# Zendesk

Attify reads Zendesk through its API, read-only, with OAuth or an API token, for customer support. It is the customer's own system, so what Attify reads there is connected data: the strongest provenance, and what a vendor verification rests on.

## As a help desk, for customer support

| Stored            | Zendesk                                                                     |
|-------------------|-----------------------------------------------------------------------------|
| id                | ticket `id`                                                                 |
| started_at        | ticket `created_at`                                                         |
| updated_at        | ticket `updated_at`                                                         |
| customer          | ticket `requester_id`                                                       |
| channel           | ticket `via.channel`, a via id read as its via type (48 is web_widget)      |
| claim             | the billed resolution count (Standard #1 v0.10): the ticket's standard      |
|                   | field "Resolution tier", found by its title in the account's ticket fields. |
|                   | Since 18 May 2026 Zendesk bills only the tier "Verified resolution", its    |
|                   | own LLM check: that tier is a claim; "Contained resolution" and "Assisted   |
|                   | escalation" are not billed, so they claim none. A ticket without a tier     |
|                   | (closed before the tiers, or an account without the field) falls back to    |
|                   | the tags `automated_resolution`, `bot_solved` or `ai_resolved`; without one |
|                   | there is no vendor claim (the reconstructed claim applies). Status solved   |
|                   | alone is not an AI claim. Zendesk documents the field, not its API key, so  |
|                   | the field is read from the ticket's `custom_fields` by the id the ticket    |
|                   | fields list gives it (or a top-level `resolution_tier`): untested live      |
| messages          | public comments, oldest first. Internal notes are left out, and so are      |
|                   | comments made by a trigger or automation (`via.channel` rule) or by a merge |
|                   | (merge): they are nobody's reply                                            |
| messages[].role   | the requester and other end users: the customer; anyone else: support       |
| messages[].author | "ai" when a support comment came through an AI agent channel (`via.channel` |
|                   | answer_bot, answer_bot_api, answer_bot_for_web_widget, answer_bot_for_sdk,  |
|                   | answer_bot_for_slack, as a via type or via id) or the ticket's              |
|                   | `support_type` is ai_agent; otherwise "human"                               |
| reopens           | audit events changing `status` from solved to new, open, pending or hold,   |
|                   | at the audit's `created_at`                                                 |
| merged_into       | on a ticket tagged closed_by_merge (a help article spells it                |
|                   | closed_by_merged), the ticket its closing comment names ("merged into       |
|                   | request #<id>", matched in English only; otherwise the surviving ticket's   |
|                   | merged_from carries the merge)                                              |
| merged_from       | on any other ticket, the tickets a merge's via names (`via.channel` merge   |
|                   | or `source.rel` merge, `source.from.ticket_id`) on its comments and audits  |
