| @@ -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. |
| @@ -224,11 +181,9 @@ | ||
| 224 | 181 | // Check if this is truly manual content (no URL at all) vs a real URL that filter_var rejects |
| 225 | 182 | // filter_var(FILTER_VALIDATE_URL) rejects valid URLs with encoded chars, non-ASCII, fragments, etc. |
| 226 | 183 | // Use a looser check: if it starts with http(s):// or has a scheme, it's a URL |
| 227 | 184 | $has_url_scheme = !empty($source_url) && preg_match('#^https?://#i', $source_url); |
| 228 | - // Treat legacy mxchat.ai source URLs as manual — old bug assigned the site URL to manual entries | |
| 229 | - $is_legacy_mxchat_url = $has_url_scheme && strpos($source_url, 'mxchat.ai') !== false; | |
| 230 | - $is_manual_content = empty($source_url) || $source_url === '' || !$has_url_scheme || $is_legacy_mxchat_url; | |
| 185 | + $is_manual_content = empty($source_url) || $source_url === '' || !$has_url_scheme; | |
| 231 | 186 | |
| 232 | 187 | if ($is_manual_content) { |
| 233 | 188 | // Generate unique identifier for manual content to prevent overwrites |
| 234 | 189 | $source_url = 'mxchat://manual-content/' . time() . '-' . wp_generate_password(8, false); |
| @@ -414,9 +369,9 @@ | ||
| 414 | 369 | 'source_url' => $url, // Can be empty for manual content |
| 415 | 370 | 'type' => $content_type, // Now supports: post, page, pdf, url, manual, product, etc. |
| 416 | 371 | 'last_updated' => time(), |
| 417 | 372 | 'created_at' => time(), // Add creation timestamp |
| 418 | - 'bot_id' => $bot_id, // Add bot identification | |
| 373 | + 'bot_id' => $bot_id // Add bot identification | |
| 419 | 374 | ); |
| 420 | 375 | |
| 421 | 376 | $vector_data = array( |
| 422 | 377 | 'id' => $vector_id, |
| @@ -488,20 +443,11 @@ | ||
| 488 | 443 | } else { |
| 489 | 444 | $bot_options = apply_filters('mxchat_get_bot_options', array(), $bot_id); |
| 490 | 445 | $options = !empty($bot_options) ? $bot_options : get_option('mxchat_options'); |
| 491 | 446 | } |
| 492 | - | |
| 493 | - // Opt-in: when the custom provider is selected for embeddings, route the KB | |
| 494 | - // INDEX side through the same custom endpoint the query side uses, so stored | |
| 495 | - // vectors and query vectors come from the same model. Default-off behavior | |
| 496 | - // below is untouched. | |
| 497 | - if (isset($options['custom_provider_for_embeddings']) && $options['custom_provider_for_embeddings'] === 'on') { | |
| 498 | - $custom = self::generate_embedding_custom($text, $options); | |
| 499 | - return is_array($custom) ? $custom : null; | |
| 500 | - } | |
| 501 | - | |
| 447 | + | |
| 502 | 448 | $selected_model = $options['embedding_model'] ?? 'text-embedding-ada-002'; |
| 503 | - | |
| 449 | + | |
| 504 | 450 | // Determine endpoint and API key based on model |
| 505 | 451 | if (strpos($selected_model, 'voyage') === 0) { |
| 506 | 452 | $endpoint = 'https://api.voyageai.com/v1/embeddings'; |
| 507 | 453 | $api_key = $options['voyage_api_key'] ?? ''; |
| @@ -573,9 +519,8 @@ | ||
| 573 | 519 | // Handle different response formats based on provider |
| 574 | 520 | if (strpos($selected_model, 'gemini-embedding') === 0) { |
| 575 | 521 | // Gemini API response format |
| 576 | 522 | if (isset($response_body['embedding']['values']) && is_array($response_body['embedding']['values'])) { |
| 577 | - self::stamp_active_embedding_model($selected_model); | |
| 578 | 523 | return $response_body['embedding']['values']; |
| 579 | 524 | } else { |
| 580 | 525 | //error_log('Invalid response received from Gemini embedding API for bot ' . $bot_id . ': ' . wp_json_encode($response_body)); |
| 581 | 526 | return null; |
| @@ -582,9 +527,8 @@ | ||
| 582 | 527 | } |
| 583 | 528 | } else { |
| 584 | 529 | // OpenAI/Voyage API response format |
| 585 | 530 | if (isset($response_body['data'][0]['embedding']) && is_array($response_body['data'][0]['embedding'])) { |
| 586 | - self::stamp_active_embedding_model($selected_model); | |
| 587 | 531 | return $response_body['data'][0]['embedding']; |
| 588 | 532 | } else { |
| 589 | 533 | //error_log('Invalid response received from embedding API for bot ' . $bot_id . ': ' . wp_json_encode($response_body)); |
| 590 | 534 | return null; |
| @@ -592,81 +536,8 @@ | ||
| 592 | 536 | } |
| 593 | 537 | } |
| 594 | 538 | |
| 595 | 539 | /** |
| 596 | - * Generate an embedding via a Custom (OpenAI-compatible) provider's /embeddings route. | |
| 597 | - * Shared by every embedding entry point so the KNOWLEDGE-BASE INDEX side and the | |
| 598 | - * QUERY side route through the same model when the opt-in | |
| 599 | - * 'custom_provider_for_embeddings' setting is on. Mirrors the query-path logic in | |
| 600 | - * MxChat_Integrator::mxchat_generate_embedding_custom() but takes an explicit | |
| 601 | - * $options array so it is callable statically from utils + knowledge-manager. | |
| 602 | - * | |
| 603 | - * Returns a numeric array (the embedding vector) on success, or a human-readable | |
| 604 | - * error string on failure (so callers expecting a string error, like the | |
| 605 | - * knowledge-manager, can surface it directly; callers expecting array|null wrap it). | |
| 606 | - * | |
| 607 | - * @param string $text Text to embed. | |
| 608 | - * @param array $options The resolved mxchat options (must contain the custom_provider_* keys). | |
| 609 | - * @return array|string Embedding vector on success; error string on failure. | |
| 610 | - */ | |
| 611 | -public static function generate_embedding_custom($text, $options) { | |
| 612 | - if (empty($text)) { | |
| 613 | - return 'No text provided for embedding generation'; | |
| 614 | - } | |
| 615 | - | |
| 616 | - $base_url = isset($options['custom_provider_base_url']) ? rtrim(trim((string) $options['custom_provider_base_url']), '/') : ''; | |
| 617 | - if (empty($base_url)) { | |
| 618 | - return 'Custom provider Base URL is not configured.'; | |
| 619 | - } | |
| 620 | - | |
| 621 | - $api_key = isset($options['custom_provider_api_key']) ? trim((string) $options['custom_provider_api_key']) : ''; | |
| 622 | - $auth_scheme = isset($options['custom_provider_auth_scheme']) ? $options['custom_provider_auth_scheme'] : 'bearer'; | |
| 623 | - $api_version = isset($options['custom_provider_api_version']) ? trim((string) $options['custom_provider_api_version']) : ''; | |
| 624 | - | |
| 625 | - // Embedding model: prefer the dedicated custom_provider_embedding_model, fall back to the chat model. | |
| 626 | - $model = (isset($options['custom_provider_embedding_model']) && trim((string) $options['custom_provider_embedding_model']) !== '') | |
| 627 | - ? trim((string) $options['custom_provider_embedding_model']) | |
| 628 | - : ((isset($options['custom_provider_model']) && trim((string) $options['custom_provider_model']) !== '') ? trim((string) $options['custom_provider_model']) : 'default'); | |
| 629 | - | |
| 630 | - $embed_url = $base_url . '/embeddings'; | |
| 631 | - if (!empty($api_version)) { | |
| 632 | - $embed_url .= (strpos($embed_url, '?') === false ? '?' : '&') . 'api-version=' . rawurlencode($api_version); | |
| 633 | - } | |
| 634 | - | |
| 635 | - $headers = ['Content-Type' => 'application/json']; | |
| 636 | - if (!empty($api_key)) { | |
| 637 | - if ($auth_scheme === 'api-key') { | |
| 638 | - $headers['api-key'] = $api_key; | |
| 639 | - } else { | |
| 640 | - $headers['Authorization'] = 'Bearer ' . $api_key; | |
| 641 | - } | |
| 642 | - } | |
| 643 | - | |
| 644 | - $response = wp_remote_post($embed_url, [ | |
| 645 | - 'headers' => $headers, | |
| 646 | - 'body' => wp_json_encode(['input' => $text, 'model' => $model]), | |
| 647 | - 'timeout' => 60, | |
| 648 | - ]); | |
| 649 | - if (is_wp_error($response)) { | |
| 650 | - return 'Connection error when generating embeddings (custom provider): ' . $response->get_error_message(); | |
| 651 | - } | |
| 652 | - | |
| 653 | - $status = wp_remote_retrieve_response_code($response); | |
| 654 | - $body = json_decode(wp_remote_retrieve_body($response), true); | |
| 655 | - if ($status !== 200) { | |
| 656 | - $msg = isset($body['error']['message']) ? $body['error']['message'] : 'HTTP ' . $status; | |
| 657 | - return 'Custom embedding endpoint error: ' . $msg; | |
| 658 | - } | |
| 659 | - if (isset($body['data'][0]['embedding']) && is_array($body['data'][0]['embedding'])) { | |
| 660 | - // Stamp the custom model identity so the active-embedding-model mismatch | |
| 661 | - // warning reflects the real (custom) model rather than the built-in setting. | |
| 662 | - self::stamp_active_embedding_model('custom:' . $model); | |
| 663 | - return $body['data'][0]['embedding']; | |
| 664 | - } | |
| 665 | - return 'Invalid embedding response from custom provider.'; | |
| 666 | -} | |
| 667 | - | |
| 668 | -/** | |
| 669 | 540 | * Submit content as multiple chunks |
| 670 | 541 | * |
| 671 | 542 | * Splits large content into chunks, generates embeddings for each, |
| 672 | 543 | * and stores them with chunk metadata for later reassembly. |
| @@ -713,36 +584,8 @@ | ||
| 713 | 584 | |
| 714 | 585 | foreach ($chunks as $index => $chunk_text) { |
| 715 | 586 | // Generate chunk metadata |
| 716 | 587 | $chunk_metadata = MxChat_Chunker::create_chunk_metadata($index, $total_chunks, $source_url); |
| 717 | - | |
| 718 | - // AI-Engine-style aliases so external consumers (Pinecone/Qdrant/Chroma) can rely on | |
| 719 | - // a stable shorthand ('source'/'part_index'/'part_total') without parsing our internal names. | |
| 720 | - $chunk_metadata['source'] = $source_url; | |
| 721 | - $chunk_metadata['part_index'] = (int) $index; | |
| 722 | - $chunk_metadata['part_total'] = (int) $total_chunks; | |
| 723 | - | |
| 724 | - /** | |
| 725 | - * Filter the per-chunk metadata blob before it's written to the KB store. | |
| 726 | - * | |
| 727 | - * @param array $chunk_metadata Metadata array (source, part_index, part_total, chunk_index, total_chunks, source_url, parent_url_hash, document_type, ...). | |
| 728 | - * @param string $chunk_text The chunk text being stored. | |
| 729 | - * @param array $context ['bot_id' => string, 'content_type' => string, 'source_url' => string, 'part_index' => int, 'part_total' => int] | |
| 730 | - * @return array Updated metadata array. | |
| 731 | - */ | |
| 732 | - $chunk_metadata = apply_filters( | |
| 733 | - 'mxchat_embedding_chunk_metadata', | |
| 734 | - $chunk_metadata, | |
| 735 | - $chunk_text, | |
| 736 | - array( | |
| 737 | - 'bot_id' => $bot_id, | |
| 738 | - 'content_type' => $content_type, | |
| 739 | - 'source_url' => $source_url, | |
| 740 | - 'part_index' => (int) $index, | |
| 741 | - 'part_total' => (int) $total_chunks, | |
| 742 | - ) | |
| 743 | - ); | |
| 744 | - | |
| 745 | 588 | $chunk_vector_id = MxChat_Chunker::generate_chunk_vector_id($source_url, $index); |
| 746 | 589 | |
| 747 | 590 | //error_log('[MXCHAT-CHUNK] Processing chunk ' . ($index + 1) . '/' . $total_chunks . ' (ID: ' . $chunk_vector_id . ')'); |
| 748 | 591 | |
| @@ -841,9 +684,9 @@ | ||
| 841 | 684 | 'total_chunks' => $chunk_metadata['total_chunks'], |
| 842 | 685 | 'parent_url_hash' => $chunk_metadata['parent_url_hash'], |
| 843 | 686 | 'last_updated' => time(), |
| 844 | 687 | 'created_at' => time(), |
| 845 | - 'bot_id' => $bot_id, | |
| 688 | + 'bot_id' => $bot_id | |
| 846 | 689 | ); |
| 847 | 690 | |
| 848 | 691 | $vector_data = array( |
| 849 | 692 | 'id' => $vector_id, |
| @@ -959,34 +802,31 @@ | ||
| 959 | 802 | |
| 960 | 803 | // Add the original single-vector ID (for non-chunked content) |
| 961 | 804 | $vectors_to_delete[] = $base_vector_id; |
| 962 | 805 | |
| 963 | - // Pinecone /vectors/list is a GET endpoint with query-string parameters; a POST here returns a | |
| 964 | - // non-200 silently and we end up only deleting the base vector, leaving chunks orphaned. | |
| 965 | - $query_params = array( | |
| 806 | + // Use Pinecone list API to find all chunk vectors with this prefix | |
| 807 | + $list_url = "https://{$host}/vectors/list"; | |
| 808 | + | |
| 809 | + $list_body = array( | |
| 966 | 810 | 'prefix' => $base_vector_id . '_chunk_', |
| 967 | - 'limit' => 100, | |
| 811 | + 'limit' => 100 | |
| 968 | 812 | ); |
| 813 | + | |
| 969 | 814 | if (!empty($namespace)) { |
| 970 | - $query_params['namespace'] = $namespace; | |
| 815 | + $list_body['namespace'] = $namespace; | |
| 971 | 816 | } |
| 972 | 817 | |
| 973 | - $list_url = "https://{$host}/vectors/list?" . http_build_query($query_params); | |
| 818 | + $list_response = wp_remote_post($list_url, array( | |
| 819 | + 'headers' => array( | |
| 820 | + 'Api-Key' => $api_key, | |
| 821 | + 'accept' => 'application/json', | |
| 822 | + 'content-type' => 'application/json' | |
| 823 | + ), | |
| 824 | + 'body' => wp_json_encode($list_body), | |
| 825 | + 'timeout' => 30 | |
| 826 | + )); | |
| 974 | 827 | |
| 975 | - // Paginate in case a URL has more than 100 chunks. | |
| 976 | - do { | |
| 977 | - $list_response = wp_remote_get($list_url, array( | |
| 978 | - 'headers' => array( | |
| 979 | - 'Api-Key' => $api_key, | |
| 980 | - 'accept' => 'application/json', | |
| 981 | - ), | |
| 982 | - 'timeout' => 30, | |
| 983 | - )); | |
| 984 | - | |
| 985 | - if (is_wp_error($list_response) || wp_remote_retrieve_response_code($list_response) !== 200) { | |
| 986 | - break; | |
| 987 | - } | |
| 988 | - | |
| 828 | + if (!is_wp_error($list_response)) { | |
| 989 | 829 | $list_data = json_decode(wp_remote_retrieve_body($list_response), true); |
| 990 | 830 | if (!empty($list_data['vectors'])) { |
| 991 | 831 | foreach ($list_data['vectors'] as $vector) { |
| 992 | 832 | if (isset($vector['id'])) { |
| @@ -993,17 +833,9 @@ | ||
| 993 | 833 | $vectors_to_delete[] = $vector['id']; |
| 994 | 834 | } |
| 995 | 835 | } |
| 996 | 836 | } |
| 997 | - | |
| 998 | - $next_token = $list_data['pagination']['next'] ?? ''; | |
| 999 | - if (empty($next_token)) { | |
| 1000 | - break; | |
| 1001 | - } | |
| 1002 | - | |
| 1003 | - $query_params['paginationToken'] = $next_token; | |
| 1004 | - $list_url = "https://{$host}/vectors/list?" . http_build_query($query_params); | |
| 1005 | - } while (true); | |
| 837 | + } | |
| 1006 | 838 | |
| 1007 | 839 | if (empty($vectors_to_delete)) { |
| 1008 | 840 | //error_log('[MXCHAT-CHUNK-DELETE] No vectors found to delete'); |
| 1009 | 841 | return true; |
