Get Ticket AI Insights
GET/api/support/agent/tickets/:ticket/ai
Returns the latest AI state of a ticket in one object: the triage (classification, suggested type and priority, product area, summary, questions to ask, whether the text proves a bug), the documentation sections that may answer it, and the drafted reply. A drafted reply is only ever a suggestion — it reaches the customer only when an agent sends it. Compare Rerun AI Triage, which starts a new pass; this endpoint only reads.
support:readGrant this scope to your token under Settings → Developer → Personal Access Tokens.
The triage shown is the newest triage run of any status; the draft is the newest completed draft.
Path parameters
ticket(integer) — the ticket id.
Response fields (data)
run_id(integer, nullable) — the draft run id, else the triage run id; use it with Record AI Feedbacktriage_run_id,draft_run_id(integer, nullable)status(string) — one of:ready— a draft is available (draft.bodyis the literalNO_REPLYwhen the triage judged the ticket spam or noise)triaged— the triage finished but no draft could be producedpending— the triage is still runningquarantined— the triage was skipped because the ticket text matched the prompt-injection screen; a human should read itfailed— the triage call failed (seeerror)none— no AI pass has run for this ticket
model(string, nullable) — the model that produced the draft, else the triageclassification(object, nullable — null unless the triage finished):classification(string) — one of: NEEDS_INFO, BUG_SUSPECTED, HOW_TO, FEATURE_REQUEST, BILLING, SPAM_OR_NOISEtype(string, nullable) — suggested ticket type: bug, question, feature_request, billingpriority(string, nullable) — suggested priority: low, medium, high, urgentarea(string, max 80) — a short product-area labelsummary(string, max 500) — one sentence for the agentquestions[](array of strings, up to 3) — what to ask the customer when information is missingbug_confirmed(boolean) — true only when the ticket text itself shows a reproducible defect
suggestions(object, may be empty) — one entry per suggested ticket property, keyed by property (type,priority,group,assignee,tenant). Each carriesvalue(the suggested value),label(a readable version of it),reason(one line, nullable),confidence(0–1, nullable),property,mode(showorauto_apply— a property set tohidenever appears) anddecision(nulluntil an agent records one, then{accepted, by_user_id, at, reason}— see Record AI Feedback). A property whose suggestion matches what the ticket already says is not offered, and a run made before a property was suggested carries nothing for itsettings(object) — the desk-wide mode for each property, see Get AI Suggestion Settingsdocs[]— up to 3 documentation matches:title,url,snippet(max 240 chars),score(0–1, best first)draft(object, nullable) —id,body(string),accepted(boolean, nullable — what the agent recorded, see Record AI Feedback),created_aterror(string, nullable) — why the triage was skipped or failedcreated_at(ISO-8601, nullable) — when the draft (else the triage) was created
Authentication: Bearer token with the support:read scope, or an authenticated session, from a caller with an active support role (any role, including viewer). No tenant context — this endpoint is central.
Request
Responses
- 200
- 401
- 403
- 404
- 429
OK
Response Headers
Unauthorized
Response Headers
Forbidden
Response Headers
Not Found
Response Headers
Rate limited — platform limit is 1,000 requests/min; individual tokens may carry lower limits. Honor the Retry-After header before retrying. See the Rate Limits guide.