When it fires
Only when the answer does not stream: after the provider answered and the answer was saved, right before kimaai_chatbot_after_response and before the JSON goes back with status 200. A streamed answer has no JSON body to filter, so the filter does not run then; use kimaai_chatbot_final_message for the answer text on both paths. It does not run for an error either, nor for a quick action whose callback answers without the provider.
The chatbot, the admin assistant and the search answer share this request runner, so the filter fires for all three; $payload['assistantSurface'] is admin_assistant or search outside the chat. KimaAI’s widget reads message, sessionId, history and renderables: add keys of your own for a client of your own, and leave those as they are unless you mean to change what the widget shows. The search answer box builds its answer, and the cached copy, from the filtered response.
Parameters
| Name | Type | Description |
|---|---|---|
response | array | The JSON answer, with message, session_id, sessionId, history and, when there are any, renderables and sources. |
ai_payload | array | The request to the provider, after kimaai_chatbot_ai_payload. |
payload | array | The normalised request payload, with prompt, sessionId and chatbot_settings among others, and assistantSurface (admin_assistant or search) outside the chat. |
chatbot_settings | array | The chatbot settings, or the admin assistant’s in its requests. |
Returns
array: The JSON answer to send.
Example
add_filter(
'kimaai_chatbot_response_payload',
function (array $response, array $ai_payload): array {
// A headless front end shows which model answered.
$response['model'] = (string) ($ai_payload['aiModel'] ?? '');
return $response;
},
10,
2
);
Related
kimaai_chatbot_after_response: fires after the answer, streamed or not.kimaai_chatbot_final_message: changes the answer text on both paths.