How conversations are handled

01

Position

Napkin is an AI teller assistant for community banks and credit unions. It works during live conversations with members. This page is the operational description of data handling, not a marketing summary.

02

Processing on Napkin’s own machines

Inference and processing run on infrastructure Napkin operates. The institution’s conversation data is not sent to a rented third-party model API for this work.

03

No third-party cloud AI APIs

Napkin does not call third-party cloud AI APIs to process these conversations. The contrast is the processing claim: work stays on machines Napkin operates.

04

Voice-only capture

The assistant hears the conversation. That is the capture surface. There is no other capture story on this product.

05

Retention and deletion controls

Retention is set by the institution. Deletion follows that policy and can be requested. A default retention period is not published until counsel confirms it.

06

Access roles

Access is role-based. A teller sees prompts and task completions for their own conversations. A supervisor sees staff outputs for their branch. A compliance officer sees required-language outcomes and audit events for the institution. An admin sets retention, roles, and deletion. These are role descriptions, not a product name.

07

Audit trail

The institution can see who accessed outputs, when, and what was changed. This is an institutional control over the record of the work. It is not a tool for watching staff.

08

Member disclosure guidance

Institutions decide how to tell members that an AI teller assistant is present. The sample below is sample language, not legal advice.

Sample — not legal advice

Our branch uses Napkin, an AI teller assistant, during this conversation. It helps our staff complete follow-up work, suggest fitting products, and use required language while we speak with you. Processing runs on Napkin’s own machines.

09

Data is never used to train outside models

Institution data is not sent to train third-party models. Napkin does not use these conversations to train models outside Napkin.

Downloadable security overview

Download Napkin-Security-Overview.pdf. This is an interim public overview. It restates this page and does not name certifications. The longer packet for an internal review is requested separately. A counsel draft of that packet is pending.

Request the packet

If you need the longer packet for an internal review, request it here. Get started remains available in the navigation.

No certification is named on this page. A certification will be listed only after it is approved.