← All changes | includes/class-mxchat-utils.php +2 -75 3.2.7 → 3.2.2 View file →

		@@ -5,51 +5,8 @@
5	5
6	6	class MxChat_Utils {
7	7
8	8	/**
9		- * Centralized embedding model registry. Single source of truth for dimensions
10		- * and provider, so model-switch protection logic doesn't drift across files.
11		- */
12		-public static function embedding_model_registry() {
13		- return array(
14		- 'text-embedding-ada-002' => array('dims' => 1536, 'provider' => 'openai', 'label' => 'Ada 2'),
15		- 'text-embedding-3-small' => array('dims' => 1536, 'provider' => 'openai', 'label' => 'TE3 Small'),
16		- 'text-embedding-3-large' => array('dims' => 3072, 'provider' => 'openai', 'label' => 'TE3 Large'),
17		- 'voyage-3-large' => array('dims' => 2048, 'provider' => 'voyage', 'label' => 'Voyage-3 Large'),
18		- 'gemini-embedding-001' => array('dims' => 1536, 'provider' => 'gemini', 'label' => 'Gemini Embedding'),
19		- );
20		-}
21		-
22		-public static function embedding_model_dimensions($model) {
23		- $registry = self::embedding_model_registry();
24		- return isset($registry[$model]) ? (int) $registry[$model]['dims'] : 0;
25		-}
26		-
27		-public static function embedding_model_label($model) {
28		- $registry = self::embedding_model_registry();
29		- return isset($registry[$model]) ? $registry[$model]['label'] : $model;
30		-}
31		-
32		-/**
33		- * Returns the model that was last used to actually write embeddings into the
34		- * KB. Differs from the user-selected setting once a switch has happened but
35		- * no re-embed has occurred yet — that's the mismatch state we warn about.
36		- */
37		-public static function get_active_embedding_model() {
38		- return get_option('mxchat_active_embedding_model', '');
39		-}
40		-
41		-/**
42		- * Stamp the model that produced the most recent successful embedding. Called
43		- * from generate_embedding() right after the API responds with a valid vector.
44		- */
45		-public static function stamp_active_embedding_model($model) {
46		- if (!empty($model) && $model !== self::get_active_embedding_model()) {
47		- update_option('mxchat_active_embedding_model', $model, false);
48		- }
49		-}
50		-
51		-/**
52	9	* UPDATED: Submit or update content (and its embedding) in the database.
53	10	* Stores in Pinecone if enabled, otherwise stores in WordPress DB.
54	11	*
55	12	* @param string $content The content to be embedded.
		@@ -414,9 +371,9 @@
414	371	'source_url' => $url, // Can be empty for manual content
415	372	'type' => $content_type, // Now supports: post, page, pdf, url, manual, product, etc.
416	373	'last_updated' => time(),
417	374	'created_at' => time(), // Add creation timestamp
418		- 'bot_id' => $bot_id, // Add bot identification
	375	+ 'bot_id' => $bot_id // Add bot identification
419	376	);
420	377
421	378	$vector_data = array(
422	379	'id' => $vector_id,
		@@ -564,9 +521,8 @@
564	521	// Handle different response formats based on provider
565	522	if (strpos($selected_model, 'gemini-embedding') === 0) {
566	523	// Gemini API response format
567	524	if (isset($response_body['embedding']['values']) && is_array($response_body['embedding']['values'])) {
568		- self::stamp_active_embedding_model($selected_model);
569	525	return $response_body['embedding']['values'];
570	526	} else {
571	527	//error_log('Invalid response received from Gemini embedding API for bot ' . $bot_id . ': ' . wp_json_encode($response_body));
572	528	return null;
		@@ -573,9 +529,8 @@
573	529	}
574	530	} else {
575	531	// OpenAI/Voyage API response format
576	532	if (isset($response_body['data'][0]['embedding']) && is_array($response_body['data'][0]['embedding'])) {
577		- self::stamp_active_embedding_model($selected_model);
578	533	return $response_body['data'][0]['embedding'];
579	534	} else {
580	535	//error_log('Invalid response received from embedding API for bot ' . $bot_id . ': ' . wp_json_encode($response_body));
581	536	return null;
		@@ -631,36 +586,8 @@
631	586
632	587	foreach ($chunks as $index => $chunk_text) {
633	588	// Generate chunk metadata
634	589	$chunk_metadata = MxChat_Chunker::create_chunk_metadata($index, $total_chunks, $source_url);
635		-
636		- // AI-Engine-style aliases so external consumers (Pinecone/Qdrant/Chroma) can rely on
637		- // a stable shorthand ('source'/'part_index'/'part_total') without parsing our internal names.
638		- $chunk_metadata['source'] = $source_url;
639		- $chunk_metadata['part_index'] = (int) $index;
640		- $chunk_metadata['part_total'] = (int) $total_chunks;
641		-
642		- /**
643		- * Filter the per-chunk metadata blob before it's written to the KB store.
644		- *
645		- * @param array $chunk_metadata Metadata array (source, part_index, part_total, chunk_index, total_chunks, source_url, parent_url_hash, document_type, ...).
646		- * @param string $chunk_text The chunk text being stored.
647		- * @param array $context ['bot_id' => string, 'content_type' => string, 'source_url' => string, 'part_index' => int, 'part_total' => int]
648		- * @return array Updated metadata array.
649		- */
650		- $chunk_metadata = apply_filters(
651		- 'mxchat_embedding_chunk_metadata',
652		- $chunk_metadata,
653		- $chunk_text,
654		- array(
655		- 'bot_id' => $bot_id,
656		- 'content_type' => $content_type,
657		- 'source_url' => $source_url,
658		- 'part_index' => (int) $index,
659		- 'part_total' => (int) $total_chunks,
660		- )
661		- );
662		-
663	590	$chunk_vector_id = MxChat_Chunker::generate_chunk_vector_id($source_url, $index);
664	591
665	592	//error_log('[MXCHAT-CHUNK] Processing chunk ' . ($index + 1) . '/' . $total_chunks . ' (ID: ' . $chunk_vector_id . ')');
666	593
		@@ -759,9 +686,9 @@
759	686	'total_chunks' => $chunk_metadata['total_chunks'],
760	687	'parent_url_hash' => $chunk_metadata['parent_url_hash'],
761	688	'last_updated' => time(),
762	689	'created_at' => time(),
763		- 'bot_id' => $bot_id,
	690	+ 'bot_id' => $bot_id
764	691	);
765	692
766	693	$vector_data = array(
767	694	'id' => $vector_id,