Name the requests a report is about with their x-request-id values, or a live session's connection id. The ids let the team trace each request in the account's logs. For a wrong output, add an example with the exact input, the output you received, and the output you expected.
A rating on at least one request id is a complete report and the quickest signal after checking an output. Ratings feed the team's quality measures and stay open. A report with a message or an example is triaged, and GET /v1/feedback returns its status and the team's note.
Feedback costs no credits and sits outside the rate limit. Each account can send 50 written reports and 1,000 rating-only reports in any 24 hours. Past either cap the report answers 429 feedback_limit and nothing is stored. Reports take no Idempotency-Key. Before sending one again, check GET /v1/feedback with request_id.
Headers#
authorizationBearer pxa_...RequiredAuthorization: Bearer pxa_... (recommended).
x-api-keypxa_...Alternative to the Authorization header. Ignored when Authorization is present.
Body#
kind"model" or "api" or "platform" or "other"RequiredWhat the report is about. Use model when an output was wrong or poor, api when the API misbehaved, platform for the dashboard, billing, or docs, and other for anything else.
rating"good" or "bad"Your judgment of the output the request ids name. A rating alone is a complete report when request_ids names at least one request.
messagestringWhat happened and what you expected, in plain text or Markdown.
request_idsarray of stringsUp to 20 x-request-id values the report is about, or the connection id of a live session. They link the report to the requests in your logs.
exampleobjectA worked case as text, with the exact input, the output you received, and the output you expected.
Show propertiesHide properties
example.inputstringThe input you sent.
example.outputstringThe output you received.
example.expectedstringThe output you expected instead.
Response#
One feedback report.
idstringRequiredReport id. Pass it as before to list older reports.
kind"model" or "api" or "platform" or "other"RequiredWhat the report is about.
rating"good" or "bad" or nullRequiredThe rating sent, or null.
messagestring or nullRequiredThe message as stored, or null.
request_idsarray of stringsRequiredThe request ids sent, lowercased, deduplicated, and sorted.
exampleobjectRequiredThe example as stored; every field is null without one.
Show propertiesHide properties
example.inputstring or nullRequiredThe example input, or null.
example.outputstring or nullRequiredThe example output, or null.
example.expectedstring or nullRequiredThe expected output, or null.
source"api" or "app"Requiredapi: sent to this endpoint with a key. app: sent from the dashboard.
status"open" or "resolved"RequiredA report is open until the team has acted on it, and resolved from then on. A report that carries only a rating feeds quality measures and stays open.
resolutionstring or nullRequiredThe team's note on the outcome; null when none was written.
created_atstringRequiredCreation time, ISO 8601.
resolved_atstring or nullRequiredResolution time, ISO 8601; null while open.