown
Human confirmation
Not established
Permission basis
Not established
Administrative control
HubSpot documents that access to generative AI features is managed in AI settings, and that both the "Give users access to generative AI tools and features" and CRM data settings must be turned on before the Data Agent: Custom prompt or Data Agent: Research actions are used to analyse record data. Super Admin or Workflows permissions are required to create workflows. HubSpot points to its AI settings, AI Trust FAQs and AI Model Cards for AI security controls and data-use configuration.
Default state
Not established
Availability
Documented as available with Marketing Hub, Sales Hub, Service Hub, Data Hub, Smart CRM and Revenue Hub Professional and Enterprise subscriptions, except where noted. Generating personalised content for automated marketing emails through this action is documented as requiring a Marketing Hub Enterprise subscription. The three Data Agent workflow actions are each listed as BETA in HubSpot's workflow action reference.
Licensing
HubSpot states that HubSpot Credits are required. Documentation notes that if an account does not have sufficient credits, the Data Agent: Custom prompt action will fail and populate null values as the output.
External model or provider
Not established
Limitations and uncertainty
The reviewed documentation does not identify the underlying model or provider; it states only that the LLM used in the Data Agent: Custom prompt is not connected to the internet. An important distinction is preserved here: the Data Agent actions generate or extract output, and any subsequent change to CRM records through branching or the "Edit record" action is a separately configured workflow action rather than an action independently decided and executed by the AI. The Fill Smart Property action is the case where HubSpot describes the AI action itself populating a smart property. The documentation does not establish whether any runtime human confirmation applies to individual executions, nor the identity or permission context under which the actions execute at runtime (permissions are documented for workflow creation, not for action execution). Default enablement state, external-system action capability, and the model/provider remain not established.

Evidence