Data handling

Which AI sees what, per agent

7 of 10 agents run without an AI model. The others are listed field by field, by HubSpot property name.

• Full Portal Audit Agent

  AI sees: record fields

  Audit-section counts and a sample of deal fields; with an Anthropic key the model reads contact, company, deal and ticket records itself.

  Default model under managed AI: gemini-2.5-pro

  Managed AI (Google Gemini, the default), or your own Google or OpenAI key

  One call per audit category with that category's counts and percentages — including how many sampled contacts have each lifecycle stage and each original source, and how many sampled tickets have each priority (the contact and ticket fields below, as counts per value, never per record) — and up to three record IDs per object; for deals, also each sampled deal's stage, open/won/lost state, amount, last-modified date and number of associated contacts. A final call sees only the category scores and findings.

  Contacts

  lifecyclestagehs_analytics_sourcehs_latest_source

  Deals

  dealstageamounths_lastmodifieddatenum_associated_contacts

  Tickets

  hs_ticket_priority

  Your own Anthropic key, or an account with no AI setting (runs on our Anthropic key)

  The model reads your records itself, through HubSpot read tools, during the run. It is told to read the fields below, plus property definitions, pipelines, workflows, lists, owners (users' names and email addresses), forms, marketing-email statistics, blog posts, landing pages, and sent, bounced and unsubscribed email events (each with the recipient's address). Its tools are not limited to those reads, so on this engine any field a HubSpot read tool returns can reach the model. Up to three entries from the AI memory you saved in Settings may be added.

  Contacts

  emailfirstnamelastnamephonecompanyjobtitleindustrylifecyclestagehs_lead_statushs_email_bouncehs_email_optouths_email_is_ineligiblecitycountrycreatedatehs_lastmodifieddatehubspot_owner_idassociatedcompanyidnum_associated_dealsnotes_last_contactedhs_analytics_sourcehs_latest_source

  Companies

  namedomainindustrynumberofemployeescitycountryphoneannualrevenuelifecyclestagecreatedatehs_lastmodifieddatehubspot_owner_idnum_associated_contactsnum_associated_deals

  Deals

  dealnameamountdealstagepipelineclosedatecreatedatehs_lastmodifieddatehubspot_owner_idhs_deal_stage_probabilitynum_associated_contactshs_closed_won_datehs_closed_lost_reasondealtypehs_is_closedhs_is_closed_won

  Tickets

  subjecths_pipelinehs_pipeline_stagehs_ticket_priorityhubspot_owner_idcreatedatehs_lastmodifieddatehs_resolution_timesource_typehs_is_closedclosed_datehs_time_to_close_sla_statushs_time_to_first_response_sla_status

• Property Cleanup Agent

  AI sees: nothing

  Property usage, descriptions and archive candidates are worked out by rules in our code; no AI model is called.

• Deduplication Agent

  AI sees: record fields

  Only ambiguous (fuzzy) matches: the matched records' display fields, for an advisory second opinion. Exact matches never reach a model.

  Default model under managed AI: gemini-2.5-flash

  Only when a run finds fuzzy matches (confidence below 95%), at most 40 duplicate groups per run

  Each matched record's ID and display fields (a value longer than 200 characters is cut, except names, email addresses, phone numbers, domains, deal names and ticket subjects), the property the match was made on, and your optional instructions. For a custom object, up to eight of that object's own text properties. The answer is advisory: it never changes which merges are proposed, and every merge still waits for your approval.

  Contacts

  emailfirstnamelastnamephonecompanyjobtitlelifecyclestagecreatedatehs_lastmodifieddatehubspot_owner_idhs_additional_emails

  Companies

  namedomainindustrycitycountrycreatedatehs_lastmodifieddatehubspot_owner_id

  Deals

  dealnameamountdealstagepipelineclosedatecreatedatehs_lastmodifieddatehubspot_owner_id

  Tickets

  subjecths_pipelinehs_pipeline_stagehs_ticket_prioritycreatedatehs_lastmodifieddate

• Data Enrichment Agent

  AI sees: nothing

  Fills empty fields from Apollo.io with your own Apollo key; no AI model is called.

  Also sent to Apollo.io (your own API key): Each contact's email address and each company's domain, to look up the values for empty fields.

• Contact Context Agent

  AI sees: record fields

  Each contact's profile, recent notes, calls, meetings and emails, and open deals and tickets — to write the contact summary.

  Default model under managed AI: gemini-2.5-flash

  Contacts in batches of five, each with its record ID and: profile fields; activity recency (dates are sent as days ago); its company's HubSpot Intent figures where the portal has Intent; up to three open deals and three open tickets; the five newest notes (first 600 characters each); the five newest calls, meetings and logged emails (title and body, first 200 characters); and up to five earlier RevvedAgents findings about the contact. The model has no HubSpot tools. Up to three entries from the AI memory you saved in Settings may be added.

  Contacts

  firstnamelastnameemailcompanyjobtitlelifecyclestagehs_lead_statusnotes_last_updatedhs_email_last_open_datehs_email_opennum_contacted_notescreatedatehs_analytics_last_visit_timestamphs_analytics_num_visitshs_analytics_num_page_viewshs_latest_sourcehs_sales_email_last_repliedhs_last_sales_activity_timestamp

  Companies

  hs_intent_visitors_last_30_dayshs_intent_page_views_last_30_dayshs_most_recent_de_anonymized_visiths_recent_intent_signals

  Deals

  dealnameamountdealstagehs_deal_stage_probabilityhs_lastmodifieddate

  Tickets

  subjecths_ticket_prioritycreatedate

  Notes

  hs_note_bodyhs_timestamp

  Calls

  hs_call_titlehs_call_bodyhs_call_directionhs_timestamp

  Meetings

  hs_meeting_titlehs_meeting_bodyhs_internal_meeting_noteshs_meeting_outcomehs_timestamp

  Logged emails

  hs_email_subjecths_email_texths_email_directionhs_email_statushs_timestamp

• Send Suspension Prevention Agent

  AI sees: nothing

  Bounce, complaint and engagement figures are computed by rules in our code; no AI model is called.

• Email Sunset Policy

  AI sees: nothing

  Unengaged cohorts and exclusions are decided by rules in our code; no AI model is called.

• Deal Risk Autopilot

  AI sees: nothing

  Risk signals are computed by rules in our code; no AI model is called.

• Renewal Playbook

  AI sees: nothing

  Renewal dates and tasks are worked out by rules in our code; no AI model is called.

• Content Integrity

  AI sees: nothing

  Blog checks are rules in our code; no AI model is called.

Which AI provider runs an agent

It depends on the AI setting your account uses (Settings → AI Configuration).

Other features that use AI

Sub-processors

The services that process your data on our behalf, and when.

How long data is kept

Windows come from our code. “Not set” means the code deletes nothing automatically — we say so rather than guess.