← All changes | admin/class-knowledge-manager.php +2288 -686 3.2.2 → trunk View file →

		@@ -9,11 +9,27 @@
9	9	exit; // Exit if accessed directly
10	10	}
11	11
12	12	class MxChat_Knowledge_Manager {
13		-
	13	+
14	14	private $options;
15		-
	15	+
	16	+ // Post IDs whose vectors were already deleted by mxchat_handle_status_transition this
	17	+ // request, so the transient-based branch in mxchat_handle_post_update can skip the
	18	+ // redundant (idempotent but network-visible) second deletion.
	19	+ private $transition_deleted_posts = array();
	20	+
	21	+ // Post IDs already INDEXED by mxchat_handle_status_transition's arrival edge this
	22	+ // request. Normal editor publishes fire transition_post_status first, then
	23	+ // post_updated — without this guard every editor publish would embed twice.
	24	+ private $transition_indexed_posts = array();
	25	+
	26	+ // Post IDs core has announced an in-flight UPDATE for. pre_post_update fires only
	27	+ // inside wp_insert_post's update branch and always before wp_transition_post_status,
	28	+ // so this is an exact "a post_updated is coming later this request" signal — which is
	29	+ // what makes it safe to arm transition_indexed_posts (plan a664f3).
	30	+ private $pending_post_update = array();
	31	+
16	32	/**
17	33	* Constructor - Register hooks for content processing
18	34	*/
19	35	public function __construct() {
		@@ -31,8 +47,10 @@
31	47	// Admin post handlers for form submissions
32	48	add_action('admin_post_mxchat_submit_content', array($this, 'mxchat_handle_content_submission'));
33	49	add_action('admin_post_mxchat_submit_sitemap', array($this, 'mxchat_handle_sitemap_submission'));
34	50	add_action('admin_post_mxchat_submit_pdf_file', array($this, 'mxchat_handle_pdf_file_submission'));
	51	+ add_action('admin_post_mxchat_submit_document_file', array($this, 'mxchat_handle_document_file_submission'));
	52	+ add_action('admin_post_mxchat_submit_youtube', array($this, 'mxchat_handle_youtube_submission'));
35	53	add_action('admin_post_mxchat_stop_processing', array($this, 'mxchat_stop_processing'));
36	54
37	55	// AJAX handlers for real-time processing and status updates
38	56	add_action('wp_ajax_mxchat_get_status_updates', array($this, 'mxchat_ajax_get_status_updates'));
		@@ -58,8 +76,9 @@
58	76	add_action('wp_ajax_mxchat_refresh_pinecone_entries', array($this, 'ajax_mxchat_refresh_pinecone_entries'));
59	77	add_action('wp_ajax_mxchat_paginate_entries', array($this, 'ajax_mxchat_paginate_entries'));
60	78	add_action('wp_ajax_mxchat_get_entry_content', array($this, 'ajax_mxchat_get_entry_content'));
61	79	add_action('wp_ajax_mxchat_save_entry_content', array($this, 'ajax_mxchat_save_entry_content'));
	80	+ add_action('wp_ajax_mxchat_inspect_entry', array($this, 'ajax_mxchat_inspect_entry'));
62	81
63	82	// WordPress post management hooks
64	83	add_action('pre_post_update', array($this, 'mxchat_store_pre_update_status'), 10, 2);
65	84	add_action('post_updated', array($this, 'mxchat_handle_post_update'), 10, 3);
		@@ -64,13 +83,27 @@
64	83	add_action('pre_post_update', array($this, 'mxchat_store_pre_update_status'), 10, 2);
65	84	add_action('post_updated', array($this, 'mxchat_handle_post_update'), 10, 3);
66	85	add_action('before_delete_post', array($this, 'mxchat_handle_post_delete'));
67	86	add_action('wp_trash_post', array($this, 'mxchat_handle_post_delete'));
	87	+ // Authoritative unpublish detection: core hands this hook the REAL previous status, so
	88	+ // removal no longer depends on the mxchat_prev_status_* transients (evictable by persistent
	89	+ // object caches, never written by paths that bypass wp_update_post, e.g. plugins flipping
	90	+ // post_status directly and calling wp_transition_post_status themselves).
	91	+ add_action('transition_post_status', array($this, 'mxchat_handle_status_transition'), 10, 3);
68	92
69	93	// ACF hook - fires AFTER ACF fields are saved, ensuring ACF data is available
70	94	// Priority 20 to run after ACF's own save (which runs at priority 10)
71	95	add_action('acf/save_post', array($this, 'mxchat_handle_acf_save'), 20);
72	96
	97	+ // One-time cleanup for vectors orphaned by unpublishes that predate the
	98	+ // transition_post_status handler (plan 816fb1): wp mxchat prune-unpublished
	99	+ if (defined('WP_CLI') && WP_CLI) {
	100	+ WP_CLI::add_command('mxchat prune-unpublished', array($this, 'cli_prune_unpublished'));
	101	+ // In-place repair for RTL KB rows imported in visual order before the
	102	+ // 32bf9e normalizer existed: wp mxchat rtl-repair (plan d1e6f7)
	103	+ WP_CLI::add_command('mxchat rtl-repair', array($this, 'cli_rtl_repair'));
	104	+ }
	105	+
73	106	add_action('wp_ajax_mxchat_mark_queue_complete', array($this, 'ajax_mxchat_mark_queue_complete'));
74	107
75	108	// WooCommerce product hooks (if WooCommerce is active)
76	109	if (class_exists('WooCommerce')) {
		@@ -118,26 +151,17 @@
118	151
119	152	// Get bot-specific options and API key
120	153	$bot_options = $this->get_bot_options($bot_id);
121	154	$options = !empty($bot_options) ? $bot_options : get_option('mxchat_options');
122		- $selected_model = $options['embedding_model'] ?? 'text-embedding-ada-002';
123		-
124		- if (strpos($selected_model, 'voyage') === 0) {
125		- $api_key = $options['voyage_api_key'] ?? '';
126		- } elseif (strpos($selected_model, 'gemini-embedding') === 0) {
127		- $api_key = $options['gemini_api_key'] ?? '';
128		- } else {
129		- $api_key = $options['api_key'] ?? '';
130		- }
131		-
132		- if (empty($api_key)) {
133		- set_transient('mxchat_admin_notice_error',
134		- esc_html__('API key is not configured. Please add your API key in the settings before submitting content.', 'mxchat'),
135		- 30
136		- );
	155	+
	156	+ // Custom-provider-aware decision; keyless custom sites must pass (plan cbd5fd).
	157	+ $preflight = MxChat_Utils::embedding_preflight($options);
	158	+ if (!$preflight['ok']) {
	159	+ set_transient('mxchat_admin_notice_error', esc_html($preflight['reason']), 30);
137	160	wp_safe_redirect(esc_url(admin_url('admin.php?page=mxchat-prompts')));
138	161	exit;
139	162	}
	163	+ $api_key = $preflight['api_key'];
140	164
141	165	// Use centralized utility function with bot_id
142	166	$result = MxChat_Utils::submit_content_to_db($article_content, $article_url, $api_key, null, $bot_id);
143	167
		@@ -156,8 +180,238 @@
156	180	wp_safe_redirect(esc_url(admin_url('admin.php?page=mxchat-prompts')));
157	181	exit;
158	182	}
159	183
	184	+/**
	185	+ * Handle the "YouTube" KB import source (admin-post form submission).
	186	+ *
	187	+ * Per-video description mode:
	188	+ * - auto: fetch oEmbed metadata (reliable) + best-effort captions transcript.
	189	+ * If no usable transcript, index the metadata anyway, tell the admin,
	190	+ * and bounce back with the manual box pre-filled (never fail silently).
	191	+ * - manual: the admin's own description is what gets indexed; metadata rides along.
	192	+ *
	193	+ * The row is stored with content_type 'youtube' and source_url = the canonical
	194	+ * watch URL, so re-importing the same video UPDATES the entry (source_url
	195	+ * duplicate handling in MxChat_Utils::store_in_wordpress_db) — that is also the
	196	+ * "augment a metadata-only entry" path.
	197	+ */
	198	+public function mxchat_handle_youtube_submission() {
	199	+ if (!isset($_POST['submit_youtube']) || !current_user_can('manage_options')) {
	200	+ wp_die(esc_html__('Unauthorized access', 'mxchat'));
	201	+ }
	202	+
	203	+ check_admin_referer('mxchat_submit_youtube_action', 'mxchat_submit_youtube_nonce');
	204	+
	205	+ $redirect_url = admin_url('admin.php?page=mxchat-prompts');
	206	+
	207	+ $youtube_url = isset($_POST['youtube_url']) ? esc_url_raw(wp_unslash($_POST['youtube_url'])) : '';
	208	+ $video_id = MxChat_Utils::parse_youtube_id($youtube_url);
	209	+
	210	+ if (empty($video_id)) {
	211	+ set_transient('mxchat_admin_notice_error',
	212	+ esc_html__('That does not look like a link to a single YouTube video. Please paste a watch, youtu.be, or Shorts URL.', 'mxchat'),
	213	+ 30
	214	+ );
	215	+ wp_safe_redirect(esc_url($redirect_url));
	216	+ exit;
	217	+ }
	218	+
	219	+ $canonical_url = 'https://www.youtube.com/watch?v=' . $video_id;
	220	+
	221	+ $description_mode = (isset($_POST['youtube_description_mode']) && $_POST['youtube_description_mode'] === 'manual') ? 'manual' : 'auto';
	222	+ $manual_description = isset($_POST['youtube_description']) ? trim(wp_kses_post(wp_unslash($_POST['youtube_description']))) : '';
	223	+
	224	+ $bot_id = isset($_POST['bot_id']) ? sanitize_key($_POST['bot_id']) : 'default';
	225	+
	226	+ // Resolve the embedding decision exactly like the sibling handlers —
	227	+ // custom-provider-aware (plan cbd5fd).
	228	+ $bot_options = $this->get_bot_options($bot_id);
	229	+ $options = !empty($bot_options) ? $bot_options : get_option('mxchat_options');
	230	+
	231	+ $preflight = MxChat_Utils::embedding_preflight($options);
	232	+ if (!$preflight['ok']) {
	233	+ set_transient('mxchat_admin_notice_error', esc_html($preflight['reason']), 30);
	234	+ wp_safe_redirect(esc_url($redirect_url));
	235	+ exit;
	236	+ }
	237	+ $api_key = $preflight['api_key'];
	238	+
	239	+ // Metadata is fetched in BOTH modes — it is the reliable half of auto, and in
	240	+ // manual mode it enriches the indexed text with the real title/channel.
	241	+ $meta = $this->mxchat_fetch_youtube_oembed($video_id);
	242	+ $video_title = isset($meta['title']) ? sanitize_text_field($meta['title']) : '';
	243	+ $video_channel = isset($meta['author_name']) ? sanitize_text_field($meta['author_name']) : '';
	244	+
	245	+ $header_lines = 'YouTube Video: ' . ($video_title !== '' ? $video_title : $canonical_url) . "\n";
	246	+ if ($video_channel !== '') {
	247	+ $header_lines .= 'Channel: ' . $video_channel . "\n";
	248	+ }
	249	+ $header_lines .= 'URL: ' . $canonical_url . "\n\n";
	250	+
	251	+ $transcript_missing = false;
	252	+
	253	+ if ($description_mode === 'manual') {
	254	+ if ($manual_description === '') {
	255	+ set_transient('mxchat_admin_notice_error',
	256	+ esc_html__('Please write a description for the video, or switch to auto-fetch.', 'mxchat'),
	257	+ 30
	258	+ );
	259	+ wp_safe_redirect(esc_url($redirect_url));
	260	+ exit;
	261	+ }
	262	+ $indexed_text = $header_lines . $manual_description;
	263	+ } else {
	264	+ $transcript = $this->mxchat_fetch_youtube_transcript($video_id);
	265	+
	266	+ if (strlen($transcript) >= 200) {
	267	+ $indexed_text = $header_lines . $transcript;
	268	+ } else {
	269	+ // Graceful fallback: captions disabled / blocked / no speech. Auto
	270	+ // reliably gets metadata; it does NOT guarantee a transcript.
	271	+ $transcript_missing = true;
	272	+
	273	+ if ($video_title === '' && $video_channel === '') {
	274	+ // Both halves failed — nothing meaningful to index.
	275	+ set_transient('mxchat_admin_notice_error',
	276	+ esc_html__('Could not retrieve any information for that video (no metadata and no captions). Please check the URL, or use the manual description option.', 'mxchat'),
	277	+ 30
	278	+ );
	279	+ wp_safe_redirect(esc_url($redirect_url));
	280	+ exit;
	281	+ }
	282	+
	283	+ $indexed_text = $header_lines . sprintf(
	284	+ /* translators: 1: video title, 2: channel name */
	285	+ __('A YouTube video titled "%1$s" from the channel %2$s.', 'mxchat'),
	286	+ $video_title !== '' ? $video_title : $canonical_url,
	287	+ $video_channel !== '' ? $video_channel : 'YouTube'
	288	+ );
	289	+ }
	290	+ }
	291	+
	292	+ $result = MxChat_Utils::submit_content_to_db($indexed_text, $canonical_url, $api_key, null, $bot_id, 'youtube');
	293	+
	294	+ if (is_wp_error($result)) {
	295	+ set_transient('mxchat_admin_notice_error',
	296	+ esc_html__('Error storing video in the knowledge base: ', 'mxchat') . $result->get_error_message(),
	297	+ 30
	298	+ );
	299	+ wp_safe_redirect(esc_url($redirect_url));
	300	+ exit;
	301	+ }
	302	+
	303	+ if ($transcript_missing) {
	304	+ set_transient('mxchat_admin_notice_success',
	305	+ esc_html__('Video indexed from its title and channel — no captions were available for a transcript. The form below is pre-filled: write your own description and import again to improve matching (it updates the same entry).', 'mxchat'),
	306	+ 30
	307	+ );
	308	+ // Bounce back with prefill args so the page reopens the YouTube form in
	309	+ // manual mode with the URL + fetched title ready to augment.
	310	+ $redirect_url = add_query_arg(array(
	311	+ 'mxchat_yt_prefill' => '1',
	312	+ 'yt_url' => rawurlencode($canonical_url),
	313	+ 'yt_title' => rawurlencode($video_title),
	314	+ ), $redirect_url);
	315	+ } else {
	316	+ set_transient('mxchat_admin_notice_success',
	317	+ esc_html__('YouTube video successfully added to the knowledge base!', 'mxchat'),
	318	+ 30
	319	+ );
	320	+ }
	321	+
	322	+ wp_safe_redirect(esc_url_raw($redirect_url));
	323	+ exit;
	324	+}
	325	+
	326	+/**
	327	+ * Fetch YouTube oEmbed metadata for a video (no API key required).
	328	+ * Returns the decoded array (title, author_name, thumbnail_url, ...) or array().
	329	+ */
	330	+private function mxchat_fetch_youtube_oembed($video_id) {
	331	+ $oembed_url = 'https://www.youtube.com/oembed?url=' . rawurlencode('https://www.youtube.com/watch?v=' . $video_id) . '&format=json';
	332	+ $response = wp_remote_get($oembed_url, array('timeout' => 15));
	333	+ if (is_wp_error($response) || wp_remote_retrieve_response_code($response) !== 200) {
	334	+ return array();
	335	+ }
	336	+ $data = json_decode(wp_remote_retrieve_body($response), true);
	337	+ return is_array($data) ? $data : array();
	338	+}
	339	+
	340	+/**
	341	+ * Best-effort captions transcript for a video. Deliberately ISOLATED: this uses
	342	+ * YouTube's unofficial timedtext route (the caption track list embedded in the
	343	+ * watch page), which YouTube has broken before and will break again. Every
	344	+ * failure mode returns '' so a break degrades to the metadata-only import path
	345	+ * instead of erroring the whole submission. Do not let anything in here throw.
	346	+ */
	347	+private function mxchat_fetch_youtube_transcript($video_id) {
	348	+ $watch_url = 'https://www.youtube.com/watch?v=' . $video_id . '&hl=en';
	349	+
	350	+ // First try the honest ingest UA; some responses omit the player config for
	351	+ // bot UAs, so retry once with a browser UA before giving up.
	352	+ $user_agents = array(
	353	+ mxchat_ingest_user_agent(),
	354	+ 'Mozilla/5.0 (Windows NT 10.0; Win64; x64) AppleWebKit/537.36 (KHTML, like Gecko) Chrome/126.0.0.0 Safari/537.36',
	355	+ );
	356	+
	357	+ $tracks = array();
	358	+ foreach ($user_agents as $ua) {
	359	+ $response = wp_remote_get($watch_url, array(
	360	+ 'timeout' => 20,
	361	+ 'user-agent' => $ua,
	362	+ ));
	363	+ if (is_wp_error($response) || wp_remote_retrieve_response_code($response) !== 200) {
	364	+ continue;
	365	+ }
	366	+ $body = wp_remote_retrieve_body($response);
	367	+ if (!is_string($body) || $body === '' || !preg_match('/"captionTracks":(\[.*?\])(?=,")/s', $body, $m)) {
	368	+ continue;
	369	+ }
	370	+ $decoded = json_decode($m[1], true);
	371	+ if (is_array($decoded) && !empty($decoded)) {
	372	+ $tracks = $decoded;
	373	+ break;
	374	+ }
	375	+ }
	376	+
	377	+ if (empty($tracks)) {
	378	+ return '';
	379	+ }
	380	+
	381	+ // Prefer an English track, else take the first offered.
	382	+ $chosen = null;
	383	+ foreach ($tracks as $track) {
	384	+ if (isset($track['languageCode']) && strpos($track['languageCode'], 'en') === 0) {
	385	+ $chosen = $track;
	386	+ break;
	387	+ }
	388	+ }
	389	+ if ($chosen === null) {
	390	+ $chosen = $tracks[0];
	391	+ }
	392	+ if (empty($chosen['baseUrl']) || !is_string($chosen['baseUrl'])) {
	393	+ return '';
	394	+ }
	395	+
	396	+ $timedtext = wp_remote_get($chosen['baseUrl'], array('timeout' => 20));
	397	+ if (is_wp_error($timedtext) || wp_remote_retrieve_response_code($timedtext) !== 200) {
	398	+ return '';
	399	+ }
	400	+ $xml = wp_remote_retrieve_body($timedtext);
	401	+ if (!is_string($xml) || strpos($xml, '<text') === false) {
	402	+ return '';
	403	+ }
	404	+
	405	+ // <text start=".." dur="..">caption</text> — strip tags, decode the
	406	+ // double-encoded entities timedtext ships, collapse whitespace.
	407	+ $text = preg_replace('/<[^>]+>/', ' ', $xml);
	408	+ $text = html_entity_decode(html_entity_decode($text, ENT_QUOTES | ENT_HTML5, 'UTF-8'), ENT_QUOTES | ENT_HTML5, 'UTF-8');
	409	+ $text = trim(preg_replace('/\s+/u', ' ', $text));
	410	+
	411	+ return $text;
	412	+}
	413	+
160	414	public function mxchat_is_pdf_url($url, $response) {
161	415	$content_type = wp_remote_retrieve_header($response, 'content-type');
162	416	$file_extension = strtolower(pathinfo($url, PATHINFO_EXTENSION));
163	417
		@@ -403,8 +657,160 @@
403	657	exit;
404	658	}
405	659
406	660	/**
	661	+ * Handle direct document upload (.docx / .txt / .md) from the knowledge base
	662	+ * page (plan 0485e5). Unlike PDF Upload there is no per-page queue: the text
	663	+ * extracts in one pass and routes through submit_content_to_db, whose chunker
	664	+ * takes over for long content. The uploaded file is read from the PHP temp
	665	+ * file and never persisted — only its extracted text enters the KB.
	666	+ *
	667	+ * Source identity matches PDF Upload's scheme: upload://<filename>, stable
	668	+ * across re-uploads so a re-import REPLACES (delete_chunks_for_url + upsert
	669	+ * per identity) instead of duplicating.
	670	+ */
	671	+public function mxchat_handle_document_file_submission() {
	672	+ if (!isset($_POST['submit_document_file']) || !current_user_can('manage_options')) {
	673	+ wp_die(esc_html__('Unauthorized access', 'mxchat'));
	674	+ }
	675	+
	676	+ check_admin_referer('mxchat_submit_document_file_action', 'mxchat_submit_document_file_nonce');
	677	+
	678	+ $redirect_url = admin_url('admin.php?page=mxchat-prompts');
	679	+
	680	+ if (empty($_FILES['document_file']) || $_FILES['document_file']['error'] !== UPLOAD_ERR_OK) {
	681	+ $error_code = isset($_FILES['document_file']['error']) ? $_FILES['document_file']['error'] : UPLOAD_ERR_NO_FILE;
	682	+ $error_messages = array(
	683	+ UPLOAD_ERR_INI_SIZE => __('The uploaded file exceeds the server upload_max_filesize limit.', 'mxchat'),
	684	+ UPLOAD_ERR_FORM_SIZE => __('The uploaded file exceeds the form MAX_FILE_SIZE limit.', 'mxchat'),
	685	+ UPLOAD_ERR_PARTIAL => __('The file was only partially uploaded.', 'mxchat'),
	686	+ UPLOAD_ERR_NO_FILE => __('No file was uploaded. Please select a document.', 'mxchat'),
	687	+ UPLOAD_ERR_NO_TMP_DIR => __('Server missing temporary folder.', 'mxchat'),
	688	+ UPLOAD_ERR_CANT_WRITE => __('Server failed to write file to disk.', 'mxchat'),
	689	+ );
	690	+ $error_msg = isset($error_messages[$error_code]) ? $error_messages[$error_code] : __('Unknown upload error.', 'mxchat');
	691	+ set_transient('mxchat_admin_notice_error', $error_msg, 30);
	692	+ wp_safe_redirect(esc_url($redirect_url));
	693	+ exit;
	694	+ }
	695	+
	696	+ $file = $_FILES['document_file'];
	697	+ $ext = strtolower(pathinfo($file['name'], PATHINFO_EXTENSION));
	698	+
	699	+ $finfo = finfo_open(FILEINFO_MIME_TYPE);
	700	+ $mime_type = finfo_file($finfo, $file['tmp_name']);
	701	+ finfo_close($finfo);
	702	+
	703	+ // Per-extension MIME expectations. finfo commonly reports .docx as
	704	+ // application/zip (it IS a Zip container) and .md as plain text.
	705	+ $mime_ok = false;
	706	+ if ($ext === 'docx') {
	707	+ $mime_ok = in_array($mime_type, array(
	708	+ 'application/vnd.openxmlformats-officedocument.wordprocessingml.document',
	709	+ 'application/zip',
	710	+ ), true);
	711	+ } elseif ($ext === 'txt' || $ext === 'md') {
	712	+ $mime_ok = (strpos((string) $mime_type, 'text/') === 0);
	713	+ }
	714	+
	715	+ if (!$mime_ok) {
	716	+ set_transient('mxchat_admin_notice_error',
	717	+ esc_html__('Invalid or unreadable document. Accepted types: .docx, .txt, .md.', 'mxchat'),
	718	+ 30
	719	+ );
	720	+ wp_safe_redirect(esc_url($redirect_url));
	721	+ exit;
	722	+ }
	723	+
	724	+ $bot_id = isset($_POST['bot_id']) ? sanitize_key($_POST['bot_id']) : 'default';
	725	+ $original_filename = sanitize_file_name($file['name']);
	726	+
	727	+ // ---- Extract text (ONE extractor for .docx — the word handler's) ----
	728	+ if ($ext === 'docx') {
	729	+ $text = MXChat_Word_Handler::extract_docx_text($file['tmp_name']);
	730	+ if ($text === false) {
	731	+ set_transient('mxchat_admin_notice_error',
	732	+ esc_html__('The .docx file could not be read. It may be corrupt, empty, or not a real Word document.', 'mxchat'),
	733	+ 30
	734	+ );
	735	+ wp_safe_redirect(esc_url($redirect_url));
	736	+ exit;
	737	+ }
	738	+ } else {
	739	+ // .txt / .md read as-is. Markdown keeps its syntax on purpose —
	740	+ // headings are useful retrieval signal.
	741	+ $text = (string) file_get_contents($file['tmp_name']);
	742	+ $text = wp_check_invalid_utf8($text);
	743	+ $text = trim($text);
	744	+ }
	745	+
	746	+ if ($text === '') {
	747	+ set_transient('mxchat_admin_notice_error',
	748	+ esc_html__('The uploaded document contains no readable text.', 'mxchat'),
	749	+ 30
	750	+ );
	751	+ wp_safe_redirect(esc_url($redirect_url));
	752	+ exit;
	753	+ }
	754	+
	755	+ // Size cap — same pdf_max_pages setting the PDF/toolbar paths use, but
	756	+ // estimated by CHARACTERS (~2500/page): the .docx cleaner collapses all
	757	+ // newlines to spaces, so a paragraph count reads 1 for any Word file.
	758	+ // Processing is synchronous — an unbounded document risks a timeout.
	759	+ $options = get_option('mxchat_options', array());
	760	+ $max_pages = isset($options['pdf_max_pages']) ? intval($options['pdf_max_pages']) : 69;
	761	+ $estimated_pages = (int) ceil(strlen($text) / 2500);
	762	+ if ($estimated_pages > $max_pages) {
	763	+ set_transient('mxchat_admin_notice_error',
	764	+ sprintf(
	765	+ esc_html__('The document is too large (about %1$d pages; the limit is %2$d). Split it into smaller files, or raise the PDF max pages setting.', 'mxchat'),
	766	+ $estimated_pages,
	767	+ $max_pages
	768	+ ),
	769	+ 30
	770	+ );
	771	+ wp_safe_redirect(esc_url($redirect_url));
	772	+ exit;
	773	+ }
	774	+
	775	+ // Embedding API key — bot-aware, same shape as the direct-content handler.
	776	+ $api_key = '';
	777	+ if ($bot_id !== 'default' && class_exists('MxChat_Multi_Bot_Manager')) {
	778	+ $bot_options = apply_filters('mxchat_get_bot_options', array(), $bot_id);
	779	+ $api_key = $bot_options['api_key'] ?? '';
	780	+ }
	781	+ if (empty($api_key)) {
	782	+ $api_key = $options['api_key'] ?? '';
	783	+ }
	784	+
	785	+ // Stable identity — PDF Upload's scheme. A re-upload of the same filename
	786	+ // replaces: clear old chunks first (covers a doc shrinking below the chunk
	787	+ // threshold, where the single-vector path would not clean them), then
	788	+ // submit — the chunked path re-deletes harmlessly.
	789	+ $source_label = 'upload://' . $original_filename;
	790	+ MxChat_Utils::delete_chunks_for_url($source_label, $bot_id);
	791	+ $result = MxChat_Utils::submit_content_to_db($text, $source_label, $api_key, null, $bot_id, 'document');
	792	+
	793	+ if (is_wp_error($result)) {
	794	+ set_transient('mxchat_admin_notice_error',
	795	+ esc_html__('Failed to import the document: ', 'mxchat') . esc_html($result->get_error_message()),
	796	+ 30
	797	+ );
	798	+ } else {
	799	+ set_transient('mxchat_admin_notice_success',
	800	+ sprintf(
	801	+ esc_html__('Document "%s" imported into the knowledge base.', 'mxchat'),
	802	+ esc_html($original_filename)
	803	+ ),
	804	+ 30
	805	+ );
	806	+ }
	807	+
	808	+ wp_safe_redirect(esc_url($redirect_url));
	809	+ exit;
	810	+}
	811	+
	812	+/**
407	813	* Validate PDF and count pages with multiple parser attempts
408	814	*/
409	815	private function mxchat_validate_and_count_pdf_pages($pdf_path) {
410	816	// Method 1: Try with Smalot PDF Parser (your current method)
		@@ -627,8 +1033,25 @@
627	1033	/**
628	1034	* AJAX: Get full content for editing — reassembles chunks if needed.
629	1035	* Works for both WordPress DB and Pinecone entries.
630	1036	*/
	1037	+/**
	1038	+ * Sanitize a knowledge entry's source_url from an AJAX request WITHOUT destroying
	1039	+ * its identity. sanitize_text_field() strips percent-encoded octets (%20, %D7%A9…),
	1040	+ * so a percent-encoded URL — every non-ASCII permalink — would md5 to a DIFFERENT
	1041	+ * id than the one it was stored under: reads miss the entry and saves write an
	1042	+ * orphan copy while the original keeps its stale text. URLs get esc_url_raw
	1043	+ * (identity-preserving, matches what import stored); non-URL keys (mxchat://,
	1044	+ * _ungrouped_) keep the old sanitizer.
	1045	+ */
	1046	+private function sanitize_entry_source_url( $raw ) {
	1047	+ $raw = trim( (string) $raw );
	1048	+ if ( preg_match( '#^https?://#i', $raw ) ) {
	1049	+ return esc_url_raw( $raw );
	1050	+ }
	1051	+ return sanitize_text_field( $raw );
	1052	+}
	1053	+
631	1054	public function ajax_mxchat_get_entry_content() {
632	1055	check_ajax_referer('mxchat_edit_entry_nonce', 'nonce');
633	1056
634	1057	if ( ! current_user_can('manage_options') ) {
		@@ -634,16 +1057,19 @@
634	1057	if ( ! current_user_can('manage_options') ) {
635	1058	wp_send_json_error( array( 'message' => 'Permission denied.' ) );
636	1059	}
637	1060
638		- $source_url = isset($_POST['source_url']) ? sanitize_text_field( wp_unslash($_POST['source_url']) ) : '';
	1061	+ $source_url = $this->sanitize_entry_source_url( isset($_POST['source_url']) ? wp_unslash($_POST['source_url']) : '' );
639	1062	$entry_id = isset($_POST['entry_id']) ? absint($_POST['entry_id']) : 0;
640	1063	$data_source = isset($_POST['data_source']) ? sanitize_key($_POST['data_source']) : 'wordpress';
641	1064	$bot_id = isset($_POST['bot_id']) ? sanitize_key($_POST['bot_id']) : 'default';
642	1065
643	1066	if ( $data_source === 'pinecone' ) {
	1067	+ // Pinecone ids are strings (md5 hashes, manual_* ids) — absint() would
	1068	+ // destroy them, so re-read the raw value for this branch only.
	1069	+ $vector_id = isset($_POST['entry_id']) ? sanitize_text_field( wp_unslash($_POST['entry_id']) ) : '';
644	1070	// Pinecone: fetch vectors by source_url, reassemble chunks
645		- $content = $this->get_pinecone_entry_content( $source_url, $entry_id, $bot_id );
	1071	+ $content = $this->get_pinecone_entry_content( $source_url, $vector_id, $bot_id );
646	1072	} else {
647	1073	// WordPress DB
648	1074	$content = $this->get_wordpress_entry_content( $source_url, $entry_id );
649	1075	}
		@@ -742,44 +1168,56 @@
742	1168	if ( empty($host) || empty($api_key) ) {
743	1169	return new WP_Error( 'pinecone_config', 'Pinecone not configured.' );
744	1170	}
745	1171
746		- // List vectors with the source_url prefix
747		- $base_id = md5( $source_url );
748		- $vector_ids = array( $base_id );
	1172	+ // Manual entries carry no source_url (their vector id is a minted manual_* string,
	1173	+ // not md5 of anything the row can hand us) — fetch the exact vector instead.
	1174	+ // '_ungrouped_' is the table view's synthetic display key for such rows.
	1175	+ if ( ( empty($source_url) || strpos($source_url, '_ungrouped_') === 0 ) && ! empty($entry_id) && is_string($entry_id) ) {
	1176	+ $vector_ids = array( $entry_id );
	1177	+ } else {
	1178	+ // List vectors with the source_url prefix
	1179	+ $base_id = md5( $source_url );
	1180	+ $vector_ids = array( $base_id );
749	1181
750		- // Find chunk vectors
751		- $list_url = "https://{$host}/vectors/list";
752		- $list_body = array( 'prefix' => $base_id . '_chunk_', 'limit' => 100 );
753		- if ( ! empty($namespace) ) {
754		- $list_body['namespace'] = $namespace;
755		- }
	1182	+ // Find chunk vectors
	1183	+ // NOTE: Pinecone's /vectors/list is a GET endpoint with query parameters.
	1184	+ // A POST is answered 200-with-an-empty-body, which reads as "no vectors".
	1185	+ $list_url = "https://{$host}/vectors/list";
	1186	+ $list_params = array( 'prefix' => $base_id . '_chunk_', 'limit' => 100 );
	1187	+ if ( ! empty($namespace) ) {
	1188	+ $list_params['namespace'] = $namespace;
	1189	+ }
756	1190
757		- $list_resp = wp_remote_post( $list_url, array(
758		- 'headers' => array( 'Api-Key' => $api_key, 'Content-Type' => 'application/json' ),
759		- 'body' => wp_json_encode( $list_body ),
760		- 'timeout' => 15,
761		- ) );
	1191	+ $list_resp = wp_remote_get( $list_url . '?' . http_build_query( $list_params ), array(
	1192	+ 'headers' => array( 'Api-Key' => $api_key, 'accept' => 'application/json' ),
	1193	+ 'timeout' => 15,
	1194	+ ) );
762	1195
763		- if ( ! is_wp_error($list_resp) ) {
764		- $list_data = json_decode( wp_remote_retrieve_body($list_resp), true );
765		- if ( ! empty($list_data['vectors']) ) {
766		- foreach ( $list_data['vectors'] as $v ) {
767		- $vector_ids[] = $v['id'];
	1196	+ if ( ! is_wp_error($list_resp) ) {
	1197	+ $list_data = json_decode( wp_remote_retrieve_body($list_resp), true );
	1198	+ if ( ! empty($list_data['vectors']) ) {
	1199	+ foreach ( $list_data['vectors'] as $v ) {
	1200	+ $vector_ids[] = $v['id'];
	1201	+ }
768	1202	}
769	1203	}
770	1204	}
771	1205
772	1206	// Fetch vectors with metadata
773		- $fetch_url = "https://{$host}/vectors/fetch";
774		- $fetch_body = array( 'ids' => $vector_ids );
	1207	+ // NOTE: /vectors/fetch is a GET endpoint too, and Pinecone expects the ids
	1208	+ // repeated (ids=a&ids=b) — http_build_query would emit ids[0]=a, so build
	1209	+ // the query string explicitly.
	1210	+ $fetch_query = array();
	1211	+ foreach ( $vector_ids as $fetch_vid ) {
	1212	+ $fetch_query[] = 'ids=' . rawurlencode( $fetch_vid );
	1213	+ }
775	1214	if ( ! empty($namespace) ) {
776		- $fetch_body['namespace'] = $namespace;
	1215	+ $fetch_query[] = 'namespace=' . rawurlencode( $namespace );
777	1216	}
778	1217
779		- $fetch_resp = wp_remote_post( $fetch_url, array(
780		- 'headers' => array( 'Api-Key' => $api_key, 'Content-Type' => 'application/json' ),
781		- 'body' => wp_json_encode( $fetch_body ),
	1218	+ $fetch_resp = wp_remote_get( "https://{$host}/vectors/fetch?" . implode( '&', $fetch_query ), array(
	1219	+ 'headers' => array( 'Api-Key' => $api_key, 'accept' => 'application/json' ),
782	1220	'timeout' => 15,
783	1221	) );
784	1222
785	1223	if ( is_wp_error($fetch_resp) ) {
		@@ -814,8 +1252,235 @@
814	1252	);
815	1253	}
816	1254
817	1255	/**
	1256	+ * AJAX: Inspect a knowledge entry — returns the per-chunk stored text + metadata
	1257	+ * WITHOUT collapsing it, so a site owner can see exactly what was indexed for an
	1258	+ * entry (plan-mxchat-20260628-d8cb4b). READ-ONLY: never re-embeds or mutates.
	1259	+ */
	1260	+public function ajax_mxchat_inspect_entry() {
	1261	+ check_ajax_referer('mxchat_inspect_entry_nonce', 'nonce');
	1262	+
	1263	+ if ( ! current_user_can('manage_options') ) {
	1264	+ wp_send_json_error( array( 'message' => esc_html__('Permission denied.', 'mxchat') ) );
	1265	+ }
	1266	+
	1267	+ $source_url = isset($_POST['source_url']) ? sanitize_text_field( wp_unslash($_POST['source_url']) ) : '';
	1268	+ $entry_id = isset($_POST['entry_id']) ? absint($_POST['entry_id']) : 0;
	1269	+ $data_source = isset($_POST['data_source']) ? sanitize_key($_POST['data_source']) : 'wordpress';
	1270	+ $bot_id = isset($_POST['bot_id']) ? sanitize_key($_POST['bot_id']) : 'default';
	1271	+
	1272	+ if ( $data_source === 'pinecone' ) {
	1273	+ $result = $this->inspect_pinecone_entry( $source_url, $entry_id, $bot_id );
	1274	+ } else {
	1275	+ $result = $this->inspect_wordpress_entry( $source_url, $entry_id );
	1276	+ }
	1277	+
	1278	+ if ( is_wp_error( $result ) ) {
	1279	+ wp_send_json_error( array( 'message' => $result->get_error_message() ) );
	1280	+ }
	1281	+
	1282	+ wp_send_json_success( $result );
	1283	+}
	1284	+
	1285	+/**
	1286	+ * Read-only inspector for WordPress-DB entries. Mirrors get_wordpress_entry_content()
	1287	+ * but returns each STORED chunk's exact text + length (no implode), plus the assembled
	1288	+ * embedded text. This shows what is actually in the index, not a re-derivation from the post.
	1289	+ */
	1290	+private function inspect_wordpress_entry( $source_url, $entry_id ) {
	1291	+ global $wpdb;
	1292	+ $table = $wpdb->prefix . 'mxchat_system_prompt_content';
	1293	+
	1294	+ $rows = array();
	1295	+
	1296	+ // Group by the real stored source_url — this INCLUDES "mxchat://" manual
	1297	+ // Direct Content entries (the spec's manual-entry case), which share one
	1298	+ // source_url across their chunk rows. Only the synthetic "_ungrouped_<id>"
	1299	+ // display key (invented by the table view for rows with no source_url) is
	1300	+ // excluded; those fall through to the entry_id lookup below.
	1301	+ if ( ! empty( $source_url ) && strpos( $source_url, '_ungrouped_' ) !== 0 ) {
	1302	+ $rows = $wpdb->get_results( $wpdb->prepare(
	1303	+ "SELECT id, article_content, source_url, content_type FROM {$table} WHERE source_url = %s ORDER BY id ASC",
	1304	+ $source_url
	1305	+ ) );
	1306	+ }
	1307	+
	1308	+ // Fallback / manual "Direct Content" entries: fetch the single row by id.
	1309	+ if ( empty( $rows ) && $entry_id > 0 ) {
	1310	+ $row = $wpdb->get_row( $wpdb->prepare(
	1311	+ "SELECT id, article_content, source_url, content_type FROM {$table} WHERE id = %d",
	1312	+ $entry_id
	1313	+ ) );
	1314	+ if ( $row ) {
	1315	+ $rows = array( $row );
	1316	+ }
	1317	+ }
	1318	+
	1319	+ if ( empty( $rows ) ) {
	1320	+ return new WP_Error( 'not_found', esc_html__('Entry not found in the local knowledge database.', 'mxchat') );
	1321	+ }
	1322	+
	1323	+ $chunks = array();
	1324	+ $content_type = '';
	1325	+ foreach ( $rows as $row ) {
	1326	+ $parsed = MxChat_Chunker::parse_stored_chunk( $row->article_content );
	1327	+ $text = isset( $parsed['text'] ) ? $parsed['text'] : '';
	1328	+ $index = isset( $parsed['metadata']['chunk_index'] ) ? intval( $parsed['metadata']['chunk_index'] ) : count( $chunks );
	1329	+ $content_type = $row->content_type;
	1330	+ $chunks[] = array(
	1331	+ 'index' => $index,
	1332	+ 'text' => $text,
	1333	+ 'length' => function_exists('mb_strlen') ? mb_strlen( $text ) : strlen( $text ),
	1334	+ 'row_id' => intval( $row->id ),
	1335	+ );
	1336	+ }
	1337	+
	1338	+ usort( $chunks, function( $a, $b ) { return $a['index'] - $b['index']; } );
	1339	+
	1340	+ $assembled = implode( "\n\n", wp_list_pluck( $chunks, 'text' ) );
	1341	+
	1342	+ return array(
	1343	+ 'store' => 'wordpress',
	1344	+ 'source_url' => $source_url,
	1345	+ 'content_type' => $content_type,
	1346	+ 'is_chunked' => count( $chunks ) > 1,
	1347	+ 'chunk_count' => count( $chunks ),
	1348	+ 'assembled' => $assembled,
	1349	+ 'assembled_length' => function_exists('mb_strlen') ? mb_strlen( $assembled ) : strlen( $assembled ),
	1350	+ 'chunks' => array_values( $chunks ),
	1351	+ // WP-DB storage carries no separate vector metadata; surface that fact
	1352	+ // rather than letting the owner guess (the spec's taxonomy question).
	1353	+ 'metadata' => array(),
	1354	+ 'metadata_note' => esc_html__('Stored in the local WordPress database. Only the assembled text shown here is embedded — there are no separate vector metadata fields (e.g. taxonomy terms are not stored unless they were injected into the text itself).', 'mxchat'),
	1355	+ );
	1356	+}
	1357	+
	1358	+/**
	1359	+ * Read-only inspector for Pinecone entries. Mirrors get_pinecone_entry_content()
	1360	+ * but keeps each vector's text + metadata instead of imploding, so the owner can
	1361	+ * confirm exactly which metadata fields (text/source_url/type/last_updated/created_at/bot_id)
	1362	+ * are present per chunk. READ-ONLY.
	1363	+ */
	1364	+private function inspect_pinecone_entry( $source_url, $entry_id, $bot_id ) {
	1365	+ if ( ! class_exists('MxChat_Pinecone_Manager') ) {
	1366	+ return new WP_Error( 'pinecone_unavailable', esc_html__('Pinecone manager not available.', 'mxchat') );
	1367	+ }
	1368	+
	1369	+ if ( $bot_id === 'default' || ! class_exists('MxChat_Multi_Bot_Manager') ) {
	1370	+ $pinecone_options = get_option('mxchat_pinecone_addon_options');
	1371	+ $api_key = $pinecone_options['mxchat_pinecone_api_key'] ?? '';
	1372	+ $host = $pinecone_options['mxchat_pinecone_host'] ?? '';
	1373	+ $namespace = $pinecone_options['mxchat_pinecone_namespace'] ?? '';
	1374	+ } else {
	1375	+ $bot_config = apply_filters('mxchat_get_bot_pinecone_config', array(), $bot_id);
	1376	+ $api_key = $bot_config['api_key'] ?? '';
	1377	+ $host = $bot_config['host'] ?? '';
	1378	+ $namespace = $bot_config['namespace'] ?? '';
	1379	+ }
	1380	+
	1381	+ if ( empty($host) || empty($api_key) ) {
	1382	+ return new WP_Error( 'pinecone_config', esc_html__('Pinecone not configured.', 'mxchat') );
	1383	+ }
	1384	+
	1385	+ $base_id = md5( $source_url );
	1386	+ $vector_ids = array( $base_id );
	1387	+
	1388	+ // NOTE: Pinecone's /vectors/list is a GET endpoint with query parameters.
	1389	+ // A POST is answered 200-with-an-empty-body, which reads as "no vectors".
	1390	+ $list_url = "https://{$host}/vectors/list";
	1391	+ $list_params = array( 'prefix' => $base_id . '_chunk_', 'limit' => 100 );
	1392	+ if ( ! empty($namespace) ) {
	1393	+ $list_params['namespace'] = $namespace;
	1394	+ }
	1395	+
	1396	+ $list_resp = wp_remote_get( $list_url . '?' . http_build_query( $list_params ), array(
	1397	+ 'headers' => array( 'Api-Key' => $api_key, 'accept' => 'application/json' ),
	1398	+ 'timeout' => 15,
	1399	+ ) );
	1400	+
	1401	+ if ( ! is_wp_error($list_resp) ) {
	1402	+ $list_data = json_decode( wp_remote_retrieve_body($list_resp), true );
	1403	+ if ( ! empty($list_data['vectors']) ) {
	1404	+ foreach ( $list_data['vectors'] as $v ) {
	1405	+ $vector_ids[] = $v['id'];
	1406	+ }
	1407	+ }
	1408	+ }
	1409	+
	1410	+ // NOTE: /vectors/fetch is a GET endpoint too, and Pinecone expects the ids
	1411	+ // repeated (ids=a&ids=b) — http_build_query would emit ids[0]=a, so build
	1412	+ // the query string explicitly.
	1413	+ $fetch_query = array();
	1414	+ foreach ( $vector_ids as $fetch_vid ) {
	1415	+ $fetch_query[] = 'ids=' . rawurlencode( $fetch_vid );
	1416	+ }
	1417	+ if ( ! empty($namespace) ) {
	1418	+ $fetch_query[] = 'namespace=' . rawurlencode( $namespace );
	1419	+ }
	1420	+
	1421	+ $fetch_resp = wp_remote_get( "https://{$host}/vectors/fetch?" . implode( '&', $fetch_query ), array(
	1422	+ 'headers' => array( 'Api-Key' => $api_key, 'accept' => 'application/json' ),
	1423	+ 'timeout' => 15,
	1424	+ ) );
	1425	+
	1426	+ if ( is_wp_error($fetch_resp) ) {
	1427	+ return new WP_Error( 'pinecone_fetch', esc_html__('Failed to fetch from Pinecone.', 'mxchat') );
	1428	+ }
	1429	+
	1430	+ $fetch_data = json_decode( wp_remote_retrieve_body($fetch_resp), true );
	1431	+ $vectors = $fetch_data['vectors'] ?? array();
	1432	+
	1433	+ if ( empty($vectors) ) {
	1434	+ return new WP_Error( 'not_found', esc_html__('Entry not found in Pinecone.', 'mxchat') );
	1435	+ }
	1436	+
	1437	+ // Whitelisted metadata fields the spec calls out — shown so devs can confirm
	1438	+ // what is (and is NOT) stored per vector.
	1439	+ $meta_fields = array( 'text', 'source_url', 'type', 'last_updated', 'created_at', 'bot_id', 'chunk_index', 'total_chunks' );
	1440	+ $chunks = array();
	1441	+ $content_type = '';
	1442	+ foreach ( $vectors as $vid => $vector ) {
	1443	+ $meta = isset($vector['metadata']) && is_array($vector['metadata']) ? $vector['metadata'] : array();
	1444	+ $text = $meta['text'] ?? '';
	1445	+ $index = isset($meta['chunk_index']) ? intval($meta['chunk_index']) : count($chunks);
	1446	+ $content_type = $meta['type'] ?? $content_type;
	1447	+
	1448	+ $clean_meta = array();
	1449	+ foreach ( $meta_fields as $field ) {
	1450	+ if ( array_key_exists( $field, $meta ) && $field !== 'text' ) {
	1451	+ $clean_meta[ $field ] = is_scalar( $meta[ $field ] ) ? (string) $meta[ $field ] : wp_json_encode( $meta[ $field ] );
	1452	+ }
	1453	+ }
	1454	+
	1455	+ $chunks[] = array(
	1456	+ 'index' => $index,
	1457	+ 'text' => $text,
	1458	+ 'length' => function_exists('mb_strlen') ? mb_strlen( $text ) : strlen( $text ),
	1459	+ 'vector_id' => (string) $vid,
	1460	+ 'metadata' => $clean_meta,
	1461	+ );
	1462	+ }
	1463	+
	1464	+ usort( $chunks, function( $a, $b ) { return $a['index'] - $b['index']; } );
	1465	+
	1466	+ $assembled = implode( "\n\n", wp_list_pluck( $chunks, 'text' ) );
	1467	+
	1468	+ return array(
	1469	+ 'store' => 'pinecone',
	1470	+ 'source_url' => $source_url,
	1471	+ 'content_type' => $content_type,
	1472	+ 'is_chunked' => count( $chunks ) > 1,
	1473	+ 'chunk_count' => count( $chunks ),
	1474	+ 'assembled' => $assembled,
	1475	+ 'assembled_length' => function_exists('mb_strlen') ? mb_strlen( $assembled ) : strlen( $assembled ),
	1476	+ 'chunks' => array_values( $chunks ),
	1477	+ 'metadata' => array(),
	1478	+ 'metadata_note' => esc_html__('Stored in Pinecone. Each chunk above lists the vector metadata fields actually present — if a field you expect (such as taxonomy terms) is missing here, it was not stored as metadata and is only searchable if it appears in the embedded text.', 'mxchat'),
	1479	+ );
	1480	+}
	1481	+
	1482	+/**
818	1483	* AJAX: Save edited content — re-chunks and re-embeds as needed.
819	1484	* Works for both WordPress DB and Pinecone entries.
820	1485	*/
821	1486	public function ajax_mxchat_save_entry_content() {
		@@ -824,9 +1489,9 @@
824	1489	if ( ! current_user_can('manage_options') ) {
825	1490	wp_send_json_error( array( 'message' => 'Permission denied.' ) );
826	1491	}
827	1492
828		- $source_url = isset($_POST['source_url']) ? sanitize_text_field( wp_unslash($_POST['source_url']) ) : '';
	1493	+ $source_url = $this->sanitize_entry_source_url( isset($_POST['source_url']) ? wp_unslash($_POST['source_url']) : '' );
829	1494	$entry_id = isset($_POST['entry_id']) ? absint($_POST['entry_id']) : 0;
830	1495	$content = isset($_POST['content']) ? wp_kses_post( wp_unslash($_POST['content']) ) : '';
831	1496	$data_source = isset($_POST['data_source']) ? sanitize_key($_POST['data_source']) : 'wordpress';
832	1497	$bot_id = isset($_POST['bot_id']) ? sanitize_key($_POST['bot_id']) : 'default';
		@@ -847,38 +1512,74 @@
847	1512	if ( empty($api_key) ) {
848	1513	$api_key = $options['api_key'] ?? '';
849	1514	}
850	1515
851		- global $wpdb;
852		- $table = $wpdb->prefix . 'mxchat_system_prompt_content';
	1516	+ if ( $data_source === 'pinecone' ) {
	1517	+ // Pinecone branch. The WP-DB manual-entry delete below must never run here:
	1518	+ // Pinecone ids are strings, and absint() on a digit-leading md5 hash would
	1519	+ // yield a real (unrelated) WP row id.
	1520	+ $raw_vector_id = isset($_POST['entry_id']) ? sanitize_text_field( wp_unslash($_POST['entry_id']) ) : '';
	1521	+ $is_manual_single = empty($source_url) || strpos($source_url, '_ungrouped_') === 0;
	1522	+ $is_manual_chunked = strpos($source_url, 'mxchat://') === 0;
853	1523
854		- // If source_url is empty but we have an entry_id, look it up
855		- if ( empty($source_url) && $entry_id > 0 && $data_source === 'wordpress' ) {
856		- $row = $wpdb->get_row( $wpdb->prepare( "SELECT source_url FROM {$table} WHERE id = %d", $entry_id ) );
857		- if ( $row && ! empty($row->source_url) ) {
858		- $source_url = $row->source_url;
	1524	+ if ( $is_manual_single || $is_manual_chunked ) {
	1525	+ // Manual content: remove the old vectors first, then store as fresh manual
	1526	+ // content — submit_content_to_db mints a new unique identity (manual_* id
	1527	+ // for a single vector, an mxchat:// chunk prefix if it now chunks).
	1528	+ if ( $is_manual_chunked ) {
	1529	+ // Minted identity: base + chunk vectors share the md5(mxchat://...) prefix.
	1530	+ MxChat_Utils::delete_chunks_for_url( $source_url, $bot_id );
	1531	+ } elseif ( ! empty($raw_vector_id) && class_exists('MxChat_Pinecone_Manager') ) {
	1532	+ $pinecone_manager = MxChat_Pinecone_Manager::get_instance();
	1533	+ $pinecone_options = $pinecone_manager->mxchat_get_bot_pinecone_options( $bot_id );
	1534	+ if ( ! empty($pinecone_options['mxchat_pinecone_api_key']) && ! empty($pinecone_options['mxchat_pinecone_host']) ) {
	1535	+ $pinecone_manager->mxchat_delete_from_pinecone_by_vector_id(
	1536	+ $raw_vector_id,
	1537	+ $pinecone_options['mxchat_pinecone_api_key'],
	1538	+ $pinecone_options['mxchat_pinecone_host'],
	1539	+ $pinecone_options['mxchat_pinecone_namespace'] ?? ''
	1540	+ );
	1541	+ }
	1542	+ }
	1543	+ $result = MxChat_Utils::submit_content_to_db( $content, '', $api_key, null, $bot_id, $content_type );
	1544	+ } else {
	1545	+ // URL-sourced entry: identity is md5(source_url). submit_content_to_db
	1546	+ // handles delete-old-chunks → re-chunk → re-embed → store, and sweeps
	1547	+ // stale chunk vectors when the content now fits in a single vector.
	1548	+ $result = MxChat_Utils::submit_content_to_db( $content, $source_url, $api_key, md5($source_url), $bot_id, $content_type );
859	1549	}
860		- }
	1550	+ } else {
	1551	+ global $wpdb;
	1552	+ $table = $wpdb->prefix . 'mxchat_system_prompt_content';
861	1553
862		- // For manual entries (no source_url or mxchat:// prefix), delete the old entry by ID first
863		- // so submit_content_to_db creates a replacement instead of a duplicate
864		- // Also treat legacy mxchat.ai source URLs as manual — old bug assigned the site URL to manual entries
865		- $is_legacy_manual = !empty($source_url) && strpos($source_url, 'mxchat.ai') !== false && strpos($source_url, 'mxchat://') !== 0;
866		- if ( $entry_id > 0 && (empty($source_url) || strpos($source_url, 'mxchat://') === 0 || $is_legacy_manual) ) {
867		- $wpdb->delete( $table, array( 'id' => $entry_id ), array( '%d' ) );
868		- // Clear legacy URL so submit_content_to_db generates a unique mxchat:// identifier
869		- // instead of reusing the shared URL (which would mass-delete other entries with the same URL)
870		- if ( $is_legacy_manual ) {
871		- $source_url = '';
	1554	+ // If source_url is empty but we have an entry_id, look it up
	1555	+ if ( empty($source_url) && $entry_id > 0 ) {
	1556	+ $row = $wpdb->get_row( $wpdb->prepare( "SELECT source_url FROM {$table} WHERE id = %d", $entry_id ) );
	1557	+ if ( $row && ! empty($row->source_url) ) {
	1558	+ $source_url = $row->source_url;
	1559	+ }
872	1560	}
873		- }
874	1561
875		- // Use the existing submit_content_to_db which handles chunking, Pinecone, and WP DB
876		- $vector_id = ! empty($source_url) ? md5($source_url) : md5('mxchat_manual_' . $entry_id);
	1562	+ // For manual entries (no source_url or mxchat:// prefix), delete the old entry by ID first
	1563	+ // so submit_content_to_db creates a replacement instead of a duplicate
	1564	+ // Also treat legacy mxchat.ai source URLs as manual — old bug assigned the site URL to manual entries
	1565	+ $is_legacy_manual = !empty($source_url) && strpos($source_url, 'mxchat.ai') !== false && strpos($source_url, 'mxchat://') !== 0;
	1566	+ if ( $entry_id > 0 && (empty($source_url) || strpos($source_url, 'mxchat://') === 0 || $is_legacy_manual) ) {
	1567	+ $wpdb->delete( $table, array( 'id' => $entry_id ), array( '%d' ) );
	1568	+ // Clear legacy URL so submit_content_to_db generates a unique mxchat:// identifier
	1569	+ // instead of reusing the shared URL (which would mass-delete other entries with the same URL)
	1570	+ if ( $is_legacy_manual ) {
	1571	+ $source_url = '';
	1572	+ }
	1573	+ }
877	1574
878		- // submit_content_to_db already handles: delete old chunks → re-chunk → re-embed → store
879		- $result = MxChat_Utils::submit_content_to_db( $content, $source_url, $api_key, $vector_id, $bot_id, $content_type );
	1575	+ // Use the existing submit_content_to_db which handles chunking, Pinecone, and WP DB
	1576	+ $vector_id = ! empty($source_url) ? md5($source_url) : md5('mxchat_manual_' . $entry_id);
880	1577
	1578	+ // submit_content_to_db already handles: delete old chunks → re-chunk → re-embed → store
	1579	+ $result = MxChat_Utils::submit_content_to_db( $content, $source_url, $api_key, $vector_id, $bot_id, $content_type );
	1580	+ }
	1581	+
881	1582	if ( is_wp_error($result) ) {
882	1583	wp_send_json_error( array( 'message' => $result->get_error_message() ) );
883	1584	}
884	1585
		@@ -964,42 +1665,34 @@
964	1665
965	1666	// Get bot_id from form submission
966	1667	$bot_id = isset($_POST['bot_id']) ? sanitize_key($_POST['bot_id']) : 'default';
967	1668
968		- // Get bot-specific options and validate API key
	1669	+ // Get bot-specific options and validate the embedding decision —
	1670	+ // custom-provider-aware (plan cbd5fd).
969	1671	$bot_options = $this->get_bot_options($bot_id);
970	1672	$options = !empty($bot_options) ? $bot_options : get_option('mxchat_options');
971		- $selected_model = $options['embedding_model'] ?? 'text-embedding-ada-002';
972		-
973		- if (strpos($selected_model, 'voyage') === 0) {
974		- $api_key = $options['voyage_api_key'] ?? '';
975		- $provider_name = 'Voyage AI';
976		- } elseif (strpos($selected_model, 'gemini-embedding') === 0) {
977		- $api_key = $options['gemini_api_key'] ?? '';
978		- $provider_name = 'Google Gemini';
979		- } else {
980		- $api_key = $options['api_key'] ?? '';
981		- $provider_name = 'OpenAI';
982		- }
983		-
984		- if (empty($api_key)) {
985		- $error_message = sprintf(
986		- esc_html__('%s API key is not configured. Please add your API key in the settings before submitting content.', 'mxchat'),
987		- $provider_name
988		- );
989		- set_transient('mxchat_admin_notice_error', $error_message, 30);
	1673	+
	1674	+ $preflight = MxChat_Utils::embedding_preflight($options);
	1675	+ if (!$preflight['ok']) {
	1676	+ set_transient('mxchat_admin_notice_error', esc_html($preflight['reason']), 30);
990	1677	wp_safe_redirect(esc_url(admin_url('admin.php?page=mxchat-prompts')));
991	1678	exit;
992	1679	}
	1680	+ $api_key = $preflight['api_key'];
993	1681
994		- // Fetch URL — use browser-like headers so servers with bot protection don't block us
	1682	+ // Fetch URL — send an honest, versioned MXChat crawler UA (not a spoofed
	1683	+ // browser). Stale browser UAs are exactly what WAFs like SiteGround's
	1684	+ // ModSecurity flag as scrapers, 403-ing the fetch (including PDFs served
	1685	+ // from the site's own media library, which route through this same call).
	1686	+ // See mxchat_ingest_user_agent(). Accept is kept for content negotiation;
	1687	+ // the browser-only Accept-Language fingerprint is dropped so it stays
	1688	+ // coherent with a bot identity.
995	1689	$response = wp_remote_get($submitted_url, array(
996	1690	'timeout' => 30,
997	1691	'sslverify' => false,
998		- 'user-agent' => 'Mozilla/5.0 (Windows NT 10.0; Win64; x64) AppleWebKit/537.36 (KHTML, like Gecko) Chrome/131.0.0.0 Safari/537.36',
	1692	+ 'user-agent' => mxchat_ingest_user_agent(),
999	1693	'headers' => array(
1000	1694	'Accept' => 'text/html,application/xhtml+xml,application/xml;q=0.9,*/*;q=0.8',
1001		- 'Accept-Language' => 'en-US,en;q=0.9',
1002	1695	),
1003	1696	));
1004	1697
1005	1698	if (is_wp_error($response) || wp_remote_retrieve_response_code($response) !== 200) {
		@@ -1070,12 +1763,22 @@
1070	1763	esc_html__('Sitemap queued for processing. Processing will start automatically.', 'mxchat'),
1071	1764	30
1072	1765	);
1073	1766	} else {
1074		- set_transient('mxchat_admin_notice_error',
1075		- esc_html__('Failed to queue sitemap processing. Please check the status below for details.', 'mxchat'),
1076		- 30
1077		- );
	1767	+ // Surface the reason the handler already computed (embedding pre-flight,
	1768	+ // empty sitemap, queue failure). The old message pointed at the status
	1769	+ // area, which is empty on this path — nothing was ever queued.
	1770	+ if (is_string($result) && $result !== '') {
	1771	+ set_transient('mxchat_admin_notice_error',
	1772	+ esc_html__('Failed to queue sitemap processing: ', 'mxchat') . esc_html($result),
	1773	+ 30
	1774	+ );
	1775	+ } else {
	1776	+ set_transient('mxchat_admin_notice_error',
	1777	+ esc_html__('Failed to queue sitemap processing.', 'mxchat'),
	1778	+ 30
	1779	+ );
	1780	+ }
1078	1781	}
1079	1782
1080	1783	wp_safe_redirect(esc_url(admin_url('admin.php?page=mxchat-prompts')));
1081	1784	exit;
		@@ -1218,8 +1921,140 @@
1218	1921	*
1219	1922	* @param string $content The content containing shortcodes
1220	1923	* @return string Content with shortcode tags removed but inner content preserved
1221	1924	*/
	1925	+/**
	1926	+ * Single-pass HTML entity decode for text entering the knowledge base.
	1927	+ * The corpus should hold what a human reads: a stored `&amp;` consumes
	1928	+ * extra tokens, distorts the vector away from the form a visitor's
	1929	+ * question uses, and can be quoted back verbatim in an answer.
	1930	+ * Deliberately NOT looped to a fixed point — a stored `&amp;amp;` is a
	1931	+ * legitimate literal `&amp;` and must not collapse further (data loss).
	1932	+ * UTF-8 charset keeps multibyte (CJK/RTL) text untouched. Both assembly
	1933	+ * paths call this at their output points so the treatment cannot drift.
	1934	+ * (Plan d2c92e.)
	1935	+ */
	1936	+private function mxchat_decode_entities_for_indexing($text) {
	1937	+ return html_entity_decode((string) $text, ENT_QUOTES | ENT_HTML5, 'UTF-8');
	1938	+}
	1939	+
	1940	+/**
	1941	+ * Price lines for a product's indexed text, pinned to the store's BASE currency.
	1942	+ *
	1943	+ * The four product assembly paths each used to call get_woocommerce_currency_symbol()
	1944	+ * with no argument, which resolves the currency active on the CURRENT request.
	1945	+ * Multi-currency plugins (CURCY, WOOCS, Aelia, WPML Multicurrency) filter that per
	1946	+ * request, so whichever currency the store happened to be serving when an import ran
	1947	+ * was frozen into every product it indexed. The amounts have the mirror problem: the
	1948	+ * woocommerce_product_get_* filters convert prices in the 'view' context but not in
	1949	+ * 'edit', so a converted amount could be paired with an unconverted symbol and produce
	1950	+ * a price that is not merely wrong but incoherent.
	1951	+ *
	1952	+ * Base currency option + 'edit' context makes both halves agree and makes the output
	1953	+ * independent of when the import ran. The currency CODE is emitted alongside the symbol
	1954	+ * so the model cannot read a bare "$" as USD on a store whose plugin swapped symbols.
	1955	+ * (Plan 7403ec.)
	1956	+ */
	1957	+private function mxchat_product_price_lines($product) {
	1958	+ if (!is_object($product) || !method_exists($product, 'get_regular_price')) {
	1959	+ return '';
	1960	+ }
	1961	+
	1962	+ $currency = get_option('woocommerce_currency');
	1963	+ $currency = is_string($currency) ? trim($currency) : '';
	1964	+ $symbol = ($currency !== '')
	1965	+ ? get_woocommerce_currency_symbol($currency)
	1966	+ : get_woocommerce_currency_symbol();
	1967	+ $symbol = $this->mxchat_decode_entities_for_indexing($symbol);
	1968	+
	1969	+ $regular_price = $product->get_regular_price('edit');
	1970	+ $sale_price = $product->get_sale_price('edit');
	1971	+ $price = $product->get_price('edit');
	1972	+
	1973	+ $lines = '';
	1974	+
	1975	+ if (!empty($regular_price)) {
	1976	+ $lines .= "Price: " . $this->mxchat_format_indexed_price($regular_price, $currency, $symbol) . "\n";
	1977	+ } elseif (!empty($price)) {
	1978	+ $lines .= "Price: " . $this->mxchat_format_indexed_price($price, $currency, $symbol) . "\n";
	1979	+ }
	1980	+
	1981	+ if (!empty($sale_price) && $sale_price !== $regular_price) {
	1982	+ $lines .= "Sale Price: " . $this->mxchat_format_indexed_price($sale_price, $currency, $symbol) . "\n";
	1983	+ }
	1984	+
	1985	+ if ($product->is_type('variable')) {
	1986	+ list($min_price, $max_price) = $this->mxchat_variation_price_range($product);
	1987	+ if ($min_price !== null && $max_price !== null && (float) $min_price !== (float) $max_price) {
	1988	+ $lines .= "Price Range: " . $this->mxchat_format_indexed_price($min_price, $currency, $symbol)
	1989	+ . " - " . $this->mxchat_format_indexed_price($max_price, $currency, $symbol) . "\n";
	1990	+ }
	1991	+ }
	1992	+
	1993	+ return $lines;
	1994	+}
	1995	+
	1996	+/**
	1997	+ * One indexed price amount, labelled with its currency code.
	1998	+ *
	1999	+ * "INR 1299.00 (Rs.1299.00)" — the code is what the model should reason from; the symbol
	2000	+ * is kept so a quoted price still reads naturally. Falls back to the old symbol-only
	2001	+ * shape when WooCommerce has no base currency configured, and drops the parenthetical
	2002	+ * when the symbol is absent or IS the code (several currencies have no distinct glyph).
	2003	+ */
	2004	+private function mxchat_format_indexed_price($amount, $currency, $symbol) {
	2005	+ $amount = (string) $amount;
	2006	+
	2007	+ if ($currency === '') {
	2008	+ return $symbol . $amount;
	2009	+ }
	2010	+
	2011	+ if ($symbol === '' || $symbol === $currency) {
	2012	+ return $currency . ' ' . $amount;
	2013	+ }
	2014	+
	2015	+ return $currency . ' ' . $amount . ' (' . $symbol . $amount . ')';
	2016	+}
	2017	+
	2018	+/**
	2019	+ * Min/max variation price read from the variations themselves in 'edit' context.
	2020	+ *
	2021	+ * get_variation_price() reads WooCommerce's display price cache, which multi-currency
	2022	+ * plugins populate with converted values — the same defect the rest of this helper
	2023	+ * exists to remove. Returns raw stored strings (not floats) so the indexed text keeps
	2024	+ * the store's own price formatting, and (null, null) when no variation carries a price.
	2025	+ */
	2026	+private function mxchat_variation_price_range($product) {
	2027	+ $min_raw = null;
	2028	+ $max_raw = null;
	2029	+ $min_val = null;
	2030	+ $max_val = null;
	2031	+
	2032	+ $children = method_exists($product, 'get_children') ? $product->get_children() : array();
	2033	+
	2034	+ foreach ($children as $child_id) {
	2035	+ $variation = wc_get_product($child_id);
	2036	+ if (!$variation) {
	2037	+ continue;
	2038	+ }
	2039	+ $raw = $variation->get_price('edit');
	2040	+ if ($raw === '' || $raw === null) {
	2041	+ continue;
	2042	+ }
	2043	+ $val = (float) $raw;
	2044	+ if ($min_val === null || $val < $min_val) {
	2045	+ $min_val = $val;
	2046	+ $min_raw = $raw;
	2047	+ }
	2048	+ if ($max_val === null || $val > $max_val) {
	2049	+ $max_val = $val;
	2050	+ $max_raw = $raw;
	2051	+ }
	2052	+ }
	2053	+
	2054	+ return array($min_raw, $max_raw);
	2055	+}
	2056	+
1222	2057	private function strip_shortcode_tags_preserve_content($content) {
1223	2058	// Single-pass regex removes all shortcode brackets: [tag], [tag attr="val"], [/tag], [tag /]
1224	2059	// Content between tags is inherently preserved since only brackets are targeted
1225	2060	$result = preg_replace('/\[\/?\w[\w-]*[^\]]*\]/', '', $content);
		@@ -1263,24 +2098,37 @@
1263	2098
1264	2099	// Ensure valid UTF-8 encoding
1265	2100	$content = wp_check_invalid_utf8($content);
1266	2101
1267		- // Remove any extremely long strings without spaces (often garbage)
1268		- $content = preg_replace('/\S{300,}/', ' ', $content);
1269		-
1270		- // Replace problematic characters that often cause database issues
1271		- $content = preg_replace('/[\x{10000}-\x{10FFFF}]/u', '', $content); // Remove emoji and other high Unicode characters
1272		-
1273		- // Replace any remaining potentially problematic characters with spaces
1274		- // BUT preserve newlines by temporarily replacing them
1275		- $content = str_replace("\n", "NEWLINE_PLACEHOLDER", $content);
1276		- $content = preg_replace('/[^\p{L}\p{N}\p{P}\p{Z}\p{Sm}]/u', ' ', $content);
1277		- $content = str_replace("NEWLINE_PLACEHOLDER", "\n", $content);
1278		-
1279		- // Limit to reasonable length if needed
	2102	+ // Remove extremely long runs with no whitespace (base64 blobs, minified JS).
	2103	+ // Counts CHARACTERS (/u), and never strips a run containing characters from a
	2104	+ // script written without spaces — Japanese, Chinese, Thai, Khmer, Lao, Myanmar —
	2105	+ // where a normal paragraph is legitimately one unbroken run.
	2106	+ $content = preg_replace_callback('/\S{300,}/u', function ($m) {
	2107	+ return preg_match('/[\p{Han}\p{Hiragana}\p{Katakana}\p{Thai}\p{Khmer}\p{Lao}\p{Myanmar}]/u', $m[0]) ? $m[0] : ' ';
	2108	+ }, $content);
	2109	+
	2110	+ // Remove emoji/symbol blocks only — not the whole supplementary plane, which
	2111	+ // also holds CJK Extension B ideographs used in real Chinese/Japanese names.
	2112	+ // A ZWJ (U+200D) BETWEEN stripped pictographs is consumed with them, so a
	2113	+ // family sequence like 👨‍👩‍👧 leaves no invisible zero-width residue behind
	2114	+ // (the joiner between NON-emoji characters — Hindi conjuncts — is untouched).
	2115	+ $content = preg_replace('/[\x{1F000}-\x{1F0FF}\x{1F300}-\x{1FAFF}](?:\x{200D}[\x{1F000}-\x{1F0FF}\x{1F300}-\x{1FAFF}])*/u', '', $content);
	2116	+
	2117	+ // There is deliberately NO catch-all character allowlist here (plan 209e57;
	2118	+ // one existed until 3.2.20). Every genuinely dangerous byte is already gone:
	2119	+ // control characters, null bytes, invalid UTF-8 and the emoji blocks are all
	2120	+ // stripped above. The allowlist's only remaining effect was to damage scripts
	2121	+ // nobody thought to enumerate — Unicode Cf (Format) was missing, so it
	2122	+ // replaced the zero-width joiner/non-joiner with spaces and silently split
	2123	+ // Persian words (می‌روم → می روم) and broke Hindi conjuncts (क्‍ष → क् ष).
	2124	+ // Do not add one back; the failure mode of an allowlist is exactly this.
	2125	+
	2126	+ // Limit to reasonable length if needed (byte limit — MySQL TEXT is byte-sized,
	2127	+ // but cut on a character boundary so a multibyte char is never split mid-sequence)
1280	2128	$max_length = 65000; // Just under MySQL TEXT field limit
1281	2129	if (strlen($content) > $max_length) {
1282		- $content = substr($content, 0, $max_length);
	2130	+ $content = mb_strcut($content, 0, $max_length, 'UTF-8');
1283	2131	}
1284	2132
1285	2133	//error_log('[MXCHAT-SANITIZE] Sanitized content preview: ' . substr($content, 0, 500) . '...');
1286	2134	return $content;
		@@ -2118,10 +2966,19 @@
2118	2966	<span style="color: var(--mxch-text-muted);"><?php esc_html_e('Manual Content', 'mxchat'); ?></span>
2119	2967	<?php endif; ?>
2120	2968	</td>
2121	2969	<td class="mxchat-actions-cell" style="padding: 12px 16px; white-space: nowrap;">
2122		- <?php if ($data_source !== 'pinecone') : ?>
2123	2970	<button type="button"
	2971	+ class="mxch-btn mxch-btn-ghost mxch-btn-sm mxchat-inspect-entry-btn"
	2972	+ data-source-url="<?php echo esc_attr($source_url); ?>"
	2973	+ data-entry-id="<?php echo esc_attr($first_prompt->id); ?>"
	2974	+ data-data-source="<?php echo esc_attr($data_source); ?>"
	2975	+ data-bot-id="<?php echo esc_attr($current_bot_id); ?>"
	2976	+ data-nonce="<?php echo wp_create_nonce('mxchat_inspect_entry_nonce'); ?>"
	2977	+ title="<?php esc_attr_e('View indexed content', 'mxchat'); ?>">
	2978	+ <span class="dashicons dashicons-visibility" style="font-size: 14px;"></span>
	2979	+ </button>
	2980	+ <button type="button"
2124	2981	class="mxch-btn mxch-btn-ghost mxch-btn-sm mxchat-edit-entry-btn"
2125	2982	data-source-url="<?php echo esc_attr($source_url); ?>"
2126	2983	data-entry-id="<?php echo esc_attr($first_prompt->id); ?>"
2127	2984	data-data-source="<?php echo esc_attr($data_source); ?>"
		@@ -2129,9 +2986,8 @@
2129	2986	data-nonce="<?php echo wp_create_nonce('mxchat_edit_entry_nonce'); ?>"
2130	2987	title="<?php esc_attr_e('Edit content', 'mxchat'); ?>">
2131	2988	<span class="dashicons dashicons-edit" style="font-size: 14px;"></span>
2132	2989	</button>
2133		- <?php endif; ?>
2134	2990	<button type="button"
2135	2991	class="mxch-btn mxch-btn-ghost mxch-btn-sm delete-button-group"
2136	2992	data-source-url="<?php echo esc_attr($source_url); ?>"
2137	2993	data-chunk-count="<?php echo esc_attr($chunk_count); ?>"
		@@ -2263,9 +3119,29 @@
2263	3119	<?php else : ?>
2264	3120	<span style="color: var(--mxch-text-muted);"><?php esc_html_e('Manual', 'mxchat'); ?></span>
2265	3121	<?php endif; ?>
2266	3122	</td>
2267		- <td style="padding: 12px 16px;">
	3123	+ <td style="padding: 12px 16px; white-space: nowrap;">
	3124	+ <button type="button"
	3125	+ class="mxch-btn mxch-btn-ghost mxch-btn-sm mxchat-inspect-entry-btn"
	3126	+ data-source-url="<?php echo esc_attr($prompt->source_url ?? ''); ?>"
	3127	+ data-entry-id="<?php echo esc_attr($prompt->id); ?>"
	3128	+ data-data-source="<?php echo esc_attr($data_source); ?>"
	3129	+ data-bot-id="<?php echo esc_attr($current_bot_id); ?>"
	3130	+ data-nonce="<?php echo wp_create_nonce('mxchat_inspect_entry_nonce'); ?>"
	3131	+ title="<?php esc_attr_e('View indexed content', 'mxchat'); ?>">
	3132	+ <span class="dashicons dashicons-visibility" style="font-size: 14px;"></span>
	3133	+ </button>
	3134	+ <button type="button"
	3135	+ class="mxch-btn mxch-btn-ghost mxch-btn-sm mxchat-edit-entry-btn"
	3136	+ data-source-url="<?php echo esc_attr($prompt->source_url ?? ''); ?>"
	3137	+ data-entry-id="<?php echo esc_attr($prompt->id); ?>"
	3138	+ data-data-source="<?php echo esc_attr($data_source); ?>"
	3139	+ data-bot-id="<?php echo esc_attr($current_bot_id); ?>"
	3140	+ data-nonce="<?php echo wp_create_nonce('mxchat_edit_entry_nonce'); ?>"
	3141	+ title="<?php esc_attr_e('Edit content', 'mxchat'); ?>">
	3142	+ <span class="dashicons dashicons-edit" style="font-size: 14px;"></span>
	3143	+ </button>
2268	3144	<button type="button" class="mxch-btn mxch-btn-ghost mxch-btn-sm delete-button-ajax" data-vector-id="<?php echo esc_attr($prompt->id); ?>" data-bot-id="<?php echo esc_attr($current_bot_id); ?>" data-nonce="<?php echo wp_create_nonce('mxchat_delete_pinecone_prompt_nonce'); ?>" style="color: var(--mxch-error);">
2269	3145	<span class="dashicons dashicons-trash" style="font-size: 14px;"></span>
2270	3146	</button>
2271	3147	</td>
		@@ -2838,9 +3714,9 @@
2838	3714	$response = wp_remote_head($url, array(
2839	3715	'timeout' => 10,
2840	3716	'sslverify' => false,
2841	3717	'redirection' => 1,
2842		- 'user-agent' => 'Mozilla/5.0 (Windows NT 10.0; Win64; x64) AppleWebKit/537.36 (KHTML, like Gecko) Chrome/131.0.0.0 Safari/537.36',
	3718	+ 'user-agent' => mxchat_ingest_user_agent(),
2843	3719	));
2844	3720
2845	3721	if (!is_wp_error($response) && wp_remote_retrieve_response_code($response) === 200) {
2846	3722	// Found a sitemap index - parse it to get sub-sitemaps
		@@ -2900,12 +3776,11 @@
2900	3776
2901	3777	$response = wp_remote_get($url, array(
2902	3778	'timeout' => 30,
2903	3779	'sslverify' => false,
2904		- 'user-agent' => 'Mozilla/5.0 (Windows NT 10.0; Win64; x64) AppleWebKit/537.36 (KHTML, like Gecko) Chrome/131.0.0.0 Safari/537.36',
	3780	+ 'user-agent' => mxchat_ingest_user_agent(),
2905	3781	'headers' => array(
2906	3782	'Accept' => 'text/html,application/xhtml+xml,application/xml;q=0.9,*/*;q=0.8',
2907		- 'Accept-Language' => 'en-US,en;q=0.9',
2908	3783	),
2909	3784	));
2910	3785
2911	3786	if (is_wp_error($response)) {
		@@ -2959,12 +3834,11 @@
2959	3834	private function get_sitemap_url_count($url) {
2960	3835	$response = wp_remote_get($url, array(
2961	3836	'timeout' => 30,
2962	3837	'sslverify' => false,
2963		- 'user-agent' => 'Mozilla/5.0 (Windows NT 10.0; Win64; x64) AppleWebKit/537.36 (KHTML, like Gecko) Chrome/131.0.0.0 Safari/537.36',
	3838	+ 'user-agent' => mxchat_ingest_user_agent(),
2964	3839	'headers' => array(
2965	3840	'Accept' => 'text/html,application/xhtml+xml,application/xml;q=0.9,*/*;q=0.8',
2966		- 'Accept-Language' => 'en-US,en;q=0.9',
2967	3841	),
2968	3842	));
2969	3843
2970	3844	if (is_wp_error($response)) {
		@@ -2990,9 +3864,9 @@
2990	3864
2991	3865	$response = wp_remote_get($robots_url, array(
2992	3866	'timeout' => 15,
2993	3867	'sslverify' => false,
2994		- 'user-agent' => 'Mozilla/5.0 (Windows NT 10.0; Win64; x64) AppleWebKit/537.36 (KHTML, like Gecko) Chrome/131.0.0.0 Safari/537.36',
	3868	+ 'user-agent' => mxchat_ingest_user_agent(),
2995	3869	));
2996	3870
2997	3871	if (is_wp_error($response)) {
2998	3872	return $sitemaps;
		@@ -3482,9 +4356,14 @@
3482	4356	}
3483	4357
3484	4358	// Get bot_id from request
3485	4359	$bot_id = isset($_POST['bot_id']) ? sanitize_key($_POST['bot_id']) : 'default';
3486		-
	4360	+
	4361	+ // ACF→PDF extraction is an install-level setting (Knowledge → ACF Fields,
	4362	+ // plan 11720c). The import modal shows a passive status line pointing
	4363	+ // there; the old per-batch checkbox and its remembered default are gone.
	4364	+ $extract_acf_pdfs = get_option('mxchat_acf_pdf_extraction', '0') === '1';
	4365	+
3487	4366	// Process only ONE post at a time to avoid request size issues
3488	4367	$post_id = reset($post_ids);
3489	4368	$post = get_post($post_id);
3490	4369
		@@ -3492,139 +4371,34 @@
3492	4371	wp_send_json_error('Post not found');
3493	4372	exit;
3494	4373	}
3495	4374
3496		- // Allow developers to modify post data before processing into knowledge base
	4375	+ /**
	4376	+ * Allow developers to modify post data before processing into the knowledge base.
	4377	+ * Applied on BOTH content-preparation paths (this manual bulk import and the
	4378	+ * auto-sync path in mxchat_handle_post_update) with the same signature, so a
	4379	+ * callback registered once covers every indexing route. Purely additive —
	4380	+ * zero behaviour change when unhooked.
	4381	+ *
	4382	+ * @param WP_Post $post The post about to be indexed.
	4383	+ * @param string $bot_id Bot context for this import.
	4384	+ */
3497	4385	$post = apply_filters('mxchat_before_process_post', $post, $bot_id);
3498		-
3499		- // Get content including title, short description (for WooCommerce), and main content
3500		- $content = $post->post_title . "\n\n";
3501		-
3502		- // Add short description if it exists (WooCommerce products use post_excerpt for short description)
3503		- if (!empty($post->post_excerpt)) {
3504		- // Remove shortcode tags but preserve content inside them
3505		- $clean_excerpt = $this->strip_shortcode_tags_preserve_content($post->post_excerpt);
3506		- $content .= "Short Description: " . wp_strip_all_tags($clean_excerpt) . "\n\n";
	4386	+ if (!($post instanceof WP_Post)) {
	4387	+ $post = get_post($post_id); // defend against a bad callback return
3507	4388	}
3508	4389
3509		- // Add main content - remove shortcode tags but preserve content inside them
3510		- $clean_content = $this->strip_shortcode_tags_preserve_content($post->post_content);
3511		- $content .= wp_strip_all_tags($clean_content);
	4390	+ // Assemble the indexable text via the shared post-kind assembler (a3d60c).
	4391	+ // Bulk import reads raw post fields, has always included product custom tabs,
	4392	+ // and passes the install-level ACF→PDF option.
	4393	+ $prepared = $this->mxchat_prepare_post_content_for_indexing($post_id, $post, array(
	4394	+ 'read_display' => false,
	4395	+ 'extract_acf_pdfs' => $extract_acf_pdfs,
	4396	+ 'include_product_tabs' => true,
	4397	+ ));
	4398	+ $content = $prepared['content'];
	4399	+ $pdf_extracted_count = $prepared['pdf_extracted_count'];
3512	4400
3513		- // ADD WOOCOMMERCE PRODUCT DATA (pricing, stock, categories, custom tabs)
3514		- if (get_post_type($post_id) === 'product' && class_exists('WooCommerce')) {
3515		- $product = wc_get_product($post_id);
3516		-
3517		- if ($product) {
3518		- // Get pricing information
3519		- $regular_price = $product->get_regular_price();
3520		- $sale_price = $product->get_sale_price();
3521		- $price = $product->get_price();
3522		- $sku = $product->get_sku();
3523		-
3524		- // Get currency symbol
3525		- $currency_symbol = get_woocommerce_currency_symbol();
3526		-
3527		- // Add pricing information
3528		- $content .= "\n";
3529		- if (!empty($regular_price)) {
3530		- $content .= "Price: " . $currency_symbol . $regular_price . "\n";
3531		- } elseif (!empty($price)) {
3532		- $content .= "Price: " . $currency_symbol . $price . "\n";
3533		- }
3534		-
3535		- if (!empty($sale_price) && $sale_price !== $regular_price) {
3536		- $content .= "Sale Price: " . $currency_symbol . $sale_price . "\n";
3537		- }
3538		-
3539		- // Handle variable products - show price range
3540		- if ($product->is_type('variable')) {
3541		- $min_price = $product->get_variation_price('min');
3542		- $max_price = $product->get_variation_price('max');
3543		- if ($min_price !== $max_price) {
3544		- $content .= "Price Range: " . $currency_symbol . $min_price . " - " . $currency_symbol . $max_price . "\n";
3545		- }
3546		- }
3547		-
3548		- if (!empty($sku)) {
3549		- $content .= "SKU: " . $sku . "\n";
3550		- }
3551		-
3552		- // Get product categories
3553		- $categories = wp_get_post_terms($post_id, 'product_cat', array('fields' => 'names'));
3554		- if (!empty($categories) && !is_wp_error($categories)) {
3555		- $content .= "Categories: " . implode(', ', $categories) . "\n";
3556		- }
3557		- }
3558		-
3559		- // Get Custom Product Tabs (supports "Custom Product Tabs for WooCommerce" by Code Parrots)
3560		- $custom_tabs = get_post_meta($post_id, 'yikes_woo_products_tabs', true);
3561		- if (!empty($custom_tabs) && is_array($custom_tabs)) {
3562		- foreach ($custom_tabs as $tab) {
3563		- $tab_title = isset($tab['title']) ? $tab['title'] : (isset($tab['tab_title']) ? $tab['tab_title'] : '');
3564		- $tab_content = isset($tab['content']) ? $tab['content'] : '';
3565		-
3566		- if (!empty($tab_title) && !empty($tab_content)) {
3567		- $content .= "\n" . $tab_title . ": " . wp_strip_all_tags($tab_content) . "\n";
3568		- }
3569		- }
3570		- }
3571		-
3572		- // Also check for reusable/saved tabs applied to this product
3573		- $applied_saved_tabs = get_post_meta($post_id, 'yikes_woo_reusable_products_tabs_applied', true);
3574		- if (!empty($applied_saved_tabs) && is_array($applied_saved_tabs)) {
3575		- $saved_tabs = get_option('yikes_woo_reusable_products_tabs', array());
3576		- if (!empty($saved_tabs) && is_array($saved_tabs)) {
3577		- foreach ($applied_saved_tabs as $saved_tab_id) {
3578		- if (isset($saved_tabs[$saved_tab_id])) {
3579		- $tab = $saved_tabs[$saved_tab_id];
3580		- $tab_title = isset($tab['title']) ? $tab['title'] : (isset($tab['tab_title']) ? $tab['tab_title'] : '');
3581		- $tab_content = isset($tab['content']) ? $tab['content'] : '';
3582		-
3583		- if (!empty($tab_title) && !empty($tab_content)) {
3584		- $content .= "\n" . $tab_title . ": " . wp_strip_all_tags($tab_content) . "\n";
3585		- }
3586		- }
3587		- }
3588		- }
3589		- }
3590		- }
3591		-
3592		- // ADD ACF FIELDS SUPPORT
3593		- $acf_fields = $this->mxchat_get_acf_fields_for_post($post_id);
3594		- if (!empty($acf_fields)) {
3595		- $acf_content_parts = array();
3596		-
3597		- foreach ($acf_fields as $field_name => $field_value) {
3598		- $formatted_value = $this->mxchat_format_acf_field_value($field_value, $field_name, $post_id);
3599		-
3600		- if (!empty($formatted_value)) {
3601		- $field_label = ucwords(str_replace('_', ' ', $field_name));
3602		- $acf_content_parts[] = $field_label . ": " . $formatted_value;
3603		- }
3604		- }
3605		-
3606		- if (!empty($acf_content_parts)) {
3607		- $content .= "\n\n" . implode("\n", $acf_content_parts);
3608		- }
3609		- }
3610		-
3611		- // ADD CUSTOM POST META SUPPORT (whitelisted non-ACF meta fields)
3612		- $custom_meta = $this->mxchat_get_whitelisted_post_meta($post_id);
3613		- if (!empty($custom_meta)) {
3614		- $meta_content_parts = array();
3615		-
3616		- foreach ($custom_meta as $meta_key => $meta_value) {
3617		- // Convert meta key to readable label
3618		- $meta_label = ucwords(str_replace(array('_', '-'), ' ', $meta_key));
3619		- $meta_content_parts[] = $meta_label . ": " . $meta_value;
3620		- }
3621		-
3622		- if (!empty($meta_content_parts)) {
3623		- $content .= "\n\n" . implode("\n", $meta_content_parts);
3624		- }
3625		- }
3626		-
3627	4401	// Debug logging for WordPress Import content
3628	4402	//error_log('[MXCHAT-WP-IMPORT-DEBUG] Post ID: ' . $post_id . ' Title: ' . $post->post_title);
3629	4403	//error_log('[MXCHAT-WP-IMPORT-DEBUG] Raw post_content length: ' . strlen($post->post_content));
3630	4404	//error_log('[MXCHAT-WP-IMPORT-DEBUG] Final content length: ' . strlen($content));
		@@ -3631,29 +4405,19 @@
3631	4405	//error_log('[MXCHAT-WP-IMPORT-DEBUG] Content preview: ' . substr($content, 0, 300));
3632	4406
3633	4407	// Note: Removed 10,000 char limit - chunking now handles large content properly
3634	4408
3635		- // Get bot-specific API key
	4409	+ // Get bot-specific embedding decision — custom-provider-aware (plan cbd5fd)
3636	4410	$bot_options = $this->get_bot_options($bot_id);
3637	4411	$options = !empty($bot_options) ? $bot_options : get_option('mxchat_options');
3638		- $selected_model = $options['embedding_model'] ?? 'text-embedding-ada-002';
3639		-
3640		- if (strpos($selected_model, 'voyage') === 0) {
3641		- $api_key = $options['voyage_api_key'] ?? '';
3642		- $provider_name = 'Voyage AI';
3643		- } elseif (strpos($selected_model, 'gemini-embedding') === 0) {
3644		- $api_key = $options['gemini_api_key'] ?? '';
3645		- $provider_name = 'Google Gemini';
3646		- } else {
3647		- $api_key = $options['api_key'] ?? '';
3648		- $provider_name = 'OpenAI';
3649		- }
3650		-
3651		- if (empty($api_key)) {
3652		- MxChat_Admin::mxchat_log_debug('api_error', $provider_name . ' API key not configured for knowledge processing');
3653		- wp_send_json_error($provider_name . ' API key not configured');
	4412	+
	4413	+ $preflight = MxChat_Utils::embedding_preflight($options);
	4414	+ if (!$preflight['ok']) {
	4415	+ MxChat_Admin::mxchat_log_debug('api_error', $preflight['reason'] . ' (knowledge processing)');
	4416	+ wp_send_json_error($preflight['reason']);
3654	4417	exit;
3655	4418	}
	4419	+ $api_key = $preflight['api_key'];
3656	4420
3657	4421	$source_url = get_permalink($post_id);
3658	4422	$vector_id = md5($source_url); // Vector ID for Pinecone
3659	4423
		@@ -3724,11 +4488,11 @@
3724	4488	// Automatically apply role restriction based on tags
3725	4489	$this->apply_role_restriction_to_post($post_id, $source_url);
3726	4490
3727	4491	$operation_type = $is_update ? 'update' : 'new';
3728		-
	4492	+
3729	4493	// Count ACF fields for debugging
3730		- $acf_field_count = count($acf_fields);
	4494	+ $acf_field_count = $prepared['acf_fields_found'];
3731	4495
3732	4496	// Success response with minimal data
3733	4497	wp_send_json_success(array(
3734	4498	'message' => $operation_type === 'update' ? 'Content updated successfully' : 'Content processed successfully',
		@@ -3736,8 +4500,9 @@
3736	4500	'title' => $post->post_title,
3737	4501	'operation_type' => $operation_type,
3738	4502	'vector_id' => $vector_id,
3739	4503	'acf_fields_found' => $acf_field_count,
	4504	+ 'pdf_extracted_count' => (int) $pdf_extracted_count,
3740	4505	'content_preview' => substr($content, 0, 100) . '...',
3741	4506	'bot_id' => $bot_id
3742	4507	));
3743	4508	exit;
		@@ -3807,9 +4572,9 @@
3807	4572	);
3808	4573	} else {
3809	4574	// Update WordPress DB
3810	4575	$table_name = $wpdb->prefix . 'mxchat_system_prompt_content';
3811		-
	4576	+
3812	4577	$wpdb->update(
3813	4578	$table_name,
3814	4579	array('role_restriction' => $highest_role),
3815	4580	array('source_url' => $source_url),
		@@ -3816,8 +4581,15 @@
3816	4581	array('%s'),
3817	4582	array('%s')
3818	4583	);
3819	4584	}
	4585	+
	4586	+ // The entry's restriction just changed — keep the OpenAI Vector Store
	4587	+ // mirror consistent: non-public pulls the file (file_search has no
	4588	+ // per-role filtering), public re-mirrors it (plan 15b5c6).
	4589	+ if (class_exists('MxChat_Vectorstore_Manager')) {
	4590	+ MxChat_Vectorstore_Manager::handle_role_change($source_url, 'default', $highest_role);
	4591	+ }
3820	4592	}
3821	4593
3822	4594	public function mxchat_get_public_post_types() {
3823	4595	// Get all public post types
		@@ -3888,64 +4660,63 @@
3888	4660
3889	4661	return $pinecone_data;
3890	4662	}
3891	4663	public function mxchat_fetch_pinecone_vectors_by_ids($pinecone_options, $vector_ids) {
3892		- //error_log('=== DEBUG: Starting mxchat_fetch_pinecone_vectors_by_ids ===');
3893		-
3894	4664	$api_key = $pinecone_options['mxchat_pinecone_api_key'] ?? '';
3895	4665	$host = $pinecone_options['mxchat_pinecone_host'] ?? '';
	4666	+ $namespace = $pinecone_options['mxchat_pinecone_namespace'] ?? '';
3896	4667
3897	4668	if (empty($api_key) || empty($host) || empty($vector_ids)) {
3898		- //error_log('DEBUG: Missing parameters for fetch by IDs');
3899	4669	return array();
3900	4670	}
3901	4671
3902	4672	try {
3903		- $fetch_url = "https://{$host}/vectors/fetch";
3904		- //error_log('DEBUG: Fetch URL: ' . $fetch_url);
3905		- //error_log('DEBUG: Fetching ' . count($vector_ids) . ' vector IDs');
	4673	+ // NOTE (plan 793b82): /vectors/fetch is a GET endpoint with the ids
	4674	+ // repeated in the query string (ids=a&ids=b — http_build_query would
	4675	+ // emit ids[0]=a); the old POST here was answered 200-with-an-empty-body,
	4676	+ // which read as "nothing indexed". Chunked at 100 ids to stay well
	4677	+ // under the measured HTTP 414 URL-length boundary.
	4678	+ $vectors = array();
	4679	+ foreach (array_chunk(array_values($vector_ids), 100) as $chunk) {
	4680	+ $fetch_query = array();
	4681	+ foreach ($chunk as $fetch_vid) {
	4682	+ $fetch_query[] = 'ids=' . rawurlencode($fetch_vid);
	4683	+ }
	4684	+ if (!empty($namespace)) {
	4685	+ $fetch_query[] = 'namespace=' . rawurlencode($namespace);
	4686	+ }
3906	4687
3907		- // Pinecone fetch API allows fetching specific vectors by ID
3908		- $fetch_data = array(
3909		- 'ids' => array_values($vector_ids)
3910		- );
	4688	+ $response = wp_remote_get("https://{$host}/vectors/fetch?" . implode('&', $fetch_query), array(
	4689	+ 'headers' => array(
	4690	+ 'Api-Key' => $api_key,
	4691	+ 'accept' => 'application/json'
	4692	+ ),
	4693	+ 'timeout' => 30
	4694	+ ));
3911	4695
3912		- $response = wp_remote_post($fetch_url, array(
3913		- 'headers' => array(
3914		- 'Api-Key' => $api_key,
3915		- 'Content-Type' => 'application/json'
3916		- ),
3917		- 'body' => json_encode($fetch_data),
3918		- 'timeout' => 30
3919		- ));
	4696	+ if (is_wp_error($response)) {
	4697	+ error_log('MxChat Pinecone: mxchat_fetch_pinecone_vectors_by_ids GET failed: ' . $response->get_error_message());
	4698	+ continue;
	4699	+ }
3920	4700
3921		- if (is_wp_error($response)) {
3922		- //error_log('DEBUG: Fetch by IDs WP error: ' . $response->get_error_message());
3923		- return array();
3924		- }
	4701	+ if (wp_remote_retrieve_response_code($response) !== 200) {
	4702	+ error_log('MxChat Pinecone: mxchat_fetch_pinecone_vectors_by_ids GET returned HTTP ' . wp_remote_retrieve_response_code($response));
	4703	+ continue;
	4704	+ }
3925	4705
3926		- $response_code = wp_remote_retrieve_response_code($response);
3927		- //error_log('DEBUG: Fetch response code: ' . $response_code);
3928		-
3929		- if ($response_code !== 200) {
3930		- $error_body = wp_remote_retrieve_body($response);
3931		- //error_log('DEBUG: Fetch failed with body: ' . $error_body);
3932		- return array();
	4706	+ $data = json_decode(wp_remote_retrieve_body($response), true);
	4707	+ if (isset($data['vectors']) && is_array($data['vectors'])) {
	4708	+ $vectors += $data['vectors'];
	4709	+ }
3933	4710	}
3934	4711
3935		- $body = wp_remote_retrieve_body($response);
3936		- $data = json_decode($body, true);
3937		-
3938		- //error_log('DEBUG: Fetch response structure: ' . print_r(array_keys($data), true));
3939		-
3940		- if (!isset($data['vectors'])) {
3941		- //error_log('DEBUG: No vectors key in response');
	4712	+ if (empty($vectors)) {
3942	4713	return array();
3943	4714	}
3944	4715
3945	4716	$processed_data = array();
3946	4717
3947		- foreach ($data['vectors'] as $vector_id => $vector_data) {
	4718	+ foreach ($vectors as $vector_id => $vector_data) {
3948	4719	$metadata = $vector_data['metadata'] ?? array();
3949	4720	$source_url = $metadata['source_url'] ?? '';
3950	4721
3951	4722	if (!empty($source_url)) {
		@@ -4016,8 +4787,11 @@
4016	4787	*/
4017	4788	public function mxchat_scan_pinecone_for_processed_content($pinecone_options) {
4018	4789	$api_key = $pinecone_options['mxchat_pinecone_api_key'] ?? '';
4019	4790	$host = $pinecone_options['mxchat_pinecone_host'] ?? '';
	4791	+ // plan 793b82: this scan was namespace-blind — on a namespaced setup it
	4792	+ // surveyed the default namespace and reported the wrong content as indexed.
	4793	+ $namespace = $pinecone_options['mxchat_pinecone_namespace'] ?? '';
4020	4794
4021	4795	if (empty($api_key) || empty($host)) {
4022	4796	return array();
4023	4797	}
		@@ -4052,8 +4826,12 @@
4052	4826	'topK' => 10000,
4053	4827	'vector' => $random_vector
4054	4828	);
4055	4829
	4830	+ if (!empty($namespace)) {
	4831	+ $query_data['namespace'] = $namespace;
	4832	+ }
	4833	+
4056	4834	$response = wp_remote_post($query_url, array(
4057	4835	'headers' => array(
4058	4836	'Api-Key' => $api_key,
4059	4837	'Content-Type' => 'application/json'
		@@ -4066,9 +4844,9 @@
4066	4844	continue;
4067	4845	}
4068	4846
4069	4847	$response_code = wp_remote_retrieve_response_code($response);
4070		-
	4848	+
4071	4849	if ($response_code !== 200) {
4072	4850	continue;
4073	4851	}
4074	4852
		@@ -4152,9 +4930,20 @@
4152	4930
4153	4931	// Get bot-specific options
4154	4932	$bot_options = $this->get_bot_options($bot_id);
4155	4933	$options = !empty($bot_options) ? $bot_options : get_option('mxchat_options');
4156		-
	4934	+
	4935	+ // Opt-in: when the custom provider is selected for embeddings, index through
	4936	+ // the same custom endpoint the query path uses so stored vectors and query
	4937	+ // vectors share a model. Returns the vector array on success, or an error
	4938	+ // string on failure (this function's existing failure contract).
	4939	+ if (isset($options['custom_provider_for_embeddings']) && $options['custom_provider_for_embeddings'] === 'on') {
	4940	+ if (!class_exists('MxChat_Utils')) {
	4941	+ require_once dirname(__FILE__) . '/../includes/class-mxchat-utils.php';
	4942	+ }
	4943	+ return MxChat_Utils::generate_embedding_custom($text, $options);
	4944	+ }
	4945	+
4157	4946	$selected_model = $options['embedding_model'] ?? 'text-embedding-ada-002';
4158	4947	//error_log('[MXCHAT-EMBED] Selected embedding model for bot ' . $bot_id . ': ' . $selected_model);
4159	4948
4160	4949	// Determine provider and endpoint
		@@ -4266,13 +5055,23 @@
4266	5055	$error_message = $error_json['error']['message'] ?? 'No message';
4267	5056	//error_log('[MXCHAT-EMBED] API Error Type: ' . $error_type);
4268	5057	//error_log('[MXCHAT-EMBED] API Error Message: ' . $error_message);
4269	5058
4270		- // Customize error message for common API errors
4271		- if ($error_type === 'invalid_request_error' && strpos($error_message, 'API key') !== false) {
4272		- $error_message = sprintf('Invalid %s API key for bot %s. Please check your API key in the bot settings.', $provider_name, $bot_id);
4273		- } elseif ($error_type === 'authentication_error') {
4274		- $error_message = sprintf('%s authentication failed for bot %s. Please verify your API key in the bot settings.', $provider_name, $bot_id);
	5059	+ // Keep the provider's own diagnostic — a restricted-key 401 names the
	5060	+ // exact missing scope, and replacing it with "check your API key" sent
	5061	+ // a customer to regenerate two keys (plan 46b596). Same shape as
	5062	+ // MxChat_Utils::embedding_failure_error() so both ingestion paths read
	5063	+ // identically. Key never appears in provider messages, but scrub anyway.
	5064	+ if ($error_type === 'invalid_request_error' || $error_type === 'authentication_error') {
	5065	+ if (is_string($api_key) && $api_key !== '') {
	5066	+ $error_message = str_replace($api_key, '[redacted]', $error_message);
	5067	+ }
	5068	+ $error_message = sprintf(
	5069	+ 'Embedding failed (%s, HTTP %d): %s',
	5070	+ $selected_model,
	5071	+ $http_code,
	5072	+ substr($error_message, 0, 300)
	5073	+ );
4275	5074	}
4276	5075
4277	5076	//error_log('[MXCHAT-EMBED] Returning error: ' . $error_message);
4278	5077	return $error_message;
		@@ -4784,25 +5583,38 @@
4784	5583	/**
4785	5584	* Get all ACF fields for a specific post, excluding any fields the user has disabled
4786	5585	*/
4787	5586	public function mxchat_get_acf_fields_for_post($post_id) {
4788		- if (!function_exists('get_fields')) {
	5587	+ if (!function_exists('get_field_objects')) {
4789	5588	return array();
4790	5589	}
4791	5590
4792		- $fields = get_fields($post_id);
4793		- if (!$fields || !is_array($fields)) {
	5591	+ // Field OBJECTS, not get_fields(): exclusion matches on the field KEY
	5592	+ // (unique per field) rather than the name (shared across groups — plan
	5593	+ // 30e81f). ACF's own get_fields() is implemented as get_field_objects()
	5594	+ // reduced to name => value, so the un-excluded reduction below is the
	5595	+ // identical shape and order the previous get_fields() call produced.
	5596	+ $field_objects = get_field_objects($post_id);
	5597	+ if (!$field_objects || !is_array($field_objects)) {
4794	5598	return array();
4795	5599	}
4796	5600
4797		- // Get excluded fields from settings
4798	5601	$excluded_fields = get_option('mxchat_acf_excluded_fields', array());
4799		- if (!empty($excluded_fields) && is_array($excluded_fields)) {
4800		- foreach ($excluded_fields as $excluded_field) {
4801		- if (isset($fields[$excluded_field])) {
4802		- unset($fields[$excluded_field]);
	5602	+ if (!is_array($excluded_fields)) {
	5603	+ $excluded_fields = array();
	5604	+ }
	5605	+
	5606	+ $fields = array();
	5607	+ foreach ($field_objects as $field_name => $field_object) {
	5608	+ if (!empty($excluded_fields)) {
	5609	+ $field_key = isset($field_object['key']) ? $field_object['key'] : '';
	5610	+ // Legacy name entries stay honored: a stored name whose group was
	5611	+ // inactive at migration time still excludes every field wearing it.
	5612	+ if (in_array($field_key, $excluded_fields, true) || in_array($field_name, $excluded_fields, true)) {
	5613	+ continue;
4803	5614	}
4804	5615	}
	5616	+ $fields[$field_name] = isset($field_object['value']) ? $field_object['value'] : null;
4805	5617	}
4806	5618
4807	5619	return $fields;
4808	5620	}
		@@ -4814,8 +5626,11 @@
4814	5626	if (!function_exists('acf_get_field_groups') || !function_exists('acf_get_fields')) {
4815	5627	return array();
4816	5628	}
4817	5629
	5630	+ // Keyed by GROUP KEY, not title (titles are not unique), and each field
	5631	+ // entry carries its ACF field key — the unique identifier every toggle,
	5632	+ // save, and index-time exclusion now runs on (plans 30e81f / bf57e0).
4818	5633	$all_fields = array();
4819	5634	$field_groups = acf_get_field_groups();
4820	5635
4821	5636	if (!empty($field_groups)) {
		@@ -4821,16 +5636,21 @@
4821	5636	if (!empty($field_groups)) {
4822	5637	foreach ($field_groups as $group) {
4823	5638	$group_fields = acf_get_fields($group['key']);
4824	5639	if (!empty($group_fields)) {
4825		- $all_fields[$group['title']] = array();
	5640	+ $entry = array(
	5641	+ 'title' => $group['title'],
	5642	+ 'fields' => array(),
	5643	+ );
4826	5644	foreach ($group_fields as $field) {
4827		- $all_fields[$group['title']][] = array(
4828		- 'name' => $field['name'],
	5645	+ $entry['fields'][] = array(
	5646	+ 'key' => $field['key'],
	5647	+ 'name' => $field['name'],
4829	5648	'label' => $field['label'],
4830		- 'type' => $field['type']
	5649	+ 'type' => $field['type']
4831	5650	);
4832	5651	}
	5652	+ $all_fields[$group['key']] = $entry;
4833	5653	}
4834	5654	}
4835	5655	}
4836	5656
		@@ -5072,8 +5892,200 @@
5072	5892	return implode(', ', array_filter($text_parts));
5073	5893	}
5074	5894
5075	5895	/**
	5896	+ * Walk an ACF field value tree and collect attachment IDs for any value that
	5897	+ * resolves to a PDF in the WordPress media library. Handles the three shapes
	5898	+ * ACF returns for File/Image/URL fields (array with ID+url, integer attachment ID,
	5899	+ * plain URL string), and recurses through repeater/group/flexible content.
	5900	+ *
	5901	+ * @param mixed $value The ACF field value (any depth)
	5902	+ * @param array $out Accumulator (passed by reference) for attachment IDs
	5903	+ * @param int $depth Recursion guard
	5904	+ */
	5905	+private function mxchat_collect_pdf_attachment_ids_from_acf_value($value, &$out, $depth = 0) {
	5906	+ if ($depth > 6) {
	5907	+ return; // prevent runaway recursion on circular/very-deep structures
	5908	+ }
	5909	+
	5910	+ if (empty($value)) {
	5911	+ return;
	5912	+ }
	5913	+
	5914	+ // Array shapes: ACF File/Image return value=array; repeaters/groups are arrays of arrays
	5915	+ if (is_array($value)) {
	5916	+ // Direct File/Image-style array (has 'url' and usually 'ID' + 'mime_type')
	5917	+ $looks_like_attachment = isset($value['url']) || isset($value['ID']) || isset($value['id']);
	5918	+ if ($looks_like_attachment) {
	5919	+ $att_id = 0;
	5920	+ if (!empty($value['ID']) && is_numeric($value['ID'])) {
	5921	+ $att_id = (int) $value['ID'];
	5922	+ } elseif (!empty($value['id']) && is_numeric($value['id'])) {
	5923	+ $att_id = (int) $value['id'];
	5924	+ } elseif (!empty($value['url']) && is_string($value['url'])) {
	5925	+ $att_id = (int) attachment_url_to_postid($value['url']);
	5926	+ }
	5927	+
	5928	+ $is_pdf = false;
	5929	+ if (!empty($value['mime_type']) && $value['mime_type'] === 'application/pdf') {
	5930	+ $is_pdf = true;
	5931	+ } elseif (!empty($value['subtype']) && strtolower((string) $value['subtype']) === 'pdf') {
	5932	+ $is_pdf = true;
	5933	+ } elseif (!empty($value['url']) && is_string($value['url']) && $this->mxchat_url_looks_like_pdf($value['url'])) {
	5934	+ $is_pdf = true;
	5935	+ } elseif ($att_id && get_post_mime_type($att_id) === 'application/pdf') {
	5936	+ $is_pdf = true;
	5937	+ }
	5938	+
	5939	+ if ($is_pdf && $att_id && get_post_mime_type($att_id) === 'application/pdf') {
	5940	+ $out[] = $att_id;
	5941	+ }
	5942	+ // An array node that represents one attachment doesn't contain other
	5943	+ // attachments inside it — done with this branch.
	5944	+ return;
	5945	+ }
	5946	+
	5947	+ // Recurse: repeater rows, flexible-content layouts, groups, etc.
	5948	+ foreach ($value as $sub) {
	5949	+ $this->mxchat_collect_pdf_attachment_ids_from_acf_value($sub, $out, $depth + 1);
	5950	+ }
	5951	+ return;
	5952	+ }
	5953	+
	5954	+ // Plain numeric attachment ID (ACF File field set to "Return: ID")
	5955	+ if (is_numeric($value)) {
	5956	+ $att_id = (int) $value;
	5957	+ if ($att_id > 0 && get_post_mime_type($att_id) === 'application/pdf') {
	5958	+ $out[] = $att_id;
	5959	+ }
	5960	+ return;
	5961	+ }
	5962	+
	5963	+ // Plain string — URL pointing at a PDF (ACF File field set to "Return: URL", or a custom URL/text field)
	5964	+ if (is_string($value)) {
	5965	+ $trimmed = trim($value);
	5966	+ if ($trimmed !== '' && $this->mxchat_url_looks_like_pdf($trimmed)) {
	5967	+ $att_id = (int) attachment_url_to_postid($trimmed);
	5968	+ if ($att_id > 0 && get_post_mime_type($att_id) === 'application/pdf') {
	5969	+ $out[] = $att_id;
	5970	+ }
	5971	+ }
	5972	+ return;
	5973	+ }
	5974	+}
	5975	+
	5976	+/**
	5977	+ * Heuristic: does this URL/string look like a PDF reference?
	5978	+ * Tolerates query strings and fragments (#page=2).
	5979	+ */
	5980	+private function mxchat_url_looks_like_pdf($url) {
	5981	+ if (!is_string($url) || $url === '') {
	5982	+ return false;
	5983	+ }
	5984	+ // Strip query + fragment before checking extension
	5985	+ $path = preg_replace('/[?#].*$/', '', $url);
	5986	+ return (bool) preg_match('/\.pdf$/i', $path);
	5987	+}
	5988	+
	5989	+/**
	5990	+ * Extract text from a PDF attachment by ID using the bundled Smalot parser.
	5991	+ * Reads the file directly from disk via get_attached_file (no HTTP fetch).
	5992	+ * Result is cached on the attachment as post_meta keyed by file mtime so we
	5993	+ * only parse the same PDF once unless the file changes on disk.
	5994	+ *
	5995	+ * @param int $attachment_id
	5996	+ * @return string Extracted plain text, or '' on failure.
	5997	+ */
	5998	+private function mxchat_extract_pdf_text_by_attachment_id($attachment_id) {
	5999	+ $attachment_id = (int) $attachment_id;
	6000	+ if ($attachment_id <= 0) {
	6001	+ return '';
	6002	+ }
	6003	+ if (get_post_mime_type($attachment_id) !== 'application/pdf') {
	6004	+ return '';
	6005	+ }
	6006	+
	6007	+ $pdf_path = get_attached_file($attachment_id);
	6008	+ if (empty($pdf_path) || !file_exists($pdf_path) || !is_readable($pdf_path)) {
	6009	+ return '';
	6010	+ }
	6011	+
	6012	+ // Raw-file size cap. Parsing very large PDFs can OOM the request; skip with a log entry
	6013	+ // and let the rest of the ACF content land in the KB. Filterable for users who need it bigger.
	6014	+ $default_max_bytes = 25 * 1024 * 1024;
	6015	+ $max_bytes = (int) apply_filters('mxchat_acf_pdf_max_bytes', $default_max_bytes, $attachment_id, $pdf_path);
	6016	+ if ($max_bytes > 0) {
	6017	+ $file_size = @filesize($pdf_path);
	6018	+ if ($file_size !== false && $file_size > $max_bytes) {
	6019	+ error_log(sprintf(
	6020	+ '[mxchat] ACF PDF skipped (over size cap): attachment %d "%s" %d bytes > cap %d',
	6021	+ $attachment_id,
	6022	+ basename($pdf_path),
	6023	+ $file_size,
	6024	+ $max_bytes
	6025	+ ));
	6026	+ return '';
	6027	+ }
	6028	+ }
	6029	+
	6030	+ $mtime = @filemtime($pdf_path);
	6031	+ $cache_meta_key = '_mxchat_acf_pdf_text_v1';
	6032	+ $cached = get_post_meta($attachment_id, $cache_meta_key, true);
	6033	+ if (is_array($cached) && isset($cached['mtime'], $cached['text']) && (int) $cached['mtime'] === (int) $mtime) {
	6034	+ return (string) $cached['text'];
	6035	+ }
	6036	+
	6037	+ $text = '';
	6038	+ try {
	6039	+ if (function_exists('mxchat_load_pdf_parser')) {
	6040	+ mxchat_load_pdf_parser();
	6041	+ }
	6042	+ if (!class_exists('\\Smalot\\PdfParser\\Parser')) {
	6043	+ return '';
	6044	+ }
	6045	+ $parser = new \Smalot\PdfParser\Parser();
	6046	+ $pdf = $parser->parseFile($pdf_path);
	6047	+ $pages = $pdf->getPages();
	6048	+ $page_texts = array();
	6049	+ $acf_page_num = 0;
	6050	+ foreach ($pages as $page) {
	6051	+ $acf_page_num++;
	6052	+ $page_text = '';
	6053	+ try {
	6054	+ $page_text = $page->getText();
	6055	+ } catch (\Exception $e) {
	6056	+ $page_text = '';
	6057	+ }
	6058	+ if (!empty($page_text)) {
	6059	+ $page_text = MxChat_Utils::normalize_pdf_rtl($page_text, 'acf_pdf attachment ' . $attachment_id . ' page ' . $acf_page_num);
	6060	+ $page_texts[] = $page_text;
	6061	+ }
	6062	+ }
	6063	+ $text = trim(implode("\n\n", $page_texts));
	6064	+ } catch (\Exception $e) {
	6065	+ error_log('[mxchat] ACF PDF extraction failed for attachment ' . $attachment_id . ': ' . $e->getMessage());
	6066	+ return '';
	6067	+ } catch (\Throwable $e) {
	6068	+ error_log('[mxchat] ACF PDF extraction error for attachment ' . $attachment_id . ': ' . $e->getMessage());
	6069	+ return '';
	6070	+ }
	6071	+
	6072	+ // Cap per-PDF text to avoid blowing up the embedding payload on enormous PDFs.
	6073	+ // The chunker downstream will still split this into multiple vectors.
	6074	+ $max_len = (int) apply_filters('mxchat_acf_pdf_text_max_length', 50000);
	6075	+ if ($max_len > 0 && strlen($text) > $max_len) {
	6076	+ $text = substr($text, 0, $max_len);
	6077	+ }
	6078	+
	6079	+ update_post_meta($attachment_id, $cache_meta_key, array(
	6080	+ 'mtime' => (int) $mtime,
	6081	+ 'text' => $text,
	6082	+ ));
	6083	+
	6084	+ return $text;
	6085	+}
	6086	+
	6087	+/**
5076	6088	* Handle ACF save - fires after ACF fields are saved
5077	6089	* This ensures ACF field data is available when syncing to knowledge base
5078	6090	*/
5079	6091	public function mxchat_handle_acf_save($post_id) {
		@@ -5143,8 +6155,13 @@
5143	6155	$this->mxchat_handle_post_update($post_id, $post, true);
5144	6156	}
5145	6157
5146	6158	public function mxchat_handle_post_update($post_id, $post, $update) {
	6159	+ // The in-flight-update marker has done its job the moment post_updated runs; drop it
	6160	+ // before any early return so it can never outlive its own save (a failed $wpdb->update
	6161	+ // inside wp_insert_post returns after pre_post_update but before the transition).
	6162	+ unset($this->pending_post_update[$post_id]);
	6163	+
5147	6164	// Basic validation checks
5148	6165	if (defined('DOING_AUTOSAVE') && DOING_AUTOSAVE || wp_is_post_revision($post_id)) {
5149	6166	return;
5150	6167	}
		@@ -5182,9 +6199,11 @@
5182	6199	if ($previous_status === 'publish' && $post->post_status !== 'publish') {
5183	6200	// Use the stored URL from when it was published, or fall back to current permalink
5184	6201	$source_url = $previous_url ?: get_permalink($post_id);
5185	6202
5186		- if ($source_url) {
	6203	+ // mxchat_handle_status_transition already deleted for this post earlier in this
	6204	+ // request (it fires first inside wp_insert_post); skip the redundant round-trip.
	6205	+ if ($source_url && empty($this->transition_deleted_posts[$post_id])) {
5187	6206	// Chunk-aware deletion (routes to Pinecone or WP DB and removes base + all chunks)
5188	6207	MxChat_Utils::delete_chunks_for_url($source_url, 'default');
5189	6208	}
5190	6209
		@@ -5213,177 +6232,402 @@
5213	6232	set_transient($previous_url_key, $current_url, DAY_IN_SECONDS);
5214	6233	}
5215	6234	}
5216	6235
5217		- // Only process currently published content for adding/updating
	6236	+ // Only process currently published content for adding/updating.
	6237	+ // transition_indexed_posts: mxchat_handle_status_transition's arrival edge may have
	6238	+ // already indexed this post earlier in this request (editor publishes fire
	6239	+ // transition_post_status first, then post_updated) — skip the duplicate embed.
	6240	+ // Consume-once: the flag is cleared when honoured, so a LATER save of the same
	6241	+ // post in one long-running process (WP-CLI scripts, importers) re-indexes normally.
5218	6242	if ($post->post_status === 'publish') {
	6243	+ if (!empty($this->transition_indexed_posts[$post_id])) {
	6244	+ unset($this->transition_indexed_posts[$post_id]);
	6245	+ } else {
	6246	+ $this->mxchat_index_published_post($post_id, $post);
	6247	+ }
	6248	+ }
	6249	+
	6250	+ // Clean up the stored previous status if not used above
	6251	+ if ($previous_status !== 'publish' || $post->post_status === 'publish') {
	6252	+ delete_transient($previous_status_key);
	6253	+ delete_transient($previous_url_key);
	6254	+ }
	6255	+}
	6256	+
	6257	+/**
	6258	+ * Index a published post into the knowledge base: preprocessing filter, content
	6259	+ * assembly (title/excerpt/body), WooCommerce product enrichment, job_listing meta,
	6260	+ * ACF fields (+ optional PDF extraction), whitelisted custom meta, embedding and
	6261	+ * upsert, then tag-based role restriction.
	6262	+ *
	6263	+ * Shared by the post_updated auto-sync path (mxchat_handle_post_update) and the
	6264	+ * transition_post_status arrival edge (mxchat_handle_status_transition), so
	6265	+ * scheduled publishes (wp_publish_post) and direct status=publish inserts index
	6266	+ * identically to editor saves (plan 3055e1). Pure extraction of the former
	6267	+ * publish branch — body indentation retained to keep the diff reviewable.
	6268	+ */
	6269	+private function mxchat_index_published_post($post_id, $post) {
	6270	+ $post_type = $post->post_type;
	6271	+
	6272	+ // WooCommerce products are owned by the WC-object assembler (plan a3d60c):
	6273	+ // whenever WooCommerce is active AND the integration is enabled, every
	6274	+ // product save also fires save_post_product, which queues
	6275	+ // mxchat_store_product_embedding on shutdown — and that writer runs LAST,
	6276	+ // overwriting the same md5(permalink) row this path would write. Assembling
	6277	+ // and embedding the product here was pure duplicate spend (measured: two
	6278	+ // embedding calls per product save, second one wins). Skip ONLY under the
	6279	+ // exact conditions the shutdown writer runs — same option read as its own
	6280	+ // gate — because with the integration off (or WooCommerce inactive) this
	6281	+ // path is the sole product indexer and must keep working.
	6282	+ if ($post_type === 'product'
	6283	+ && class_exists('WooCommerce')
	6284	+ && isset($this->options['enable_woocommerce_integration'])
	6285	+ && in_array($this->options['enable_woocommerce_integration'], ['1', 'on'])) {
	6286	+ return;
	6287	+ }
	6288	+
5219	6289	// Get the source URL
5220	6290	$source_url = get_permalink($post_id);
5221		-
5222		- // Get content with proper formatting (matching ajax_mxchat_process_selected_content)
5223		- $title = get_the_title($post_id);
5224		- $content = get_post_field('post_content', $post_id);
5225		- $excerpt = get_post_field('post_excerpt', $post_id);
5226	6291
5227		- // Remove shortcode tags but preserve content inside them
5228		- $content = $this->strip_shortcode_tags_preserve_content($content);
5229		- $excerpt = $this->strip_shortcode_tags_preserve_content($excerpt);
	6292	+ // A draft published programmatically (wp_publish_post) can reach this
	6293	+ // point with an EMPTY post_name — wp_insert_post skips slug generation
	6294	+ // for draft/pending — and get_permalink() then resolves to the bare
	6295	+ // site root. A knowledge row keyed to the homepage cites the wrong URL
	6296	+ // and answers homepage questions with this post's body, so refuse to
	6297	+ // write it; the post indexes correctly on its next save, once the slug
	6298	+ // exists. The empty-post_name test is what keeps a legitimate static
	6299	+ // front page (which has a slug but a root permalink) indexable.
	6300	+ // (Plan d138c4.)
	6301	+ if ('' === $post->post_name
	6302	+ && untrailingslashit($source_url) === untrailingslashit(home_url())) {
	6303	+ return;
	6304	+ }
5230	6305
5231		- // Strip tags but preserve structure (don't use 'the_content' filter as it may re-add shortcodes)
5232		- $content = wp_strip_all_tags($content);
	6306	+ /**
	6307	+ * Allow developers to modify post data before processing into the knowledge base.
	6308	+ * Same filter and signature as the manual bulk-import path
	6309	+ * (ajax_mxchat_process_selected_content), so a callback registered once covers
	6310	+ * every indexing route. Purely additive — zero behaviour change when unhooked.
	6311	+ * Auto-sync runs under the 'default' bot context, matching the rest of this
	6312	+ * function.
	6313	+ *
	6314	+ * @param WP_Post $post The post about to be indexed.
	6315	+ * @param string $bot_id Bot context ('default' on auto-sync).
	6316	+ */
	6317	+ $post = apply_filters('mxchat_before_process_post', $post, 'default');
	6318	+ if (!($post instanceof WP_Post)) {
	6319	+ $post = get_post($post_id); // defend against a bad callback return
	6320	+ }
5233	6321
5234		- // Combine title, short description (if exists), and content
5235		- $final_content = $title . "\n\n";
	6322	+ // Assemble the indexable text via the shared post-kind assembler (a3d60c),
	6323	+ // reading from the FILTERED post object — not re-fetched by ID, which would
	6324	+ // discard it. Auto-sync reads content/excerpt in its historical
	6325	+ // get_post_field() display context, never appended product custom tabs
	6326	+ // (its product branch is reachable only with the WooCommerce integration
	6327	+ // off), and gates ACF→PDF extraction behind its own opt-in option —
	6328	+ // default OFF, because re-parsing every ACF PDF on every editor save is
	6329	+ // expensive and most sites don't want it (the 25 MB size cap lives in the
	6330	+ // shared extractor either way).
	6331	+ $prepared = $this->mxchat_prepare_post_content_for_indexing($post_id, $post, array(
	6332	+ 'read_display' => true,
	6333	+ 'extract_acf_pdfs' => get_option('mxchat_auto_sync_acf_pdfs', '0') === '1',
	6334	+ 'include_product_tabs' => false,
	6335	+ ));
	6336	+ $final_content = $prepared['content'];
5236	6337
5237		- // Add short description if it exists (WooCommerce products use post_excerpt for short description)
5238		- if (!empty($excerpt)) {
5239		- $final_content .= "Short Description: " . wp_strip_all_tags($excerpt) . "\n\n";
	6338	+ // Embedding decision — custom-provider-aware. Gating on a cloud API key
	6339	+ // here silently killed auto-sync on keyless custom-embeddings sites,
	6340	+ // because generate_embedding() routes custom FIRST and never needs the
	6341	+ // key (plan cbd5fd). Silent-return shape preserved.
	6342	+ $preflight = MxChat_Utils::embedding_preflight(get_option('mxchat_options'));
	6343	+ if (!$preflight['ok']) {
	6344	+ return;
5240	6345	}
	6346	+ $api_key = $preflight['api_key'];
5241	6347
5242		- $final_content .= $content;
	6348	+ // Use the centralized utility function for storage
	6349	+ $result = MxChat_Utils::submit_content_to_db(
	6350	+ $final_content,
	6351	+ $source_url,
	6352	+ $api_key,
	6353	+ md5($source_url) // Vector ID for Pinecone
	6354	+ );
	6355	+
	6356	+ // After successful storage, apply role restriction based on tags
	6357	+ if (!is_wp_error($result)) {
	6358	+ $this->apply_role_restriction_to_post($post_id, $source_url);
	6359	+ }
	6360	+}
5243	6361
5244		- // For WooCommerce products, include pricing and product details
5245		- if ($post_type === 'product' && class_exists('WooCommerce')) {
5246		- $product = wc_get_product($post_id);
	6362	+/**
	6363	+ * Shared post-fields content assembler (plan a3d60c) — the ONE body behind both
	6364	+ * post-kind ingestion paths: manual bulk import (ajax_mxchat_process_selected_content)
	6365	+ * and auto-sync (mxchat_index_published_post). Behavior-preserving extraction; the
	6366	+ * measured per-caller differences ride $args instead of living as drifting copies:
	6367	+ *
	6368	+ * 'read_display' bool Auto-sync historically reads content/excerpt via
	6369	+ * get_post_field() in its default 'display' context
	6370	+ * (the post_content / post_excerpt display filters
	6371	+ * fire); bulk import reads the raw properties. Inert
	6372	+ * on a stock install — preserved per-path, not converged.
	6373	+ * 'extract_acf_pdfs' bool Each caller passes its OWN option (bulk:
	6374	+ * mxchat_acf_pdf_extraction; auto-sync:
	6375	+ * mxchat_auto_sync_acf_pdfs) — the two-option design
	6376	+ * is deliberate (plan 11720c). Gates BOTH the PDF-id
	6377	+ * collection walk and the extraction loop; the ids are
	6378	+ * only ever read inside the extraction branch, so
	6379	+ * gating collection is output-identical on every install.
	6380	+ * 'include_product_tabs' bool The bulk path has always appended yikes_woo custom
	6381	+ * tabs to product content; the auto-sync product branch
	6382	+ * (reachable only with the WooCommerce integration off)
	6383	+ * never did. Preserved per-path — converging it would be
	6384	+ * a behavior change, recorded on the plan instead.
	6385	+ *
	6386	+ * Returns array: 'content' (the assembled indexable text), 'acf_fields_found' and
	6387	+ * 'pdf_extracted_count' (the bulk path reports both in its AJAX response).
	6388	+ */
	6389	+private function mxchat_prepare_post_content_for_indexing($post_id, $post, $args) {
	6390	+ $read_display = !empty($args['read_display']);
	6391	+ $extract_acf_pdfs = !empty($args['extract_acf_pdfs']);
	6392	+ $include_product_tabs = !empty($args['include_product_tabs']);
5247	6393
5248		- if ($product) {
5249		- // Get pricing information
5250		- $regular_price = $product->get_regular_price();
5251		- $sale_price = $product->get_sale_price();
5252		- $price = $product->get_price();
5253		- $sku = $product->get_sku();
	6394	+ // Raw post_title, NOT get_the_title(): the_title applies wptexturize +
	6395	+ // convert_chars and prepends the "Protected:" / "Private:" display chrome.
	6396	+ // The knowledge base stores facts, not display strings. Entity decode at
	6397	+ // output time (single-pass, shared helper) — a stored `&amp;` embeds worse
	6398	+ // than `&` and gets quoted back to visitors (d2c92e).
	6399	+ $content = $this->mxchat_decode_entities_for_indexing($post->post_title) . "\n\n";
5254	6400
5255		- // Get currency symbol
5256		- $currency_symbol = get_woocommerce_currency_symbol();
	6401	+ $raw_excerpt = $read_display ? get_post_field('post_excerpt', $post) : $post->post_excerpt;
	6402	+ $raw_content = $read_display ? get_post_field('post_content', $post) : $post->post_content;
5257	6403
5258		- // Add pricing information
5259		- $final_content .= "\n";
5260		- if (!empty($regular_price)) {
5261		- $final_content .= "Price: " . $currency_symbol . $regular_price . "\n";
5262		- } elseif (!empty($price)) {
5263		- $final_content .= "Price: " . $currency_symbol . $price . "\n";
5264		- }
	6404	+ // Add short description if it exists (WooCommerce products use post_excerpt for short description)
	6405	+ // Strip FIRST, then test: an excerpt that is nothing but shortcodes strips to
	6406	+ // empty, and testing the raw value emitted a bare "Short Description: " label
	6407	+ // with no value after it. trim() only in the TEST — the emitted value is
	6408	+ // untouched, so a populated excerpt is byte-identical to before. A
	6409	+ // whitespace-only excerpt is an empty excerpt and must not produce a labelled
	6410	+ // line with nothing after it.
	6411	+ $clean_excerpt = $this->strip_shortcode_tags_preserve_content($raw_excerpt);
	6412	+ if (trim($clean_excerpt) !== '') {
	6413	+ $content .= "Short Description: " . $this->mxchat_decode_entities_for_indexing(wp_strip_all_tags($clean_excerpt)) . "\n\n";
	6414	+ }
5265	6415
5266		- if (!empty($sale_price) && $sale_price !== $regular_price) {
5267		- $final_content .= "Sale Price: " . $currency_symbol . $sale_price . "\n";
5268		- }
	6416	+ // Main content — remove shortcode tags but preserve content inside them, then
	6417	+ // strip tags (don't use 'the_content' filter as it may re-add shortcodes).
	6418	+ $clean_content = $this->strip_shortcode_tags_preserve_content($raw_content);
	6419	+ $content .= $this->mxchat_decode_entities_for_indexing(wp_strip_all_tags($clean_content));
5269	6420
5270		- // Handle variable products - show price range
5271		- if ($product->is_type('variable')) {
5272		- $min_price = $product->get_variation_price('min');
5273		- $max_price = $product->get_variation_price('max');
5274		- if ($min_price !== $max_price) {
5275		- $final_content .= "Price Range: " . $currency_symbol . $min_price . " - " . $currency_symbol . $max_price . "\n";
5276		- }
5277		- }
	6421	+ // WooCommerce product enrichment (post-fields kind). The WC-object assembler
	6422	+ // (mxchat_prepare_product_content_for_indexing) owns product rows whenever the
	6423	+ // integration is on; this branch serves the bulk import (all configurations)
	6424	+ // and auto-sync with the integration off.
	6425	+ if (get_post_type($post_id) === 'product' && class_exists('WooCommerce')) {
	6426	+ $product = wc_get_product($post_id);
5278	6427
5279		- if (!empty($sku)) {
5280		- $final_content .= "SKU: " . $sku . "\n";
5281		- }
	6428	+ if ($product) {
	6429	+ $content .= "\n";
	6430	+ $content .= $this->mxchat_woo_product_summary_lines($product);
	6431	+ }
5282	6432
5283		- // Get product categories
5284		- $categories = wp_get_post_terms($post_id, 'product_cat', array('fields' => 'names'));
5285		- if (!empty($categories) && !is_wp_error($categories)) {
5286		- $final_content .= "Categories: " . implode(', ', $categories) . "\n";
5287		- }
5288		- }
	6433	+ if ($include_product_tabs) {
	6434	+ $content .= $this->mxchat_woo_custom_tabs_text($post_id);
5289	6435	}
	6436	+ }
5290	6437
5291		- // For custom post types like job_listing, include additional fields
5292		- if ($post_type === 'job_listing') {
5293		- // Add job-specific meta if available
5294		- $job_location = get_post_meta($post_id, '_job_location', true);
5295		- if (!empty($job_location)) {
5296		- $final_content .= "\n\nLocation: " . $job_location;
5297		- }
	6438	+ // For custom post types like job_listing, include additional fields
	6439	+ if (get_post_type($post_id) === 'job_listing') {
	6440	+ // Add job-specific meta if available
	6441	+ $job_location = get_post_meta($post_id, '_job_location', true);
	6442	+ if (!empty($job_location)) {
	6443	+ $content .= "\n\nLocation: " . $job_location;
	6444	+ }
5298	6445
5299		- // Get job type terms
5300		- $job_types = get_the_terms($post_id, 'job_listing_type');
5301		- if (!empty($job_types) && !is_wp_error($job_types)) {
5302		- $types = array();
5303		- foreach ($job_types as $type) {
5304		- $types[] = $type->name;
5305		- }
5306		- $final_content .= "\n\nJob Type: " . implode(', ', $types);
	6446	+ // Get job type terms
	6447	+ $job_types = get_the_terms($post_id, 'job_listing_type');
	6448	+ if (!empty($job_types) && !is_wp_error($job_types)) {
	6449	+ $types = array();
	6450	+ foreach ($job_types as $type) {
	6451	+ $types[] = $type->name;
5307	6452	}
	6453	+ $content .= "\n\nJob Type: " . implode(', ', $types);
	6454	+ }
5308	6455
5309		- // Get company name if available
5310		- $company_name = get_post_meta($post_id, '_company_name', true);
5311		- if (!empty($company_name)) {
5312		- $final_content .= "\n\nCompany: " . $company_name;
5313		- }
	6456	+ // Get company name if available
	6457	+ $company_name = get_post_meta($post_id, '_company_name', true);
	6458	+ if (!empty($company_name)) {
	6459	+ $content .= "\n\nCompany: " . $company_name;
5314	6460	}
	6461	+ }
5315	6462
5316		- // ADD ACF FIELDS SUPPORT (matches ajax_mxchat_process_selected_content behavior)
5317		- $acf_fields = $this->mxchat_get_acf_fields_for_post($post_id);
5318		- if (!empty($acf_fields)) {
5319		- $acf_content_parts = array();
	6463	+ // ADD ACF FIELDS SUPPORT
	6464	+ $acf_fields = $this->mxchat_get_acf_fields_for_post($post_id);
	6465	+ $pdf_extracted_count = 0;
	6466	+ if (!empty($acf_fields)) {
	6467	+ $acf_content_parts = array();
	6468	+ $pdf_attachment_ids = array();
5320	6469
5321		- foreach ($acf_fields as $field_name => $field_value) {
5322		- $formatted_value = $this->mxchat_format_acf_field_value($field_value, $field_name, $post_id);
5323		- if (!empty($formatted_value)) {
5324		- // Convert field name to readable label
5325		- $field_label = ucwords(str_replace(['_', '-'], ' ', $field_name));
5326		- $acf_content_parts[] = $field_label . ": " . $formatted_value;
5327		- }
	6470	+ foreach ($acf_fields as $field_name => $field_value) {
	6471	+ $formatted_value = $this->mxchat_format_acf_field_value($field_value, $field_name, $post_id);
	6472	+
	6473	+ if (!empty($formatted_value)) {
	6474	+ // Both separators: a hyphenated ACF name should read as words.
	6475	+ $field_label = ucwords(str_replace(['_', '-'], ' ', $field_name));
	6476	+ $acf_content_parts[] = $field_label . ": " . $formatted_value;
5328	6477	}
5329	6478
5330		- if (!empty($acf_content_parts)) {
5331		- $final_content .= "\n\n" . implode("\n", $acf_content_parts);
	6479	+ // Walk this field's value tree for any PDF attachment references and
	6480	+ // queue them for extraction — only when this caller's PDF option is on.
	6481	+ if ($extract_acf_pdfs) {
	6482	+ $this->mxchat_collect_pdf_attachment_ids_from_acf_value($field_value, $pdf_attachment_ids);
5332	6483	}
5333	6484	}
5334	6485
5335		- // ADD CUSTOM POST META SUPPORT (whitelisted non-ACF meta fields)
5336		- $custom_meta = $this->mxchat_get_whitelisted_post_meta($post_id);
5337		- if (!empty($custom_meta)) {
5338		- $meta_content_parts = array();
	6486	+ if (!empty($acf_content_parts)) {
	6487	+ $content .= "\n\n" . implode("\n", $acf_content_parts);
	6488	+ }
5339	6489
5340		- foreach ($custom_meta as $meta_key => $meta_value) {
5341		- // Convert meta key to readable label
5342		- $meta_label = ucwords(str_replace(array('_', '-'), ' ', $meta_key));
5343		- $meta_content_parts[] = $meta_label . ": " . $meta_value;
	6490	+ // Extract text from each unique PDF found in ACF fields and append as a labeled section
	6491	+ if ($extract_acf_pdfs && !empty($pdf_attachment_ids)) {
	6492	+ $pdf_attachment_ids = array_unique(array_filter(array_map('intval', $pdf_attachment_ids)));
	6493	+ $pdf_sections = array();
	6494	+ foreach ($pdf_attachment_ids as $att_id) {
	6495	+ $pdf_text = $this->mxchat_extract_pdf_text_by_attachment_id($att_id);
	6496	+ if (!empty($pdf_text)) {
	6497	+ $pdf_title = get_the_title($att_id);
	6498	+ $pdf_url = wp_get_attachment_url($att_id);
	6499	+ $header = 'PDF Attachment';
	6500	+ if (!empty($pdf_title)) {
	6501	+ $header .= ': ' . $pdf_title;
	6502	+ }
	6503	+ if (!empty($pdf_url)) {
	6504	+ $header .= ' (' . $pdf_url . ')';
	6505	+ }
	6506	+ $pdf_sections[] = $header . "\n" . $pdf_text;
	6507	+ $pdf_extracted_count++;
	6508	+ }
5344	6509	}
5345		-
5346		- if (!empty($meta_content_parts)) {
5347		- $final_content .= "\n\n" . implode("\n", $meta_content_parts);
	6510	+ if (!empty($pdf_sections)) {
	6511	+ $content .= "\n\nAttached PDFs:\n" . implode("\n\n", $pdf_sections);
5348	6512	}
5349	6513	}
	6514	+ }
5350	6515
5351		- // Get API key with proper model detection
5352		- $options = get_option('mxchat_options');
5353		- $selected_model = $options['embedding_model'] ?? 'text-embedding-ada-002';
5354		-
5355		- if (strpos($selected_model, 'voyage') === 0) {
5356		- $api_key = $options['voyage_api_key'] ?? '';
5357		- } elseif (strpos($selected_model, 'gemini-embedding') === 0) {
5358		- $api_key = $options['gemini_api_key'] ?? '';
5359		- } else {
5360		- $api_key = $options['api_key'] ?? '';
	6516	+ // ADD CUSTOM POST META SUPPORT (whitelisted non-ACF meta fields)
	6517	+ $custom_meta = $this->mxchat_get_whitelisted_post_meta($post_id);
	6518	+ if (!empty($custom_meta)) {
	6519	+ $meta_content_parts = array();
	6520	+
	6521	+ foreach ($custom_meta as $meta_key => $meta_value) {
	6522	+ // Convert meta key to readable label
	6523	+ $meta_label = ucwords(str_replace(array('_', '-'), ' ', $meta_key));
	6524	+ $meta_content_parts[] = $meta_label . ": " . $meta_value;
5361	6525	}
5362		-
5363		- if (empty($api_key)) {
5364		- return;
	6526	+
	6527	+ if (!empty($meta_content_parts)) {
	6528	+ $content .= "\n\n" . implode("\n", $meta_content_parts);
5365	6529	}
5366		-
5367		- // Use the centralized utility function for storage
5368		- $result = MxChat_Utils::submit_content_to_db(
5369		- $final_content,
5370		- $source_url,
5371		- $api_key,
5372		- md5($source_url) // Vector ID for Pinecone
5373		- );
5374		-
5375		- // After successful storage, apply role restriction based on tags
5376		- if (!is_wp_error($result)) {
5377		- $this->apply_role_restriction_to_post($post_id, $source_url);
	6530	+ }
	6531	+
	6532	+ return array(
	6533	+ 'content' => $content,
	6534	+ 'acf_fields_found' => count($acf_fields),
	6535	+ 'pdf_extracted_count' => $pdf_extracted_count,
	6536	+ );
	6537	+}
	6538	+
	6539	+/**
	6540	+ * Shared WC-object product assembler (plan a3d60c) — the ONE body behind the two
	6541	+ * WooCommerce-object ingestion paths: the auto-sync product writer
	6542	+ * (mxchat_store_product_embedding) and the URL/sitemap product import
	6543	+ * (mxchat_extract_woocommerce_product_content). Assembles from the WC_Product,
	6544	+ * the authoritative source for product rows (scope decision on the plan).
	6545	+ */
	6546	+private function mxchat_prepare_product_content_for_indexing($product) {
	6547	+ $title = $product->get_name();
	6548	+ $description = $product->get_description();
	6549	+ $short_description = $product->get_short_description();
	6550	+
	6551	+ // Format content consistently
	6552	+ $content = $title . "\n\n";
	6553	+
	6554	+ if (!empty($short_description)) {
	6555	+ $content .= "Short Description: " . wp_strip_all_tags($short_description) . "\n\n";
	6556	+ }
	6557	+
	6558	+ if (!empty($description)) {
	6559	+ $content .= wp_strip_all_tags($description) . "\n\n";
	6560	+ }
	6561	+
	6562	+ $content .= $this->mxchat_woo_product_summary_lines($product);
	6563	+ $content .= $this->mxchat_woo_custom_tabs_text($product->get_id());
	6564	+
	6565	+ return $content;
	6566	+}
	6567	+
	6568	+/**
	6569	+ * Pricing + SKU + categories lines for a product — shared by both assembler kinds
	6570	+ * (the post-fields product enrichment and the WC-object assembler).
	6571	+ */
	6572	+private function mxchat_woo_product_summary_lines($product) {
	6573	+ // Add pricing information (base-currency pinned — see mxchat_product_price_lines)
	6574	+ $lines = $this->mxchat_product_price_lines($product);
	6575	+
	6576	+ $sku = $product->get_sku();
	6577	+ if (!empty($sku)) {
	6578	+ $lines .= "SKU: " . $sku . "\n";
	6579	+ }
	6580	+
	6581	+ // Get product categories
	6582	+ $categories = wp_get_post_terms($product->get_id(), 'product_cat', array('fields' => 'names'));
	6583	+ if (!empty($categories) && !is_wp_error($categories)) {
	6584	+ $lines .= "Categories: " . implode(', ', $categories) . "\n";
	6585	+ }
	6586	+
	6587	+ return $lines;
	6588	+}
	6589	+
	6590	+/**
	6591	+ * Custom Product Tabs text (supports "Custom Product Tabs for WooCommerce" by
	6592	+ * Code Parrots) — direct tabs plus applied reusable/saved tabs. The ONE copy of
	6593	+ * the yikes_woo logic; three sites carried byte-identical clones before a3d60c.
	6594	+ */
	6595	+private function mxchat_woo_custom_tabs_text($product_id) {
	6596	+ $text = '';
	6597	+
	6598	+ $custom_tabs = get_post_meta($product_id, 'yikes_woo_products_tabs', true);
	6599	+ if (!empty($custom_tabs) && is_array($custom_tabs)) {
	6600	+ foreach ($custom_tabs as $tab) {
	6601	+ $tab_title = isset($tab['title']) ? $tab['title'] : (isset($tab['tab_title']) ? $tab['tab_title'] : '');
	6602	+ $tab_content = isset($tab['content']) ? $tab['content'] : '';
	6603	+
	6604	+ if (!empty($tab_title) && !empty($tab_content)) {
	6605	+ $text .= "\n" . $tab_title . ": " . wp_strip_all_tags($tab_content) . "\n";
	6606	+ }
5378	6607	}
5379	6608	}
5380		-
5381		- // Clean up the stored previous status if not used above
5382		- if ($previous_status !== 'publish' || $post->post_status === 'publish') {
5383		- delete_transient($previous_status_key);
5384		- delete_transient($previous_url_key);
	6609	+
	6610	+ // Also check for reusable/saved tabs applied to this product
	6611	+ $applied_saved_tabs = get_post_meta($product_id, 'yikes_woo_reusable_products_tabs_applied', true);
	6612	+ if (!empty($applied_saved_tabs) && is_array($applied_saved_tabs)) {
	6613	+ $saved_tabs = get_option('yikes_woo_reusable_products_tabs', array());
	6614	+ if (!empty($saved_tabs) && is_array($saved_tabs)) {
	6615	+ foreach ($applied_saved_tabs as $saved_tab_id) {
	6616	+ if (isset($saved_tabs[$saved_tab_id])) {
	6617	+ $tab = $saved_tabs[$saved_tab_id];
	6618	+ $tab_title = isset($tab['title']) ? $tab['title'] : (isset($tab['tab_title']) ? $tab['tab_title'] : '');
	6619	+ $tab_content = isset($tab['content']) ? $tab['content'] : '';
	6620	+
	6621	+ if (!empty($tab_title) && !empty($tab_content)) {
	6622	+ $text .= "\n" . $tab_title . ": " . wp_strip_all_tags($tab_content) . "\n";
	6623	+ }
	6624	+ }
	6625	+ }
	6626	+ }
5385	6627	}
	6628	+
	6629	+ return $text;
5386	6630	}
5387	6631
5388	6632	/**
5389	6633	* Store the post status and URL before update to detect status transitions
		@@ -5389,8 +6633,12 @@
5389	6633	* Store the post status and URL before update to detect status transitions
5390	6634	* This runs before the post is actually updated in the database
5391	6635	*/
5392	6636	public function mxchat_store_pre_update_status($post_id, $data) {
	6637	+ // Core is inside wp_insert_post's update branch, so a post_updated WILL fire later
	6638	+ // this request and can consume the arrival-edge guard (plan a664f3).
	6639	+ $this->pending_post_update[$post_id] = true;
	6640	+
5393	6641	// Get the current post from database (before update)
5394	6642	$current_post = get_post($post_id);
5395	6643
5396	6644	if ($current_post) {
		@@ -5406,8 +6654,407 @@
5406	6654	}
5407	6655	}
5408	6656	}
5409	6657
	6658	+/**
	6659	+ * Whether auto-sync is enabled for a post type (mirrors the checks used by the
	6660	+ * update/delete handlers; kept as one helper so new call sites cannot drift).
	6661	+ */
	6662	+private function mxchat_is_auto_sync_enabled($post_type) {
	6663	+ if ($post_type === 'post') {
	6664	+ return get_option('mxchat_auto_sync_posts') === '1';
	6665	+ }
	6666	+ if ($post_type === 'page') {
	6667	+ return get_option('mxchat_auto_sync_pages') === '1';
	6668	+ }
	6669	+ return get_option('mxchat_auto_sync_' . $post_type) === '1';
	6670	+}
	6671	+
	6672	+/**
	6673	+ * Remove a post's vectors the moment it leaves 'publish', using the authoritative
	6674	+ * old status core passes to transition_post_status — no transient involved (plan 816fb1).
	6675	+ *
	6676	+ * Covers status changes that never route through wp_update_post (scheduled-expiry
	6677	+ * plugins and others that flip post_status directly and call wp_transition_post_status),
	6678	+ * where neither pre_post_update nor post_updated fires and the old detection missed.
	6679	+ */
	6680	+public function mxchat_handle_status_transition($new_status, $old_status, $post) {
	6681	+ if (!($post instanceof WP_Post) || wp_is_post_revision($post->ID)) {
	6682	+ return;
	6683	+ }
	6684	+
	6685	+ // Arrival edge (plan 3055e1): a post BECOMING published is indexed here, because
	6686	+ // wp_publish_post() — the path scheduled posts take via check_and_publish_future_post —
	6687	+ // and direct wp_insert_post(status=publish) creates never fire post_updated, so the
	6688	+ // auto-sync ADD path alone misses them. Editor publishes also pass through here;
	6689	+ // the transition_indexed_posts guard keeps mxchat_handle_post_update from embedding
	6690	+ // a second time in the same request.
	6691	+ if ($new_status === 'publish' && $old_status !== 'publish') {
	6692	+ if ($this->mxchat_is_auto_sync_enabled($post->post_type)) {
	6693	+ $this->mxchat_index_published_post($post->ID, $post);
	6694	+
	6695	+ // Arm the double-fire guard ONLY when a post_updated is actually coming to
	6696	+ // consume it (plan a664f3). Two publish paths never fire post_updated at all:
	6697	+ // a direct wp_insert_post(status=publish) create, and wp_publish_post() — the
	6698	+ // call check_and_publish_future_post() makes for scheduled posts. Arming the
	6699	+ // guard unconditionally left it set with nothing to consume it, so the NEXT
	6700	+ // update of that post was swallowed entirely: zero embed calls, no knowledge
	6701	+ // -base row, silently. Consume-once on this side too, so a guard can never
	6702	+ // outlive the single save it was armed for.
	6703	+ if (!empty($this->pending_post_update[$post->ID])) {
	6704	+ unset($this->pending_post_update[$post->ID]);
	6705	+ $this->transition_indexed_posts[$post->ID] = true;
	6706	+ }
	6707	+ }
	6708	+ return;
	6709	+ }
	6710	+
	6711	+ // Only the publish -> not-publish edge matters here.
	6712	+ if ($old_status !== 'publish' || $new_status === 'publish') {
	6713	+ return;
	6714	+ }
	6715	+ // Trash is handled by mxchat_handle_post_delete (wp_trash_post) with pre-trash URL
	6716	+ // resolution; skip to avoid a second network round-trip per trash.
	6717	+ if ($new_status === 'trash') {
	6718	+ return;
	6719	+ }
	6720	+ if (!$this->mxchat_is_auto_sync_enabled($post->post_type)) {
	6721	+ return;
	6722	+ }
	6723	+
	6724	+ $urls = array();
	6725	+
	6726	+ // The DB may already hold the new status when this fires, so get_permalink() on the
	6727	+ // live post could build a draft-style URL whose md5 misses the stored vector IDs.
	6728	+ // Reconstruct the published permalink from a clone instead.
	6729	+ $published_clone = clone $post;
	6730	+ $published_clone->post_status = 'publish';
	6731	+ $published_url = get_permalink($published_clone);
	6732	+ if ($published_url) {
	6733	+ $urls[] = $published_url;
	6734	+ }
	6735	+
	6736	+ // Honour the pre-update capture when present (covers a slug change in the same save).
	6737	+ $previous_url = get_transient('mxchat_prev_url_' . $post->ID);
	6738	+ if (!empty($previous_url)) {
	6739	+ $urls[] = $previous_url;
	6740	+ }
	6741	+
	6742	+ foreach (array_unique($urls) as $url) {
	6743	+ MxChat_Utils::delete_chunks_for_url($url, 'default');
	6744	+ }
	6745	+
	6746	+ if (!empty($urls)) {
	6747	+ $this->transition_deleted_posts[$post->ID] = true;
	6748	+ }
	6749	+}
	6750	+
	6751	+/**
	6752	+ * WP-CLI: remove knowledge-base entries left behind by posts that were unpublished,
	6753	+ * trashed, or made private before the transition_post_status handler existed.
	6754	+ *
	6755	+ * Walks every auto-synced post type's non-published posts, reconstructs each one's
	6756	+ * published-era permalink, and deletes its vectors (routes to Pinecone or the WP table).
	6757	+ * Deletion is idempotent, so never-indexed posts are a cheap no-op.
	6758	+ *
	6759	+ * ## OPTIONS
	6760	+ *
	6761	+ * [--dry-run]
	6762	+ * : Report what would be removed without deleting anything.
	6763	+ *
	6764	+ * ## EXAMPLES
	6765	+ *
	6766	+ * wp mxchat prune-unpublished --dry-run
	6767	+ * wp mxchat prune-unpublished
	6768	+ */
	6769	+public function cli_prune_unpublished($args, $assoc_args) {
	6770	+ global $wpdb;
	6771	+ $dry_run = !empty($assoc_args['dry-run']);
	6772	+ $table = $wpdb->prefix . 'mxchat_system_prompt_content';
	6773	+
	6774	+ $candidate_types = array_merge(array('post', 'page'), array_values(get_post_types(array('_builtin' => false), 'names')));
	6775	+ $synced_types = array();
	6776	+ foreach ($candidate_types as $type) {
	6777	+ if ($this->mxchat_is_auto_sync_enabled($type)) {
	6778	+ $synced_types[] = $type;
	6779	+ }
	6780	+ }
	6781	+ if (empty($synced_types)) {
	6782	+ WP_CLI::success('No post types have auto-sync enabled; nothing to prune.');
	6783	+ return;
	6784	+ }
	6785	+
	6786	+ $scanned = 0;
	6787	+ $pruned = 0;
	6788	+ $paged = 1;
	6789	+ do {
	6790	+ $query = new WP_Query(array(
	6791	+ 'post_type' => $synced_types,
	6792	+ 'post_status' => array('draft', 'pending', 'private', 'future', 'trash'),
	6793	+ 'posts_per_page' => 100,
	6794	+ 'paged' => $paged,
	6795	+ 'fields' => 'ids',
	6796	+ ));
	6797	+ foreach ($query->posts as $post_id) {
	6798	+ $post = get_post($post_id);
	6799	+ if (!$post) {
	6800	+ continue;
	6801	+ }
	6802	+ $scanned++;
	6803	+
	6804	+ // Rebuild the permalink the post had while published: publish-status clone,
	6805	+ // with wp_trash_post's __trashed slug suffix stripped for trashed posts.
	6806	+ $clone = clone $post;
	6807	+ $clone->post_status = 'publish';
	6808	+ if (substr($clone->post_name, -9) === '__trashed') {
	6809	+ $clone->post_name = substr($clone->post_name, 0, -9);
	6810	+ }
	6811	+ $url = get_permalink($clone);
	6812	+ if (!$url) {
	6813	+ continue;
	6814	+ }
	6815	+
	6816	+ // Local-table row count is exact in WordPress-DB mode; in Pinecone mode it
	6817	+ // reads 0 but the delete below still routes to Pinecone and is idempotent.
	6818	+ $local_rows = (int) $wpdb->get_var($wpdb->prepare(
	6819	+ "SELECT COUNT(*) FROM {$table} WHERE source_url = %s", $url
	6820	+ ));
	6821	+
	6822	+ if ($dry_run) {
	6823	+ if ($local_rows > 0) {
	6824	+ WP_CLI::log(sprintf('Would remove %d row(s): %s (post %d, %s)', $local_rows, $url, $post_id, $post->post_status));
	6825	+ $pruned += $local_rows;
	6826	+ }
	6827	+ continue;
	6828	+ }
	6829	+
	6830	+ MxChat_Utils::delete_chunks_for_url($url, 'default');
	6831	+ if ($local_rows > 0) {
	6832	+ WP_CLI::log(sprintf('Removed %d row(s): %s (post %d, %s)', $local_rows, $url, $post_id, $post->post_status));
	6833	+ $pruned += $local_rows;
	6834	+ }
	6835	+ }
	6836	+ $more = $paged < $query->max_num_pages;
	6837	+ $paged++;
	6838	+ } while ($more);
	6839	+
	6840	+ WP_CLI::success(sprintf(
	6841	+ '%s %d local knowledge row(s) across %d non-published post(s) scanned.%s',
	6842	+ $dry_run ? 'Would remove' : 'Removed',
	6843	+ $pruned,
	6844	+ $scanned,
	6845	+ ' (Pinecone-mode deletions are not counted locally.)'
	6846	+ ));
	6847	+}
	6848	+
	6849	+/**
	6850	+ * WP-CLI: repair knowledge-base rows whose PDF text was imported in visual
	6851	+ * (reversed) order before the RTL normalizer existed. 32bf9e fixed new
	6852	+ * imports only; this fixes rows already in the table without the customer
	6853	+ * having to re-source and re-upload the original PDFs (plan d1e6f7).
	6854	+ *
	6855	+ * Detection reuses MxChat_Utils::normalize_pdf_rtl() on the stored text: a
	6856	+ * row is a candidate exactly when the normalizer would change it, so the
	6857	+ * import-time heuristic and the repair heuristic can never disagree.
	6858	+ * Repaired rows are RE-EMBEDDED — the stored vector was computed over
	6859	+ * reversed text and is as broken as the text — so a wet run calls the
	6860	+ * embedding provider once per repaired row on the site's API key. Runs
	6861	+ * beyond 25 rows therefore require --yes.
	6862	+ *
	6863	+ * Scope notes:
	6864	+ * - Scans the WordPress knowledge table. Pinecone-mode entries live in
	6865	+ * Pinecone, not this table, and are not scanned; if a scanned row's bot
	6866	+ * ALSO has Pinecone enabled (hybrid drift), the repaired entry is
	6867	+ * re-submitted through the normal import path so the md5-keyed Pinecone
	6868	+ * vector is replaced too.
	6869	+ * - Knowledge rows do not carry a bot id; --bot only selects whose
	6870	+ * embedding configuration (model + key) is used for re-embedding.
	6871	+ * - The mxchat_pdf_rtl_normalize filter is honoured: a site that disabled
	6872	+ * normalization gets detections of zero, not surprise rewrites.
	6873	+ * - The metadata header the PDF importer stores before the text separator
	6874	+ * is preserved byte-identical; only the text segment is repaired.
	6875	+ *
	6876	+ * ## OPTIONS
	6877	+ *
	6878	+ * [--dry-run]
	6879	+ * : List the rows that would be repaired without changing anything.
	6880	+ *
	6881	+ * [--bot=<id>]
	6882	+ * : Embedding configuration to use for re-embedding. Default: default.
	6883	+ *
	6884	+ * [--all-content]
	6885	+ * : Scan every row containing right-to-left text, not just rows with PDF
	6886	+ * provenance (a page anchor in the source URL, or pdf content type).
	6887	+ *
	6888	+ * [--yes]
	6889	+ * : Proceed even when more than 25 rows need re-embedding (API cost gate).
	6890	+ *
	6891	+ * ## EXAMPLES
	6892	+ *
	6893	+ * wp mxchat rtl-repair --dry-run
	6894	+ * wp mxchat rtl-repair
	6895	+ * wp mxchat rtl-repair --all-content --yes
	6896	+ */
	6897	+public function cli_rtl_repair($args, $assoc_args) {
	6898	+ global $wpdb;
	6899	+ $dry_run = !empty($assoc_args['dry-run']);
	6900	+ $all = !empty($assoc_args['all-content']);
	6901	+ $yes = !empty($assoc_args['yes']);
	6902	+ $bot_id = isset($assoc_args['bot']) ? sanitize_key($assoc_args['bot']) : 'default';
	6903	+ $table = $wpdb->prefix . 'mxchat_system_prompt_content';
	6904	+
	6905	+ // Detection pass — no API calls. Walk the table in id batches so a large
	6906	+ // knowledge base never loads at once.
	6907	+ $rtl_re = '/[\x{0590}-\x{05FF}\x{0600}-\x{06FF}\x{0750}-\x{077F}\x{FB50}-\x{FDFF}\x{FE70}-\x{FEFF}]/u';
	6908	+ $candidates = array();
	6909	+ $scanned = 0;
	6910	+ $last_id = 0;
	6911	+ do {
	6912	+ if ($all) {
	6913	+ $rows = $wpdb->get_results($wpdb->prepare(
	6914	+ "SELECT id, article_content, source_url, content_type FROM {$table}
	6915	+ WHERE id > %d ORDER BY id ASC LIMIT 200",
	6916	+ $last_id
	6917	+ ));
	6918	+ } else {
	6919	+ $rows = $wpdb->get_results($wpdb->prepare(
	6920	+ "SELECT id, article_content, source_url, content_type FROM {$table}
	6921	+ WHERE id > %d AND (source_url LIKE %s OR content_type = 'pdf')
	6922	+ ORDER BY id ASC LIMIT 200",
	6923	+ $last_id,
	6924	+ '%' . $wpdb->esc_like('#page=') . '%'
	6925	+ ));
	6926	+ }
	6927	+ foreach ($rows as $row) {
	6928	+ $last_id = (int) $row->id;
	6929	+ $scanned++;
	6930	+ $content = (string) $row->article_content;
	6931	+ if (!preg_match($rtl_re, $content)) {
	6932	+ continue;
	6933	+ }
	6934	+ list($header, $text) = $this->mxchat_rtl_repair_split($content);
	6935	+ $normalized = MxChat_Utils::normalize_pdf_rtl($text, 'rtl-repair row ' . $row->id);
	6936	+ if (is_string($normalized) && $normalized !== $text) {
	6937	+ $candidates[] = array(
	6938	+ 'id' => (int) $row->id,
	6939	+ 'source_url' => (string) $row->source_url,
	6940	+ 'content_type' => (string) $row->content_type,
	6941	+ 'new_content' => $header . $normalized,
	6942	+ );
	6943	+ }
	6944	+ }
	6945	+ } while (count($rows) === 200);
	6946	+
	6947	+ WP_CLI::log(sprintf('Scanned %d row(s); %d stored in reversed (visual) order.', $scanned, count($candidates)));
	6948	+ if (empty($candidates)) {
	6949	+ WP_CLI::success('No reversed RTL rows found — nothing to repair.');
	6950	+ return;
	6951	+ }
	6952	+
	6953	+ foreach ($candidates as $c) {
	6954	+ WP_CLI::log(sprintf('%s row %d %s', $dry_run ? 'Would repair' : 'Will repair', $c['id'], $c['source_url']));
	6955	+ }
	6956	+ if ($dry_run) {
	6957	+ WP_CLI::success(sprintf('Dry run: %d row(s) would be repaired and re-embedded. Run without --dry-run to apply.', count($candidates)));
	6958	+ return;
	6959	+ }
	6960	+
	6961	+ // Cost gate: re-embedding spends the customer's API budget.
	6962	+ WP_CLI::log(sprintf('Re-embedding will call the embedding provider once per row — %d call(s) on this site\'s API key.', count($candidates)));
	6963	+ if (count($candidates) > 25 && !$yes) {
	6964	+ WP_CLI::error(sprintf('%d rows need re-embedding (more than 25). Re-run with --yes to confirm the API cost. No rows were changed.', count($candidates)));
	6965	+ }
	6966	+
	6967	+ $bot_options = $this->get_bot_options($bot_id);
	6968	+ $options = !empty($bot_options) ? $bot_options : get_option('mxchat_options');
	6969	+ $preflight = MxChat_Utils::embedding_preflight($options);
	6970	+ if (!$preflight['ok']) {
	6971	+ WP_CLI::error('Embedding configuration problem: ' . $preflight['reason']);
	6972	+ }
	6973	+ $api_key = $preflight['api_key'];
	6974	+
	6975	+ $pinecone_hybrid = $this->mxchat_rtl_repair_pinecone_enabled($bot_id);
	6976	+ $repaired = 0;
	6977	+ $failed = 0;
	6978	+ foreach ($candidates as $c) {
	6979	+ $vector = MxChat_Utils::regenerate_embedding($c['new_content'], $api_key, $bot_id);
	6980	+ if (!is_array($vector)) {
	6981	+ $failed++;
	6982	+ $reason = is_wp_error($vector) ? $vector->get_error_message() : 'embedding request failed';
	6983	+ // Text and vector must stay consistent: never write repaired text
	6984	+ // beside the stale reversed-text vector.
	6985	+ WP_CLI::warning(sprintf('Row %d NOT repaired — %s. Row left unchanged.', $c['id'], $reason));
	6986	+ continue;
	6987	+ }
	6988	+ $wpdb->update(
	6989	+ $table,
	6990	+ array(
	6991	+ 'article_content' => $c['new_content'],
	6992	+ 'embedding_vector' => maybe_serialize($vector),
	6993	+ ),
	6994	+ array('id' => $c['id']),
	6995	+ array('%s', '%s'),
	6996	+ array('%d')
	6997	+ );
	6998	+ $repaired++;
	6999	+ if (class_exists('MxChat_Admin')) {
	7000	+ MxChat_Admin::mxchat_log_debug('pdf_rtl_repaired', 'Stored KB row restored to logical order and re-embedded', array(
	7001	+ 'row_id' => $c['id'],
	7002	+ 'source_url' => $c['source_url'],
	7003	+ 'bot' => $bot_id,
	7004	+ ));
	7005	+ }
	7006	+ // Hybrid drift: the bot indexes into Pinecone but this row sat in the
	7007	+ // WP table — push the repaired entry through the normal import path so
	7008	+ // the md5(source_url)-keyed Pinecone vector is replaced as well.
	7009	+ if ($pinecone_hybrid) {
	7010	+ MxChat_Utils::submit_content_to_db(
	7011	+ $c['new_content'],
	7012	+ $c['source_url'],
	7013	+ $api_key,
	7014	+ null,
	7015	+ $bot_id,
	7016	+ $c['content_type'] !== '' ? $c['content_type'] : 'pdf'
	7017	+ );
	7018	+ }
	7019	+ }
	7020	+
	7021	+ WP_CLI::success(sprintf('Repaired + re-embedded %d row(s); %d failed; %d scanned.', $repaired, $failed, $scanned));
	7022	+}
	7023	+
	7024	+/**
	7025	+ * Split a stored KB row into (metadata header incl. separator, text segment).
	7026	+ * The PDF importer stores wp_json_encode($metadata) . "\n---\n" . $text —
	7027	+ * repair must touch only the text and keep the header byte-identical.
	7028	+ */
	7029	+private function mxchat_rtl_repair_split($content) {
	7030	+ $sep = "\n---\n";
	7031	+ $pos = strpos($content, $sep);
	7032	+ if ($pos !== false && $pos > 0 && $content[0] === '{') {
	7033	+ $maybe_json = substr($content, 0, $pos);
	7034	+ if (json_decode($maybe_json) !== null) {
	7035	+ return array(substr($content, 0, $pos + strlen($sep)), substr($content, $pos + strlen($sep)));
	7036	+ }
	7037	+ }
	7038	+ return array('', $content);
	7039	+}
	7040	+
	7041	+/**
	7042	+ * Mirror of MxChat_Utils::is_pinecone_enabled_for_bot() (private there) for
	7043	+ * the repair CLI's hybrid-drift check.
	7044	+ */
	7045	+private function mxchat_rtl_repair_pinecone_enabled($bot_id) {
	7046	+ if ($bot_id !== 'default' && class_exists('MxChat_Multi_Bot_Manager')) {
	7047	+ $cfg = apply_filters('mxchat_get_bot_pinecone_config', array(), $bot_id);
	7048	+ if (!empty($cfg)) {
	7049	+ return !empty($cfg['use_pinecone']) && !empty($cfg['api_key']) && !empty($cfg['host']);
	7050	+ }
	7051	+ }
	7052	+ $po = get_option('mxchat_pinecone_addon_options');
	7053	+ return !empty($po['mxchat_use_pinecone']) && $po['mxchat_use_pinecone'] !== '0'
	7054	+ && !empty($po['mxchat_pinecone_api_key']) && !empty($po['mxchat_pinecone_host']);
	7055	+}
	7056	+
5410	7057	public function mxchat_handle_post_delete($post_id) {
5411	7058	// Get post data before it's deleted
5412	7059	$post = get_post($post_id);
5413	7060
		@@ -5507,110 +7154,20 @@
5507	7154
5508	7155	$source_url = get_permalink($product->get_id());
5509	7156	$product_id = $product->get_id();
5510	7157
5511		- // Build product content
5512		- $title = $product->get_name();
5513		- $description = $product->get_description();
5514		- $short_description = $product->get_short_description();
5515		- $regular_price = $product->get_regular_price();
5516		- $sale_price = $product->get_sale_price();
5517		- $price = $product->get_price();
5518		- $sku = $product->get_sku();
	7158	+ // Build product content via the shared WC-object assembler (a3d60c) — this
	7159	+ // writer owns product rows whenever the integration is on.
	7160	+ $content = $this->mxchat_prepare_product_content_for_indexing($product);
5519	7161
5520		- // Get currency symbol
5521		- $currency_symbol = get_woocommerce_currency_symbol();
5522		-
5523		- // Format content consistently
5524		- $content = $title . "\n\n";
5525		-
5526		- if (!empty($short_description)) {
5527		- $content .= "Short Description: " . wp_strip_all_tags($short_description) . "\n\n";
5528		- }
5529		-
5530		- if (!empty($description)) {
5531		- $content .= wp_strip_all_tags($description) . "\n\n";
5532		- }
5533		-
5534		- // Add pricing information
5535		- if (!empty($regular_price)) {
5536		- $content .= "Price: " . $currency_symbol . $regular_price . "\n";
5537		- } elseif (!empty($price)) {
5538		- $content .= "Price: " . $currency_symbol . $price . "\n";
5539		- }
5540		-
5541		- if (!empty($sale_price) && $sale_price !== $regular_price) {
5542		- $content .= "Sale Price: " . $currency_symbol . $sale_price . "\n";
5543		- }
5544		-
5545		- // Handle variable products - show price range
5546		- if ($product->is_type('variable')) {
5547		- $min_price = $product->get_variation_price('min');
5548		- $max_price = $product->get_variation_price('max');
5549		- if ($min_price !== $max_price) {
5550		- $content .= "Price Range: " . $currency_symbol . $min_price . " - " . $currency_symbol . $max_price . "\n";
5551		- }
5552		- }
5553		-
5554		- if (!empty($sku)) {
5555		- $content .= "SKU: " . $sku . "\n";
5556		- }
5557		-
5558		- // Get product categories
5559		- $categories = wp_get_post_terms($product_id, 'product_cat', array('fields' => 'names'));
5560		- if (!empty($categories) && !is_wp_error($categories)) {
5561		- $content .= "Categories: " . implode(', ', $categories) . "\n";
5562		- }
5563		-
5564		- // Get Custom Product Tabs (supports "Custom Product Tabs for WooCommerce" by Code Parrots)
5565		- $custom_tabs = get_post_meta($product_id, 'yikes_woo_products_tabs', true);
5566		- if (!empty($custom_tabs) && is_array($custom_tabs)) {
5567		- foreach ($custom_tabs as $tab) {
5568		- $tab_title = isset($tab['title']) ? $tab['title'] : (isset($tab['tab_title']) ? $tab['tab_title'] : '');
5569		- $tab_content = isset($tab['content']) ? $tab['content'] : '';
5570		-
5571		- if (!empty($tab_title) && !empty($tab_content)) {
5572		- $content .= "\n" . $tab_title . ": " . wp_strip_all_tags($tab_content) . "\n";
5573		- }
5574		- }
5575		- }
5576		-
5577		- // Also check for reusable/saved tabs applied to this product
5578		- $applied_saved_tabs = get_post_meta($product_id, 'yikes_woo_reusable_products_tabs_applied', true);
5579		- if (!empty($applied_saved_tabs) && is_array($applied_saved_tabs)) {
5580		- // Get the saved tabs option
5581		- $saved_tabs = get_option('yikes_woo_reusable_products_tabs', array());
5582		- if (!empty($saved_tabs) && is_array($saved_tabs)) {
5583		- foreach ($applied_saved_tabs as $saved_tab_id) {
5584		- if (isset($saved_tabs[$saved_tab_id])) {
5585		- $tab = $saved_tabs[$saved_tab_id];
5586		- $tab_title = isset($tab['title']) ? $tab['title'] : (isset($tab['tab_title']) ? $tab['tab_title'] : '');
5587		- $tab_content = isset($tab['content']) ? $tab['content'] : '';
5588		-
5589		- if (!empty($tab_title) && !empty($tab_content)) {
5590		- $content .= "\n" . $tab_title . ": " . wp_strip_all_tags($tab_content) . "\n";
5591		- }
5592		- }
5593		- }
5594		- }
5595		- }
5596		-
5597		- // Get API key with proper model detection
5598		- $options = get_option('mxchat_options');
5599		- $selected_model = $options['embedding_model'] ?? 'text-embedding-ada-002';
5600		-
5601		- if (strpos($selected_model, 'voyage') === 0) {
5602		- $api_key = $options['voyage_api_key'] ?? '';
5603		- } elseif (strpos($selected_model, 'gemini-embedding') === 0) {
5604		- $api_key = $options['gemini_api_key'] ?? '';
5605		- } else {
5606		- $api_key = $options['api_key'] ?? '';
5607		- }
5608		-
5609		- if (empty($api_key)) {
5610		- //error_log('MxChat Auto-sync: No API key configured for embedding model');
	7162	+ // Embedding decision — custom-provider-aware (plan cbd5fd); silent-return
	7163	+ // shape preserved.
	7164	+ $preflight = MxChat_Utils::embedding_preflight(get_option('mxchat_options'));
	7165	+ if (!$preflight['ok']) {
	7166	+ //error_log('MxChat Auto-sync: embedding pre-flight failed: ' . $preflight['reason']);
5611	7167	return;
5612	7168	}
	7169	+ $api_key = $preflight['api_key'];
5613	7170
5614	7171	// Use the centralized utility function for storage
5615	7172	$result = MxChat_Utils::submit_content_to_db(
5616	7173	$content,
		@@ -5686,17 +7243,23 @@
5686	7243
5687	7244	// Delete from Pinecone
5688	7245	$pinecone_manager = MxChat_Pinecone_Manager::get_instance();
5689	7246	$result = $pinecone_manager->mxchat_delete_from_pinecone_by_vector_id(
5690		- $vector_id,
5691		- $pinecone_options['mxchat_pinecone_api_key'],
5692		- $pinecone_options['mxchat_pinecone_host']
	7247	+ $vector_id,
	7248	+ $pinecone_options['mxchat_pinecone_api_key'],
	7249	+ $pinecone_options['mxchat_pinecone_host'],
	7250	+ $pinecone_options['mxchat_pinecone_namespace'] ?? ''
5693	7251	);
5694	7252
5695	7253	if ($result['success']) {
	7254	+ // Mirror the removal to the OpenAI Vector Store mapping (plan 15b5c6);
	7255	+ // a chunk vector id reduces to its base entry there.
	7256	+ if (class_exists('MxChat_Vectorstore_Manager')) {
	7257	+ MxChat_Vectorstore_Manager::sync_delete_by_key($vector_id, 'default');
	7258	+ }
5696	7259	// No cache clearing needed since we removed caching
5697		- set_transient('mxchat_admin_notice_success',
5698		- esc_html__('Entry deleted successfully from Pinecone.', 'mxchat'),
	7260	+ set_transient('mxchat_admin_notice_success',
	7261	+ esc_html__('Entry deleted successfully from Pinecone.', 'mxchat'),
5699	7262	30
5700	7263	);
5701	7264	} else {
5702	7265	set_transient('mxchat_admin_notice_error',
		@@ -5741,16 +7304,21 @@
5741	7304	wp_send_json_error('Pinecone is not properly configured for bot: ' . $bot_id);
5742	7305	exit;
5743	7306	}
5744	7307
5745		- // Delete from the correct Pinecone index
	7308	+ // Delete from the correct Pinecone index and namespace
5746	7309	$result = $pinecone_manager->mxchat_delete_from_pinecone_by_vector_id(
5747		- $vector_id,
5748		- $pinecone_options['mxchat_pinecone_api_key'],
5749		- $pinecone_options['mxchat_pinecone_host']
	7310	+ $vector_id,
	7311	+ $pinecone_options['mxchat_pinecone_api_key'],
	7312	+ $pinecone_options['mxchat_pinecone_host'],
	7313	+ $pinecone_options['mxchat_pinecone_namespace'] ?? ''
5750	7314	);
5751	7315
5752	7316	if ($result['success']) {
	7317	+ // Mirror the removal to the OpenAI Vector Store mapping (plan 15b5c6)
	7318	+ if (class_exists('MxChat_Vectorstore_Manager')) {
	7319	+ MxChat_Vectorstore_Manager::sync_delete_by_key($vector_id, $bot_id);
	7320	+ }
5753	7321	// No cache clearing needed since we removed caching
5754	7322	wp_send_json_success(array(
5755	7323	'message' => 'Entry deleted successfully from Pinecone',
5756	7324	'vector_id' => $vector_id,
		@@ -5849,8 +7417,13 @@
5849	7417	}
5850	7418	}
5851	7419
5852	7420	if (empty($vectors_to_delete)) {
	7421	+ // Entry already gone from Pinecone — still clear any mirrored
	7422	+ // Vector Store file so it can't outlive the entry (plan 15b5c6).
	7423	+ if (class_exists('MxChat_Vectorstore_Manager')) {
	7424	+ MxChat_Vectorstore_Manager::sync_delete_entry($source_url, $bot_id);
	7425	+ }
5853	7426	wp_send_json_success(array(
5854	7427	'message' => 'No vectors found to delete',
5855	7428	'source_url' => $source_url
5856	7429	));
		@@ -5891,8 +7464,13 @@
5891	7464	wp_send_json_error('Pinecone API error (HTTP ' . $response_code . ')');
5892	7465	exit;
5893	7466	}
5894	7467
	7468	+ // Mirror the removal to the OpenAI Vector Store (plan 15b5c6)
	7469	+ if (class_exists('MxChat_Vectorstore_Manager')) {
	7470	+ MxChat_Vectorstore_Manager::sync_delete_entry($source_url, $bot_id);
	7471	+ }
	7472	+
5895	7473	wp_send_json_success(array(
5896	7474	'message' => 'All chunks deleted successfully from Pinecone',
5897	7475	'source_url' => $source_url,
5898	7476	'deleted_count' => count($vectors_to_delete)
		@@ -5914,8 +7492,13 @@
5914	7492	wp_send_json_error('Failed to delete from database: ' . $wpdb->last_error);
5915	7493	exit;
5916	7494	}
5917	7495
	7496	+ // Mirror the removal to the OpenAI Vector Store (plan 15b5c6)
	7497	+ if (class_exists('MxChat_Vectorstore_Manager')) {
	7498	+ MxChat_Vectorstore_Manager::sync_delete_entry($source_url, $bot_id);
	7499	+ }
	7500	+
5918	7501	wp_send_json_success(array(
5919	7502	'message' => 'All chunks deleted successfully from database',
5920	7503	'source_url' => $source_url,
5921	7504	'deleted_count' => $result
		@@ -5950,8 +7533,15 @@
5950	7533
5951	7534	global $wpdb;
5952	7535	$table_name = $wpdb->prefix . 'mxchat_system_prompt_content';
5953	7536
	7537	+ // Capture the identity BEFORE the row disappears — needed to mirror the
	7538	+ // change into the Vector Store (plan 15b5c6).
	7539	+ $source_url = $wpdb->get_var($wpdb->prepare(
	7540	+ "SELECT source_url FROM {$table_name} WHERE id = %d",
	7541	+ $entry_id
	7542	+ ));
	7543	+
5954	7544	// Clear cache for this entry
5955	7545	wp_cache_delete('prompt_' . $entry_id, 'mxchat_prompts');
5956	7546
5957	7547	// Delete from database
		@@ -5961,8 +7551,23 @@
5961	7551	array('%d')
5962	7552	);
5963	7553
5964	7554	if ($result !== false) {
	7555	+ // Mirror to the Vector Store: if sibling rows remain (this was one
	7556	+ // chunk of a larger entry) the entry's file is REFRESHED from what's
	7557	+ // left; if none remain, the file is removed.
	7558	+ if (!empty($source_url) && class_exists('MxChat_Vectorstore_Manager')) {
	7559	+ $remaining = (int) $wpdb->get_var($wpdb->prepare(
	7560	+ "SELECT COUNT(*) FROM {$table_name} WHERE source_url = %s",
	7561	+ $source_url
	7562	+ ));
	7563	+ if ($remaining > 0) {
	7564	+ MxChat_Vectorstore_Manager::sync_upsert_entry($source_url, '', 'default');
	7565	+ } else {
	7566	+ MxChat_Vectorstore_Manager::sync_delete_entry($source_url, 'default');
	7567	+ }
	7568	+ }
	7569	+
5965	7570	wp_send_json_success(array(
5966	7571	'message' => 'Entry deleted successfully',
5967	7572	'entry_id' => $entry_id
5968	7573	));
		@@ -6015,8 +7620,9 @@
6015	7620	$use_pinecone = ($pinecone_options['mxchat_use_pinecone'] ?? '0') === '1';
6016	7621
6017	7622	$api_key = $pinecone_options['mxchat_pinecone_api_key'] ?? '';
6018	7623	$host = $pinecone_options['mxchat_pinecone_host'] ?? '';
	7624	+ $namespace = $pinecone_options['mxchat_pinecone_namespace'] ?? '';
6019	7625
6020	7626	// =============================================
6021	7627	// PHASE 1: Collect all Pinecone vector IDs
6022	7628	// and separate WordPress entries
		@@ -6023,8 +7629,10 @@
6023	7629	// =============================================
6024	7630	$pinecone_entry_ids = array(); // entry IDs that are pinecone-sourced
6025	7631	$wordpress_entries = array(); // entries for WordPress DB deletion
6026	7632	$all_vector_ids = array(); // all pinecone vector IDs to delete in one batch
	7633	+ $vs_mirror_urls = array(); // Vector Store mirror: URLs to delete (plan 15b5c6)
	7634	+ $vs_mirror_keys = array(); // Vector Store mirror: bare vector ids to delete
6027	7635
6028	7636	foreach ($entries as $entry) {
6029	7637	$entry_id = sanitize_text_field($entry['id'] ?? '');
6030	7638	$source = sanitize_text_field($entry['source'] ?? 'wordpress');
		@@ -6049,8 +7657,11 @@
6049	7657	$base_vector_id = md5($source_url);
6050	7658	$all_vector_ids[] = $base_vector_id;
6051	7659
6052	7660	$list_url = 'https://' . $host . '/vectors/list?prefix=' . urlencode($base_vector_id . '_chunk_') . '&limit=100';
	7661	+ if (!empty($namespace)) {
	7662	+ $list_url .= '&namespace=' . rawurlencode($namespace);
	7663	+ }
6053	7664	$list_response = wp_remote_get($list_url, array(
6054	7665	'headers' => array(
6055	7666	'Api-Key' => $api_key,
6056	7667	'accept' => 'application/json'
		@@ -6071,8 +7682,14 @@
6071	7682	} else {
6072	7683	// Single entry: the entry_id IS the vector ID
6073	7684	$all_vector_ids[] = $entry_id;
6074	7685	}
	7686	+
	7687	+ if (!empty($source_url)) {
	7688	+ $vs_mirror_urls[] = $source_url;
	7689	+ } else {
	7690	+ $vs_mirror_keys[] = $entry_id;
	7691	+ }
6075	7692	} else {
6076	7693	$wordpress_entries[] = $entry;
6077	7694	}
6078	7695	}
		@@ -6085,8 +7702,12 @@
6085	7702	$pinecone_success = true;
6086	7703	$batches = array_chunk($all_vector_ids, 100);
6087	7704
6088	7705	foreach ($batches as $batch) {
	7706	+ $delete_body = array('ids' => $batch);
	7707	+ if (!empty($namespace)) {
	7708	+ $delete_body['namespace'] = $namespace;
	7709	+ }
6089	7710	$delete_response = wp_remote_post("https://{$host}/vectors/delete", array(
6090	7711	'headers' => array(
6091	7712	'Api-Key' => $api_key,
6092	7713	'accept' => 'application/json',
		@@ -6091,9 +7712,9 @@
6091	7712	'Api-Key' => $api_key,
6092	7713	'accept' => 'application/json',
6093	7714	'content-type' => 'application/json'
6094	7715	),
6095		- 'body' => wp_json_encode(array('ids' => $batch)),
	7716	+ 'body' => wp_json_encode($delete_body),
6096	7717	'timeout' => 60
6097	7718	));
6098	7719
6099	7720	if (is_wp_error($delete_response)) {
		@@ -6118,8 +7739,18 @@
6118	7739	} else {
6119	7740	$failed_ids[] = $eid;
6120	7741	}
6121	7742	}
	7743	+
	7744	+ // Mirror the removals to the OpenAI Vector Store (plan 15b5c6)
	7745	+ if ($pinecone_success && class_exists('MxChat_Vectorstore_Manager')) {
	7746	+ foreach (array_unique($vs_mirror_urls) as $vs_url) {
	7747	+ MxChat_Vectorstore_Manager::sync_delete_entry($vs_url, $bot_id);
	7748	+ }
	7749	+ foreach (array_unique($vs_mirror_keys) as $vs_key) {
	7750	+ MxChat_Vectorstore_Manager::sync_delete_by_key($vs_key, $bot_id);
	7751	+ }
	7752	+ }
6122	7753	}
6123	7754
6124	7755	// =============================================
6125	7756	// PHASE 3: WordPress database deletions
		@@ -6139,9 +7770,15 @@
6139	7770	$table_name,
6140	7771	array('source_url' => $source_url),
6141	7772	array('%s')
6142	7773	);
	7774	+ $row_url = $source_url;
6143	7775	} else {
	7776	+ // Identity captured pre-delete for the Vector Store mirror
	7777	+ $row_url = $wpdb->get_var($wpdb->prepare(
	7778	+ "SELECT source_url FROM {$table_name} WHERE id = %d",
	7779	+ intval($entry_id)
	7780	+ ));
6144	7781	wp_cache_delete('prompt_' . $entry_id, 'mxchat_prompts');
6145	7782	$result = $wpdb->delete(
6146	7783	$table_name,
6147	7784	array('id' => intval($entry_id)),
		@@ -6150,8 +7787,21 @@
6150	7787	}
6151	7788
6152	7789	if ($result !== false) {
6153	7790	$success_ids[] = $entry_id;
	7791	+ // Mirror to the Vector Store: refresh the entry's file when
	7792	+ // sibling chunk rows survive, remove it when none do.
	7793	+ if (!empty($row_url) && class_exists('MxChat_Vectorstore_Manager')) {
	7794	+ $remaining = (int) $wpdb->get_var($wpdb->prepare(
	7795	+ "SELECT COUNT(*) FROM {$table_name} WHERE source_url = %s",
	7796	+ $row_url
	7797	+ ));
	7798	+ if ($remaining > 0) {
	7799	+ MxChat_Vectorstore_Manager::sync_upsert_entry($row_url, '', $bot_id);
	7800	+ } else {
	7801	+ MxChat_Vectorstore_Manager::sync_delete_entry($row_url, $bot_id);
	7802	+ }
	7803	+ }
6154	7804	} else {
6155	7805	$failed_ids[] = $entry_id;
6156	7806	$errors[] = "Database error for entry: $entry_id";
6157	7807	}
		@@ -6303,9 +7953,29 @@
6303	7953	if ($result === false) {
6304	7954	wp_send_json_error('Database update failed: ' . $wpdb->last_error);
6305	7955	exit;
6306	7956	}
6307		-
	7957	+
	7958	+ // Keep the OpenAI Vector Store mirror consistent with the new restriction
	7959	+ // (plan 15b5c6): non-public pulls the mirrored file, public re-mirrors.
	7960	+ if (class_exists('MxChat_Vectorstore_Manager')) {
	7961	+ if ($data_source === 'pinecone') {
	7962	+ if ($role_restriction !== 'public') {
	7963	+ MxChat_Vectorstore_Manager::sync_delete_by_key($entry_id, 'default');
	7964	+ }
	7965	+ // Public again: Pinecone-mode content isn't held locally, so the
	7966	+ // entry re-mirrors on its next save/import rather than here.
	7967	+ } else {
	7968	+ $row_url = $wpdb->get_var($wpdb->prepare(
	7969	+ "SELECT source_url FROM {$wpdb->prefix}mxchat_system_prompt_content WHERE id = %d",
	7970	+ absint($entry_id)
	7971	+ ));
	7972	+ if (!empty($row_url)) {
	7973	+ MxChat_Vectorstore_Manager::handle_role_change($row_url, 'default', $role_restriction);
	7974	+ }
	7975	+ }
	7976	+ }
	7977	+
6308	7978	wp_send_json_success(array(
6309	7979	'message' => 'Role restriction updated successfully',
6310	7980	'role_restriction' => $role_restriction,
6311	7981	'data_source' => $data_source,
		@@ -6348,16 +8018,16 @@
6348	8018	wp_send_json_error('Unauthorized access');
6349	8019	exit;
6350	8020	}
6351	8021
6352		- $tag_slug = isset($_POST['tag_slug']) ? sanitize_text_field($_POST['tag_slug']) : '';
	8022	+ $tag_input = isset($_POST['tag_slug']) ? sanitize_text_field($_POST['tag_slug']) : '';
6353	8023	$role_restriction = isset($_POST['role_restriction']) ? sanitize_text_field($_POST['role_restriction']) : 'public';
6354		-
6355		- if (empty($tag_slug)) {
6356		- wp_send_json_error('Tag slug is required');
	8024	+
	8025	+ if (empty($tag_input)) {
	8026	+ wp_send_json_error('Please enter a tag name or slug');
6357	8027	exit;
6358	8028	}
6359		-
	8029	+
6360	8030	// Validate role restriction
6361	8031	$valid_roles = array_keys($this->mxchat_get_role_options());
6362	8032	if (!in_array($role_restriction, $valid_roles)) {
6363	8033	wp_send_json_error('Invalid role restriction');
		@@ -6362,16 +8032,27 @@
6362	8032	if (!in_array($role_restriction, $valid_roles)) {
6363	8033	wp_send_json_error('Invalid role restriction');
6364	8034	exit;
6365	8035	}
6366		-
6367		- // Check if tag exists in WordPress
6368		- $term = get_term_by('slug', $tag_slug, 'post_tag');
	8036	+
	8037	+ // Resolve the tag by slug first, then fall back to its display name, so users can
	8038	+ // enter either "premium-content" or "Premium Content". (plan b8bcf5 — the field is
	8039	+ // labeled by name but previously validated by slug only, producing the confusing
	8040	+ // "Tag does not exist in WordPress" error when a real tag's name was typed.)
	8041	+ $term = get_term_by('slug', $tag_input, 'post_tag');
6369	8042	if (!$term) {
6370		- wp_send_json_error('Tag does not exist in WordPress');
	8043	+ $term = get_term_by('name', $tag_input, 'post_tag');
	8044	+ }
	8045	+ if (!$term) {
	8046	+ wp_send_json_error('No tag with that name or slug exists yet. Create it under Posts → Tags first, then enter its name or slug.');
6371	8047	exit;
6372	8048	}
6373		-
	8049	+
	8050	+ // Always key the mapping by the RESOLVED slug — apply_role_restriction_to_post()
	8051	+ // compares against each post's tag slugs, so the stored key must be a slug,
	8052	+ // never the raw (possibly display-name) input.
	8053	+ $tag_slug = $term->slug;
	8054	+
6374	8055	// Get existing mappings
6375	8056	$mappings = get_option('mxchat_tag_role_mappings', array());
6376	8057
6377	8058	// Check if mapping already exists
		@@ -6646,9 +8327,9 @@
6646	8327	);
6647	8328	} else {
6648	8329	// Update WordPress DB
6649	8330	$table_name = $wpdb->prefix . 'mxchat_system_prompt_content';
6650		-
	8331	+
6651	8332	$wpdb->update(
6652	8333	$table_name,
6653	8334	array('role_restriction' => $highest_role),
6654	8335	array('source_url' => $source_url),
		@@ -6655,8 +8336,15 @@
6655	8336	array('%s'),
6656	8337	array('%s')
6657	8338	);
6658	8339	}
	8340	+
	8341	+ // The entry's restriction just changed — keep the OpenAI Vector Store
	8342	+ // mirror consistent: non-public pulls the file (file_search has no
	8343	+ // per-role filtering), public re-mirrors it (plan 15b5c6).
	8344	+ if (class_exists('MxChat_Vectorstore_Manager')) {
	8345	+ MxChat_Vectorstore_Manager::handle_role_change($source_url, 'default', $highest_role);
	8346	+ }
6659	8347	}
6660	8348
6661	8349	/**
6662	8350	* Apply role restriction after content is stored (for auto-sync)
		@@ -6725,9 +8413,9 @@
6725	8413	);
6726	8414	} else {
6727	8415	// Update WordPress DB
6728	8416	$table_name = $wpdb->prefix . 'mxchat_system_prompt_content';
6729		-
	8417	+
6730	8418	$wpdb->update(
6731	8419	$table_name,
6732	8420	array('role_restriction' => $highest_role),
6733	8421	array('source_url' => $source_url),
		@@ -6734,8 +8422,15 @@
6734	8422	array('%s'),
6735	8423	array('%s')
6736	8424	);
6737	8425	}
	8426	+
	8427	+ // The entry's restriction just changed — keep the OpenAI Vector Store
	8428	+ // mirror consistent: non-public pulls the file (file_search has no
	8429	+ // per-role filtering), public re-mirrors it (plan 15b5c6).
	8430	+ if (class_exists('MxChat_Vectorstore_Manager')) {
	8431	+ MxChat_Vectorstore_Manager::handle_role_change($source_url, 'default', $highest_role);
	8432	+ }
6738	8433	}
6739	8434
6740	8435
6741	8436	// ========================================
		@@ -7196,25 +8891,19 @@
7196	8891	if (empty($url)) {
7197	8892	return new WP_Error('invalid_url', 'URL is empty');
7198	8893	}
7199	8894
7200		- // Get bot-specific API key early (needed for both paths)
	8895	+ // Get bot-specific embedding decision early (needed for both paths) —
	8896	+ // custom-provider-aware (plan cbd5fd). Error code preserved.
7201	8897	$bot_options = $this->get_bot_options($bot_id);
7202	8898	$options = !empty($bot_options) ? $bot_options : get_option('mxchat_options');
7203		- $selected_model = $options['embedding_model'] ?? 'text-embedding-ada-002';
7204	8899
7205		- if (strpos($selected_model, 'voyage') === 0) {
7206		- $api_key = $options['voyage_api_key'] ?? '';
7207		- } elseif (strpos($selected_model, 'gemini-embedding') === 0) {
7208		- $api_key = $options['gemini_api_key'] ?? '';
7209		- } else {
7210		- $api_key = $options['api_key'] ?? '';
	8900	+ $preflight = MxChat_Utils::embedding_preflight($options);
	8901	+ if (!$preflight['ok']) {
	8902	+ return new WP_Error('no_api_key', $preflight['reason'] . ' (bot: ' . $bot_id . ')');
7211	8903	}
	8904	+ $api_key = $preflight['api_key'];
7212	8905
7213		- if (empty($api_key)) {
7214		- return new WP_Error('no_api_key', 'API key not configured for bot: ' . $bot_id);
7215		- }
7216		-
7217	8906	// Check if this is a WooCommerce product URL and WooCommerce is active
7218	8907	$is_product_url = (strpos($url, '/product/') !== false || strpos($url, '/shop/') !== false);
7219	8908	$content_type = $is_product_url ? 'product' : 'url';
7220	8909
		@@ -7241,9 +8930,9 @@
7241	8930	$is_likely_pdf = (strtolower(pathinfo(parse_url($url, PHP_URL_PATH) ?: '', PATHINFO_EXTENSION)) === 'pdf');
7242	8931	$response = wp_remote_get($url, array(
7243	8932	'timeout' => $is_likely_pdf ? 120 : 30,
7244	8933	'redirection' => 5,
7245		- 'user-agent' => 'MxChat/1.0'
	8934	+ 'user-agent' => mxchat_ingest_user_agent(),
7246	8935	));
7247	8936
7248	8937	if (is_wp_error($response)) {
7249	8938	return $response;
		@@ -7408,8 +9097,9 @@
7408	9097
7409	9098	for ($i = 0; $i < $total_pages; $i++) {
7410	9099	$page_num = $i + 1;
7411	9100	$text = $pages[$i]->getText();
	9101	+ $text = MxChat_Utils::normalize_pdf_rtl($text, 'kb_pdf_import page ' . $page_num);
7412	9102	if (empty($text)) {
7413	9103	$skipped_pages[] = 'Page ' . $page_num . ': No text could be extracted — page may contain only images, links, or non-standard encoding';
7414	9104	continue;
7415	9105	}
		@@ -7496,95 +9186,12 @@
7496	9186	if (!$product) {
7497	9187	return false;
7498	9188	}
7499	9189
7500		- // Build product content with pricing (similar to mxchat_store_product_embedding)
7501		- $title = $product->get_name();
7502		- $description = $product->get_description();
7503		- $short_description = $product->get_short_description();
7504		- $sku = $product->get_sku();
	9190	+ // Build product content via the shared WC-object assembler (a3d60c) — same
	9191	+ // body as the auto-sync product writer, so the two paths can never drift.
	9192	+ $content = $this->mxchat_prepare_product_content_for_indexing($product);
7505	9193
7506		- // Get pricing information
7507		- $regular_price = $product->get_regular_price();
7508		- $sale_price = $product->get_sale_price();
7509		- $price = $product->get_price(); // Current active price
7510		-
7511		- // Get currency symbol
7512		- $currency_symbol = get_woocommerce_currency_symbol();
7513		-
7514		- // Format content
7515		- $content = $title . "\n\n";
7516		-
7517		- if (!empty($short_description)) {
7518		- $content .= "Short Description: " . wp_strip_all_tags($short_description) . "\n\n";
7519		- }
7520		-
7521		- if (!empty($description)) {
7522		- $content .= wp_strip_all_tags($description) . "\n\n";
7523		- }
7524		-
7525		- // Add pricing information
7526		- if (!empty($regular_price)) {
7527		- $content .= "Price: " . $currency_symbol . $regular_price . "\n";
7528		- } elseif (!empty($price)) {
7529		- $content .= "Price: " . $currency_symbol . $price . "\n";
7530		- }
7531		-
7532		- if (!empty($sale_price) && $sale_price !== $regular_price) {
7533		- $content .= "Sale Price: " . $currency_symbol . $sale_price . "\n";
7534		- }
7535		-
7536		- // Handle variable products - show price range
7537		- if ($product->is_type('variable')) {
7538		- $min_price = $product->get_variation_price('min');
7539		- $max_price = $product->get_variation_price('max');
7540		- if ($min_price !== $max_price) {
7541		- $content .= "Price Range: " . $currency_symbol . $min_price . " - " . $currency_symbol . $max_price . "\n";
7542		- }
7543		- }
7544		-
7545		- if (!empty($sku)) {
7546		- $content .= "SKU: " . $sku . "\n";
7547		- }
7548		-
7549		- // Get product categories
7550		- $categories = wp_get_post_terms($product_id, 'product_cat', array('fields' => 'names'));
7551		- if (!empty($categories) && !is_wp_error($categories)) {
7552		- $content .= "Categories: " . implode(', ', $categories) . "\n";
7553		- }
7554		-
7555		- // Get Custom Product Tabs (supports "Custom Product Tabs for WooCommerce" by Code Parrots)
7556		- $custom_tabs = get_post_meta($product_id, 'yikes_woo_products_tabs', true);
7557		- if (!empty($custom_tabs) && is_array($custom_tabs)) {
7558		- foreach ($custom_tabs as $tab) {
7559		- $tab_title = isset($tab['title']) ? $tab['title'] : (isset($tab['tab_title']) ? $tab['tab_title'] : '');
7560		- $tab_content = isset($tab['content']) ? $tab['content'] : '';
7561		-
7562		- if (!empty($tab_title) && !empty($tab_content)) {
7563		- $content .= "\n" . $tab_title . ": " . wp_strip_all_tags($tab_content) . "\n";
7564		- }
7565		- }
7566		- }
7567		-
7568		- // Also check for reusable/saved tabs applied to this product
7569		- $applied_saved_tabs = get_post_meta($product_id, 'yikes_woo_reusable_products_tabs_applied', true);
7570		- if (!empty($applied_saved_tabs) && is_array($applied_saved_tabs)) {
7571		- $saved_tabs = get_option('yikes_woo_reusable_products_tabs', array());
7572		- if (!empty($saved_tabs) && is_array($saved_tabs)) {
7573		- foreach ($applied_saved_tabs as $saved_tab_id) {
7574		- if (isset($saved_tabs[$saved_tab_id])) {
7575		- $tab = $saved_tabs[$saved_tab_id];
7576		- $tab_title = isset($tab['title']) ? $tab['title'] : (isset($tab['tab_title']) ? $tab['tab_title'] : '');
7577		- $tab_content = isset($tab['content']) ? $tab['content'] : '';
7578		-
7579		- if (!empty($tab_title) && !empty($tab_content)) {
7580		- $content .= "\n" . $tab_title . ": " . wp_strip_all_tags($tab_content) . "\n";
7581		- }
7582		- }
7583		- }
7584		- }
7585		- }
7586		-
7587	9194	return $this->mxchat_sanitize_content_for_api($content);
7588	9195	}
7589	9196
7590	9197	/**
		@@ -7614,8 +9221,9 @@
7614	9221	return new WP_Error('page_not_found', 'Page ' . $page_number . ' not found in PDF');
7615	9222	}
7616	9223
7617	9224	$text = $pages[$page_number - 1]->getText();
	9225	+ $text = MxChat_Utils::normalize_pdf_rtl($text, 'kb_pdf_page page ' . $page_number);
7618	9226
7619	9227	if (empty($text)) {
7620	9228	return new WP_Error('empty_page', 'Page ' . $page_number . ': No text could be extracted — page may contain only images, links, or non-standard encoding');
7621	9229	}
		@@ -7636,24 +9244,18 @@
7636	9244
7637	9245	$content_with_metadata = wp_json_encode($metadata) . "\n---\n" . $sanitized;
7638	9246	$page_url = esc_url($pdf_url . "#page=" . $page_number);
7639	9247
7640		- // Get bot-specific API key
	9248	+ // Get bot-specific embedding decision — custom-provider-aware
	9249	+ // (plan cbd5fd). Error code preserved.
7641	9250	$bot_options = $this->get_bot_options($bot_id);
7642	9251	$options = !empty($bot_options) ? $bot_options : get_option('mxchat_options');
7643		- $selected_model = $options['embedding_model'] ?? 'text-embedding-ada-002';
7644		-
7645		- if (strpos($selected_model, 'voyage') === 0) {
7646		- $api_key = $options['voyage_api_key'] ?? '';
7647		- } elseif (strpos($selected_model, 'gemini-embedding') === 0) {
7648		- $api_key = $options['gemini_api_key'] ?? '';
7649		- } else {
7650		- $api_key = $options['api_key'] ?? '';
	9252	+
	9253	+ $preflight = MxChat_Utils::embedding_preflight($options);
	9254	+ if (!$preflight['ok']) {
	9255	+ return new WP_Error('no_api_key', $preflight['reason'] . ' (bot: ' . $bot_id . ')');
7651	9256	}
7652		-
7653		- if (empty($api_key)) {
7654		- return new WP_Error('no_api_key', 'API key not configured for bot: ' . $bot_id);
7655		- }
	9257	+ $api_key = $preflight['api_key'];
7656	9258
7657	9259	// Submit to database - UPDATED 2.5.6: Added content_type 'pdf'
7658	9260	$result = MxChat_Utils::submit_content_to_db(
7659	9261	$content_with_metadata,