{"id":7461,"date":"2026-08-19T19:10:05","date_gmt":"2026-08-19T19:10:05","guid":{"rendered":"https:\/\/www.wiven.ai\/wiven-llm\/docs\/demarrage\/modeles\/"},"modified":"2026-08-19T19:10:05","modified_gmt":"2026-08-19T19:10:05","slug":"modeles","status":"publish","type":"wvn_doc","link":"https:\/\/www.wiven.ai\/en\/wiven-llm\/docs\/demarrage\/modeles\/","title":{"rendered":"Choosing your models"},"content":{"rendered":"<h1>Choosing your models<\/h1>\n<p>The language model determines the quality, speed, and cost of responses.,<br \/>\nas well as the question of whether data is falling outside your scope.<br \/>\nWivenLLM does not impose any of them on you.<\/p>\n<hr \/>\n<h2>Three families<\/h2>\n<h3>1. Integrated local models<\/h3>\n<p>WivenLLM downloads and runs open models itself, without dependencies<br \/>\nExternal. An integrated catalog offers models in various sizes, from the model<br \/>\nlightweight, with a few billion parameters compared to significantly heavier models.<\/p>\n<ul>\n<li><strong>No data is being released<\/strong>, No API key, no usage costs.<\/li>\n<li>The speed depends entirely on your equipment.<\/li>\n<li>The model is loaded into memory on first use and unloaded<br \/>\n  automatically after a period of inactivity, to free up resources.<\/li>\n<\/ul>\n<p><strong>Hardware diagnosis<\/strong> (<code>Settings \u2192 Hardware diagnostics<\/code>) analyze your<br \/>\nmachine, performs a short measurement and indicates for each model in the catalog<br \/>\nif it is <strong>comfortable<\/strong>, <strong>limit<\/strong> Or <strong>out of reach<\/strong>. This is the way<br \/>\nfaster to know what to download.<\/p>\n<h3>2. Internal Inference Server<\/h3>\n<p>Does your organization already operate a template server? WivenLLM connects to it.<br \/>\nas a customer. The principle remains the same \u2014 nothing leaves the network \u2014 but it&#039;s<br \/>\nYour server manages the model lifecycle, not WivenLLM.<\/p>\n<p>Any server exposing an API compatible with the OpenAI standard is suitable, y<br \/>\nincluding the most widespread local solutions.<\/p>\n<h3>3. External Suppliers<\/h3>\n<p>Large commercial suppliers are supported. They offer<br \/>\ngenerally the best quality and the best speed, without requirements<br \/>\nmaterial.<\/p>\n<p><strong>In return:<\/strong> your messages and the associated documentary context are<br \/>\nThis information is sent to the supplier, and usage is billed. This choice must be conscious and<br \/>\ndocumented with users.<\/p>\n<p>\u2192 <a href=\"\/en\/wiven-llm\/docs\/reference\/fournisseurs\/\">Supported providers<\/a><\/p>\n<hr \/>\n<h2>What equipment is needed for a local model?<\/h2>\n<p>These orders of magnitude relate to the quantified models of the integrated catalogue.<\/p>\n<table>\n<thead>\n<tr>\n<th>Model size<\/th>\n<th>Useful RAM<\/th>\n<th>No GPU<\/th>\n<th>With a suitable GPU<\/th>\n<\/tr>\n<\/thead>\n<tbody>\n<tr>\n<td>3 billion parameters<\/td>\n<td>8 GB<\/td>\n<td>Usable<\/td>\n<td>Very fast<\/td>\n<\/tr>\n<tr>\n<td>7\u20139 billion<\/td>\n<td>16 GB<\/td>\n<td>Slow but doable<\/td>\n<td>Fast<\/td>\n<\/tr>\n<tr>\n<td>14 billion<\/td>\n<td>24 GB<\/td>\n<td>Difficult<\/td>\n<td>Comfortable<\/td>\n<\/tr>\n<tr>\n<td>32 billion<\/td>\n<td>32 GB and more<\/td>\n<td>Not recommended<\/td>\n<td>Comfortable with a sized GPU<\/td>\n<\/tr>\n<tr>\n<td>70 billion<\/td>\n<td>48 GB and more<\/td>\n<td>No<\/td>\n<td>Requires a high-end GPU<\/td>\n<\/tr>\n<\/tbody>\n<\/table>\n<p><strong>Two points that are often underestimated:<\/strong><\/p>\n<ul>\n<li><strong>Memory is the hard constraint.<\/strong> A model that is not stored in memory does not<br \/>\n  It does not start, or it runs via disk swapping at an unusable speed.<\/li>\n<li><strong>The GPU changes the scale, not the result.<\/strong> It multiplies the speed of<br \/>\n  generation; it doesn&#039;t make a small model smarter.<\/li>\n<\/ul>\n<hr \/>\n<h2>The agent model may be different<\/h2>\n<p>A workspace distinguishes between two models:<\/p>\n<ul>\n<li>THE <strong>conversation model<\/strong>, who answers questions; ;<\/li>\n<li>THE <strong>agent model<\/strong>, which controls the sequence of tools.<\/li>\n<\/ul>\n<p>The agents demand more: the model must decide which tool to call, with<br \/>\nWhat arguments, and how to interpret the result. A model that converses<br \/>\nCorrectly, it can fail as an agent.<\/p>\n<p><strong>Recommendation :<\/strong> If agents matter to you, assign them the template<br \/>\nthe most capable one you have available, even if the current conversation revolves around<br \/>\na lighter model.<\/p>\n<p>\u2192 <a href=\"\/en\/wiven-llm\/docs\/agents\/\">Agents Verification Status<\/a><\/p>\n<hr \/>\n<h2>The embeddings engine is a separate choice<\/h2>\n<p>He doesn&#039;t write anything: he translates the text into vectors for research.<br \/>\ndocumentary. The integrated engine runs locally and is suitable for most<br \/>\nuses.<\/p>\n<blockquote>\n<p><strong>Structuring constraint:<\/strong> all vectors of an index must come from the<br \/>\nsame engine. Changing it renders existing indexes unusable and imposes a<br \/>\nComplete reindexing of all spaces. Decide this at startup.<\/p>\n<\/blockquote>\n<p>An external engine may be justified for demanding multilingual corpora or<br \/>\nvery high search quality requirements \u2014 at the cost of a network output<br \/>\neach indexing and each question.<\/p>\n<hr \/>\n<h2>Arbitrate<\/h2>\n<table>\n<thead>\n<tr>\n<th>Your priority<\/th>\n<th>Configuration<\/th>\n<\/tr>\n<\/thead>\n<tbody>\n<tr>\n<td>Strict confidentiality<\/td>\n<td>Integrated local model + native embeddings + local vector database<\/td>\n<\/tr>\n<tr>\n<td>Maximum quality<\/td>\n<td>External provider, explicitly mentioned to users<\/td>\n<\/tr>\n<tr>\n<td>Zero cost in use<\/td>\n<td>Any premises<\/td>\n<\/tr>\n<tr>\n<td>Speed on modest equipment<\/td>\n<td>External provider, or shared internal inference server<\/td>\n<\/tr>\n<tr>\n<td>Both, depending on the case<\/td>\n<td><a href=\"\/en\/wiven-llm\/docs\/administration\/routeur\/\">Model router<\/a> local by default, escalation based on rules<\/td>\n<\/tr>\n<\/tbody>\n<\/table>\n<hr \/>\n<h2>Change your mind later<\/h2>\n<ul>\n<li><strong>Change language model<\/strong> : without consequence. The documents remain<br \/>\n  Once indexed, the conversations remain readable.<\/li>\n<li><strong>Change your embeddings engine<\/strong> : requires a complete re-indexing.<\/li>\n<li><strong>Change vector basis<\/strong> : requires a complete re-indexing.<\/li>\n<\/ul>\n<p>In other words, only the first of these three choices is truly reversible without<br \/>\neffort. That&#039;s why the other two deserve a thoughtful decision.<br \/>\nthe installation.<\/p>","protected":false},"parent":7449,"menu_order":203,"template":"","meta":[],"class_list":["post-7461","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>Choisir ses mod\u00e8les \u2014 Documentation WivenLLM<\/title>\n<meta name=\"description\" content=\"Local, serveur d&#039;inf\u00e9rence interne ou fournisseur externe \u2014 comment arbitrer, et quel mat\u00e9riel pr\u00e9voir.\" \/>\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\/demarrage\/modeles\/\" \/>\n<meta property=\"og:locale\" content=\"en_US\" \/>\n<meta property=\"og:type\" content=\"article\" \/>\n<meta property=\"og:title\" content=\"Choisir ses mod\u00e8les \u2014 Documentation WivenLLM\" \/>\n<meta property=\"og:description\" content=\"Local, serveur d&#039;inf\u00e9rence interne ou fournisseur externe \u2014 comment arbitrer, et quel mat\u00e9riel pr\u00e9voir.\" \/>\n<meta property=\"og:url\" content=\"https:\/\/www.wiven.ai\/en\/wiven-llm\/docs\/demarrage\/modeles\/\" \/>\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\\\/demarrage\\\/modeles\\\/\",\"url\":\"https:\\\/\\\/www.wiven.ai\\\/wiven-llm\\\/docs\\\/demarrage\\\/modeles\\\/\",\"name\":\"Choisir ses mod\u00e8les \u2014 Documentation WivenLLM\",\"isPartOf\":{\"@id\":\"https:\\\/\\\/www.wiven.ai\\\/#website\"},\"datePublished\":\"2026-08-19T19:10:05+00:00\",\"description\":\"Local, serveur d'inf\u00e9rence interne ou fournisseur externe \u2014 comment arbitrer, et quel mat\u00e9riel pr\u00e9voir.\",\"breadcrumb\":{\"@id\":\"https:\\\/\\\/www.wiven.ai\\\/wiven-llm\\\/docs\\\/demarrage\\\/modeles\\\/#breadcrumb\"},\"inLanguage\":\"en\",\"potentialAction\":[{\"@type\":\"ReadAction\",\"target\":[\"https:\\\/\\\/www.wiven.ai\\\/wiven-llm\\\/docs\\\/demarrage\\\/modeles\\\/\"]}]},{\"@type\":\"BreadcrumbList\",\"@id\":\"https:\\\/\\\/www.wiven.ai\\\/wiven-llm\\\/docs\\\/demarrage\\\/modeles\\\/#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\":\"D\u00e9marrage\",\"item\":\"https:\\\/\\\/www.wiven.ai\\\/wiven-llm\\\/docs\\\/demarrage\\\/\"},{\"@type\":\"ListItem\",\"position\":4,\"name\":\"Choisir ses mod\u00e8les\"}]},{\"@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":"Choosing your templates \u2014 WivenLLM documentation","description":"Local, internal inference server or external provider \u2014 how to decide, and what hardware to plan for.","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\/demarrage\/modeles\/","og_locale":"en_US","og_type":"article","og_title":"Choisir ses mod\u00e8les \u2014 Documentation WivenLLM","og_description":"Local, serveur d'inf\u00e9rence interne ou fournisseur externe \u2014 comment arbitrer, et quel mat\u00e9riel pr\u00e9voir.","og_url":"https:\/\/www.wiven.ai\/en\/wiven-llm\/docs\/demarrage\/modeles\/","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\/demarrage\/modeles\/","url":"https:\/\/www.wiven.ai\/wiven-llm\/docs\/demarrage\/modeles\/","name":"Choosing your templates \u2014 WivenLLM documentation","isPartOf":{"@id":"https:\/\/www.wiven.ai\/#website"},"datePublished":"2026-08-19T19:10:05+00:00","description":"Local, internal inference server or external provider \u2014 how to decide, and what hardware to plan for.","breadcrumb":{"@id":"https:\/\/www.wiven.ai\/wiven-llm\/docs\/demarrage\/modeles\/#breadcrumb"},"inLanguage":"en","potentialAction":[{"@type":"ReadAction","target":["https:\/\/www.wiven.ai\/wiven-llm\/docs\/demarrage\/modeles\/"]}]},{"@type":"BreadcrumbList","@id":"https:\/\/www.wiven.ai\/wiven-llm\/docs\/demarrage\/modeles\/#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":"D\u00e9marrage","item":"https:\/\/www.wiven.ai\/wiven-llm\/docs\/demarrage\/"},{"@type":"ListItem","position":4,"name":"Choisir ses mod\u00e8les"}]},{"@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\/7461","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\/7449"}],"wp:attachment":[{"href":"https:\/\/www.wiven.ai\/en\/wp-json\/wp\/v2\/media?parent=7461"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}