Skip to legal content
pullup Legal centre Back to Pullup
All documents/AI & Automation

USEFUL ASSISTANCE. SPECIFIC PERMISSIONS.

AI & Automation

A purpose-by-purpose account of AI, permitted context, providers, confirmation gates, moderation and human review.

Revision 2026-11-165 min read8 sections
Reviewed
Approved month
February 2026
Effective
2026-11-16
Status
effective
pullupat

4th Floor 20 Fenchurch Street London EC3M 3BY United Kingdom

Country of establishment: United Kingdom

support@pullup.at

AT A GLANCE

  • Personal Pilot and optional Mail intelligence have separate choices.
  • An assistant proposal does not save itself.
  • Generated answers and sentiment need verification against original evidence.

A reading guide. Read the full clauses for their conditions and scope.

On this page

Use your browser’s Find control to search the full text on mobile.

1. Different purposes, different flows

There is no single permission for every model use.

AI functions can include general help, personal dashboard assistance, Mail meaning/answer features, outbound safety moderation, event/search assistance, translation and poster/image generation. Their data, recipients, permissions and effects differ.

The Privacy Policy explains applicable grounds and rights. Feature permission is not blanket consent to all purposes or authority to disclose other people's information. Availability varies by release/configuration.

2. General help and personal Pilot

Personal context requires the separate agreement and scopes.

General mode sends the chosen conversation and broad context needed for an answer without enabling personal account tools. Personal Pilot requires its specific policy acceptance and selected data scopes: host events, host insights, tickets and/or account, as enabled.

Personal tools enforce account/role permissions and are designed to exclude guest contact lists, medical notes, QR credentials, passwords and card/bank details. Free text can still contain personal information if you enter it. Review a prompt before sending.

The existing Personal Assistant Terms and versioned consent remain in force for that feature. This Legal centre does not silently broaden the saved scope or replace an already recorded consent version.

3. Proposed changes and confirmation

Inspect the proposed value and the saved result.

Supported personal actions can prepare changes such as event details/schedule or account display name. The required Confirm and save action is separate from generation; access is checked again for the actual save.

Pilot does not gain authority to refund, pay out, purchase, send guest email, delete an event or change permissions merely because you request it. A successful conversational answer does not prove a database change succeeded.

Confirm the record, date/time, visibility and value before saving. A saved event edit is not necessarily an attendee notification. Reject a proposal that lacks required authority or would expose a private address.

4. Model recipients and retention

A request not to store is not universal zero retention.

Dashboard Pilot can use OpenAI directly or an enabled Vercel AI Gateway route. The implementation requests store:false where supported. That is a provider response-storage setting, not a guarantee of no abuse/security logs or an organisation-wide zero-retention agreement.

Optional Mail intelligence uses Groq for permitted matching-message excerpts. Outbound Mail moderation uses an OpenAI model through Vercel AI Gateway for a separate operational purpose. Disabling the optional Mail feature does not describe the moderation path.

Provider retention, subprocessors, processing location and any approved special data controls depend on the relevant provider agreement/account configuration. We do not promise that zero-data retention is enabled or that every provider has identical training/retention rules. Necessary request/usage and consent/audit metadata can remain.

5. Mail search, answers and organisation

The selected evidence limits the answer.

Meaning search and knowledge answers operate on the authorised mailbox and the implemented search/excerpt limits, rather than granting a model unrestricted access to every user's history. The reviewed discovery limit is up to 120 matching messages and up to 1,800 characters per excerpt.

Summaries may omit qualifications, miss a later thread or confuse a proposal with acceptance. A citation helps locate evidence but does not certify a binding agreement or its current status.

Labels, topic suggestions, rules and sentiment are aids. They are not established facts about a person's character, protected traits, health or intent, and should not determine an important employment, financial or admission decision.

6. Translation, generation and external tools

Third-party tools have their own account and terms.

Browser translation can send selected text to a Google translation endpoint; locale/IP detection and browser caching are separate functions. Do not submit confidential content solely because translation appears inside the Pullup interface.

Event discovery/generation can use configured model services such as Groq with the entered criteria. Structured planning may also use ordinary deterministic availability logic; an interface label alone is not proof that every operation uses a model.

Poster/image creation can use a browser-side Puter integration, a separate account, provider terms and charges. Inspect generated art for rights, accuracy and safety. We do not promise exclusive ownership, copyright protection or confidentiality beyond the applicable controls.

7. Operational automation and consequential outcomes

Ask for review of a decision that appears wrong.

Spam, fraud, payment-risk, account-safety, eligibility and rate-limit rules can use available signals, reports and outcomes. A moderation flag is not a finding of guilt; a successful verification is not a safety guarantee.

The applicable law determines safeguards for solely automated decisions with significant effects, including any explanation, representation, objection, contest and human review rights. UK and EU rules are not assumed identical. Contact support about an important adverse outcome or incorrect source record.

Do not rely on outputs as professional legal, medical, financial, emergency or safety advice. Use the original record and qualified judgement where required; Chat/support/AI are not emergency services.

8. Choice, records and rights

Clearing a conversation and revoking a permission have different effects.

Use the relevant personal-AI or mailbox choice to stop future optional use. The interface can clear session conversation on mode/opt-out changes; versioned consent, selected scopes, timestamps and approved-change audits may remain for accountability.

Do not paste identity documents, secrets, medical information or another person's confidential communications without the needed authority and lawful basis. Role access alone does not establish that an external model disclosure is appropriate.

For access, correction, objection, erasure, explanation or human review where available, contact support@pullup.at with the feature and approximate time. We may need proportionate verification and must also protect other people's rights.