{"id":7478,"date":"2026-08-19T19:10:08","date_gmt":"2026-08-19T19:10:08","guid":{"rendered":"https:\/\/www.wiven.ai\/wiven-llm\/docs\/administration\/vectoriel\/"},"modified":"2026-08-19T19:10:08","modified_gmt":"2026-08-19T19:10:08","slug":"vectoriel","status":"publish","type":"wvn_doc","link":"https:\/\/www.wiven.ai\/en\/wiven-llm\/docs\/administration\/vectoriel\/","title":{"rendered":"Vector base and embeddings"},"content":{"rendered":"<h1>Vector base and embeddings<\/h1>\n<p>These two building blocks determine the quality of the documentary research. They are<br \/>\nalso the two most expensive choices to change afterwards.<\/p>\n<hr \/>\n<h2>The mechanism<\/h2>\n<ol>\n<li>Each document fragment is transformed into <strong>vector<\/strong> by the engine<br \/>\n   d&#039;embeddings \u2014 a list of numbers representing its meaning.<\/li>\n<li>These vectors are stored in the <strong>vector base<\/strong>.<\/li>\n<li>The question is vectorized by the <strong>same engine<\/strong>, then compared<br \/>\n   to stored vectors.<\/li>\n<li>The closest fragments are selected and provided to the model.<\/li>\n<\/ol>\n<p>Hence the central constraint: <strong>all vectors of an index must come from the<br \/>\nsame embeddings engine<\/strong>. Two different engines produce<br \/>\nunparalleled performances.<\/p>\n<hr \/>\n<h2>The embeddings engine<\/h2>\n<p><code>Settings \u2192 Embedding preferences<\/code>.<\/p>\n<p><strong>integrated motor<\/strong> <em>(by default)<\/em> \u2014 runs locally, does not require a key,<br \/>\nNo network output. Suitable for the vast majority of uses, including<br \/>\nmultilingual.<\/p>\n<p><strong>External motors<\/strong> \u2014 Several suppliers are supported. They can<br \/>\nto bring about gains on demanding or highly specialized multilingual corpora, at<br \/>\nprice of a network output <strong>with each indexing and each question<\/strong>, and a<br \/>\ncost per use.<\/p>\n<blockquote>\n<p><strong>Changing the engine invalidates all existing indexes.<\/strong> No error message<br \/>\nIf you don&#039;t report it, the search results will simply become poor. It&#039;s necessary.<br \/>\nReindex all spaces. Decide at startup.<\/p>\n<\/blockquote>\n<h3>Text segmentation<\/h3>\n<p><code>Settings \u2192 Text Clipping<\/code> two parameters are set:<\/p>\n<ul>\n<li><strong>Fragment size<\/strong> \u2014 shorter, more precise but less contextual; more<br \/>\n  long, the opposite.<\/li>\n<li><strong>Recovery<\/strong> \u2014 the text repeated from one fragment to the next, so as not to<br \/>\n  to split an idea in two.<\/li>\n<\/ul>\n<p>The default values are suitable for most corpora. A change<br \/>\n<strong>only affects documents indexed later<\/strong>.<\/p>\n<p><strong>When to adjust:<\/strong> highly structured and concise documents (fact sheets, questions)<br \/>\nand responses) benefit from smaller fragments; longer argumentative texts<br \/>\n(contracts, case law) benefit from larger fragments.<\/p>\n<hr \/>\n<h2>The vector basis<\/h2>\n<p><code>Settings \u2192 Vector Base<\/code>.<\/p>\n<p><strong>Integrated local base<\/strong> <em>(by default)<\/em> \u2014 stored in the directory of<br \/>\nThe installation. No service to operate, no network output, backed up with<br \/>\nthe rest. She can easily handle corpora of several tens of thousands<br \/>\nof fragments, which covers most organizations.<\/p>\n<p><strong>External databases<\/strong> \u2014 several market solutions are supported, in<br \/>\nSelf-hosted or managed service version.<br \/>\n\u2192 <a href=\"\/en\/wiven-llm\/docs\/reference\/fournisseurs\/\">Full list<\/a><\/p>\n<h3>When to consider an external base<\/h3>\n<ul>\n<li><strong>The volume<\/strong> exceeds what the local base can comfortably handle.<\/li>\n<li><strong>A database already exists<\/strong> in your organization and you want the<br \/>\n  pool resources.<\/li>\n<li><strong>Several applications<\/strong> must share the same index.<\/li>\n<li><strong>High availability<\/strong> is required.<\/li>\n<\/ul>\n<p>In other cases, the local base is the best choice: fewer parts<br \/>\nMobile devices, no network output, unified backup.<\/p>\n<blockquote>\n<p>A managed service means that <strong>your vectors and their metadata are coming out of<br \/>\nyour perimeter<\/strong>. On a site where sovereignty is a key requirement,<br \/>\nThis negates the benefit of a local model.<\/p>\n<\/blockquote>\n<hr \/>\n<h2>Index organization<\/h2>\n<p>Each workspace has its own <strong>own namespace<\/strong> in the database.<br \/>\nThis is what guarantees that a search in a space cannot reach the<br \/>\ndocuments from another person.<\/p>\n<p>Each user&#039;s Second Brain also has its own, separate from the<br \/>\nworkspaces.<\/p>\n<hr \/>\n<h2>Maintenance operations<\/h2>\n<p><strong>Reindex a space.<\/strong> Required after a change of the embeddings engine,<br \/>\nor when an index is suspected of being inconsistent. The operation removes the vectors and<br \/>\nrecalculates them from the documents.<\/p>\n<p><strong>Reset a space.<\/strong> Erase the vectors from the space. The documents<br \/>\nremain in the library and must be added again.<\/p>\n<p><strong>Check the number of vectors.<\/strong> The administration displays the indexed total \u2014<br \/>\nuseful for confirming that indexing has been successful.<\/p>\n<hr \/>\n<h2>Diagnosing a bad search<\/h2>\n<table>\n<thead>\n<tr>\n<th>Symptom<\/th>\n<th>Track<\/th>\n<\/tr>\n<\/thead>\n<tbody>\n<tr>\n<td>\u00ab&quot;He doesn&#039;t realize it&#039;s written there.&quot;\u00bb<\/td>\n<td>Similarity threshold too high; or document not added to the space; or text incorrectly extracted<\/td>\n<\/tr>\n<tr>\n<td>\u00ab&quot;He quotes irrelevant passages.&quot;\u00bb<\/td>\n<td>Threshold too low, or too many fragments requested<\/td>\n<\/tr>\n<tr>\n<td>\u00ab&quot;He only finds part of it.&quot;\u00bb<\/td>\n<td>Number of fragments too low; increase gradually<\/td>\n<\/tr>\n<tr>\n<td>\u00ab&quot;The responses deteriorated suddenly.&quot;\u00bb<\/td>\n<td>Changing the embeddings engine without reindexing<\/td>\n<\/tr>\n<tr>\n<td>\u00ab&quot;He finds it in an outdated document&quot;\u00bb<\/td>\n<td>Two versions coexist in the index; remove the old one<\/td>\n<\/tr>\n<\/tbody>\n<\/table>\n<p>The thresholds and the number of fragments are adjustable. <strong>per workspace<\/strong>.<br \/>\n\u2192 <a href=\"\/en\/wiven-llm\/docs\/utilisation\/conversations\/\">Conversations<\/a><\/p>","protected":false},"parent":7451,"menu_order":503,"template":"","meta":[],"class_list":["post-7478","wvn_doc","type-wvn_doc","status-publish","hentry"],"yoast_head":"<!-- This site is optimized with the Yoast SEO plugin v28.6 - https:\/\/yoast.com\/product\/yoast-seo-wordpress\/ -->\n<title>Base vectorielle et embeddings \u2014 Documentation WivenLLM<\/title>\n<meta name=\"description\" content=\"Comment fonctionne la recherche documentaire, quel moteur choisir, et quand changer de base.\" \/>\n<meta name=\"robots\" content=\"index, follow, max-snippet:-1, max-image-preview:large, max-video-preview:-1\" \/>\n<link rel=\"canonical\" href=\"https:\/\/www.wiven.ai\/en\/wiven-llm\/docs\/administration\/vectoriel\/\" \/>\n<meta property=\"og:locale\" content=\"en_US\" \/>\n<meta property=\"og:type\" content=\"article\" \/>\n<meta property=\"og:title\" content=\"Base vectorielle et embeddings \u2014 Documentation WivenLLM\" \/>\n<meta property=\"og:description\" content=\"Comment fonctionne la recherche documentaire, quel moteur choisir, et quand changer de base.\" \/>\n<meta property=\"og:url\" content=\"https:\/\/www.wiven.ai\/en\/wiven-llm\/docs\/administration\/vectoriel\/\" \/>\n<meta property=\"og:site_name\" content=\"Wiven AI\" \/>\n<meta name=\"twitter:card\" content=\"summary_large_image\" \/>\n<meta name=\"twitter:label1\" content=\"Est. reading time\" \/>\n\t<meta name=\"twitter:data1\" content=\"4 minutes\" \/>\n<script type=\"application\/ld+json\" class=\"yoast-schema-graph\">{\"@context\":\"https:\\\/\\\/schema.org\",\"@graph\":[{\"@type\":\"WebPage\",\"@id\":\"https:\\\/\\\/www.wiven.ai\\\/wiven-llm\\\/docs\\\/administration\\\/vectoriel\\\/\",\"url\":\"https:\\\/\\\/www.wiven.ai\\\/wiven-llm\\\/docs\\\/administration\\\/vectoriel\\\/\",\"name\":\"Base vectorielle et embeddings \u2014 Documentation WivenLLM\",\"isPartOf\":{\"@id\":\"https:\\\/\\\/www.wiven.ai\\\/#website\"},\"datePublished\":\"2026-08-19T19:10:08+00:00\",\"description\":\"Comment fonctionne la recherche documentaire, quel moteur choisir, et quand changer de base.\",\"breadcrumb\":{\"@id\":\"https:\\\/\\\/www.wiven.ai\\\/wiven-llm\\\/docs\\\/administration\\\/vectoriel\\\/#breadcrumb\"},\"inLanguage\":\"en\",\"potentialAction\":[{\"@type\":\"ReadAction\",\"target\":[\"https:\\\/\\\/www.wiven.ai\\\/wiven-llm\\\/docs\\\/administration\\\/vectoriel\\\/\"]}]},{\"@type\":\"BreadcrumbList\",\"@id\":\"https:\\\/\\\/www.wiven.ai\\\/wiven-llm\\\/docs\\\/administration\\\/vectoriel\\\/#breadcrumb\",\"itemListElement\":[{\"@type\":\"ListItem\",\"position\":1,\"name\":\"Home\",\"item\":\"https:\\\/\\\/www.wiven.ai\\\/\"},{\"@type\":\"ListItem\",\"position\":2,\"name\":\"Documentation\",\"item\":\"https:\\\/\\\/www.wiven.ai\\\/wiven-llm\\\/docs\\\/\"},{\"@type\":\"ListItem\",\"position\":3,\"name\":\"Administration\",\"item\":\"https:\\\/\\\/www.wiven.ai\\\/wiven-llm\\\/docs\\\/administration\\\/\"},{\"@type\":\"ListItem\",\"position\":4,\"name\":\"Base vectorielle et embeddings\"}]},{\"@type\":\"WebSite\",\"@id\":\"https:\\\/\\\/www.wiven.ai\\\/#website\",\"url\":\"https:\\\/\\\/www.wiven.ai\\\/\",\"name\":\"Wiven AI\",\"description\":\"\",\"potentialAction\":[{\"@type\":\"SearchAction\",\"target\":{\"@type\":\"EntryPoint\",\"urlTemplate\":\"https:\\\/\\\/www.wiven.ai\\\/?s={search_term_string}\"},\"query-input\":{\"@type\":\"PropertyValueSpecification\",\"valueRequired\":true,\"valueName\":\"search_term_string\"}}],\"inLanguage\":\"en\"}]}<\/script>\n<!-- \/ Yoast SEO plugin. -->","yoast_head_json":{"title":"Vector base and embeddings \u2014 WivenLLM documentation","description":"How does document retrieval work, which search engine to choose, and when to change databases?.","robots":{"index":"index","follow":"follow","max-snippet":"max-snippet:-1","max-image-preview":"max-image-preview:large","max-video-preview":"max-video-preview:-1"},"canonical":"https:\/\/www.wiven.ai\/en\/wiven-llm\/docs\/administration\/vectoriel\/","og_locale":"en_US","og_type":"article","og_title":"Base vectorielle et embeddings \u2014 Documentation WivenLLM","og_description":"Comment fonctionne la recherche documentaire, quel moteur choisir, et quand changer de base.","og_url":"https:\/\/www.wiven.ai\/en\/wiven-llm\/docs\/administration\/vectoriel\/","og_site_name":"Wiven AI","twitter_card":"summary_large_image","twitter_misc":{"Est. reading time":"4 minutes"},"schema":{"@context":"https:\/\/schema.org","@graph":[{"@type":"WebPage","@id":"https:\/\/www.wiven.ai\/wiven-llm\/docs\/administration\/vectoriel\/","url":"https:\/\/www.wiven.ai\/wiven-llm\/docs\/administration\/vectoriel\/","name":"Vector base and embeddings \u2014 WivenLLM documentation","isPartOf":{"@id":"https:\/\/www.wiven.ai\/#website"},"datePublished":"2026-08-19T19:10:08+00:00","description":"How does document retrieval work, which search engine to choose, and when to change databases?.","breadcrumb":{"@id":"https:\/\/www.wiven.ai\/wiven-llm\/docs\/administration\/vectoriel\/#breadcrumb"},"inLanguage":"en","potentialAction":[{"@type":"ReadAction","target":["https:\/\/www.wiven.ai\/wiven-llm\/docs\/administration\/vectoriel\/"]}]},{"@type":"BreadcrumbList","@id":"https:\/\/www.wiven.ai\/wiven-llm\/docs\/administration\/vectoriel\/#breadcrumb","itemListElement":[{"@type":"ListItem","position":1,"name":"Home","item":"https:\/\/www.wiven.ai\/"},{"@type":"ListItem","position":2,"name":"Documentation","item":"https:\/\/www.wiven.ai\/wiven-llm\/docs\/"},{"@type":"ListItem","position":3,"name":"Administration","item":"https:\/\/www.wiven.ai\/wiven-llm\/docs\/administration\/"},{"@type":"ListItem","position":4,"name":"Base vectorielle et embeddings"}]},{"@type":"WebSite","@id":"https:\/\/www.wiven.ai\/#website","url":"https:\/\/www.wiven.ai\/","name":"Wiven AI","description":"","potentialAction":[{"@type":"SearchAction","target":{"@type":"EntryPoint","urlTemplate":"https:\/\/www.wiven.ai\/?s={search_term_string}"},"query-input":{"@type":"PropertyValueSpecification","valueRequired":true,"valueName":"search_term_string"}}],"inLanguage":"en"}]}},"_links":{"self":[{"href":"https:\/\/www.wiven.ai\/en\/wp-json\/wp\/v2\/wvn_doc\/7478","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/www.wiven.ai\/en\/wp-json\/wp\/v2\/wvn_doc"}],"about":[{"href":"https:\/\/www.wiven.ai\/en\/wp-json\/wp\/v2\/types\/wvn_doc"}],"up":[{"embeddable":true,"href":"https:\/\/www.wiven.ai\/en\/wp-json\/wp\/v2\/wvn_doc\/7451"}],"wp:attachment":[{"href":"https:\/\/www.wiven.ai\/en\/wp-json\/wp\/v2\/media?parent=7478"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}