{"id":81463,"date":"2026-10-04T07:30:00","date_gmt":"2026-10-04T07:30:00","guid":{"rendered":"https:\/\/www.arimetrics.com\/?post_type=encyclopedia&#038;p=81463"},"modified":"2026-10-04T07:42:14","modified_gmt":"2026-10-04T07:42:14","slug":"jev","status":"publish","type":"encyclopedia","link":"https:\/\/www.arimetrics.com\/en\/digital-glossary\/jev","title":{"rendered":"Jev"},"content":{"rendered":"<p><img decoding=\"async\" class=\"boxpad alignright wp-image-81464 size-full\" src=\"https:\/\/www.arimetrics.com\/wp-content\/uploads\/2026\/10\/jev-square.jpg\" style=\"margin-top:0;\" alt=\"Jev logo in turquoise\" width=\"300\" height=\"300\" srcset=\"https:\/\/www.arimetrics.com\/wp-content\/uploads\/2026\/10\/jev-square.jpg 300w, https:\/\/www.arimetrics.com\/wp-content\/uploads\/2026\/10\/jev-square-150x150.jpg 150w\" sizes=\"(max-width: 300px) 100vw, 300px\" \/> <strong>Definition:<\/strong><\/p>\n<p><strong>Jev<\/strong> is an artificial intelligence model developed by TypeSafe AI to make fast, structured decisions inside software applications. It receives a state containing information and one or more typed questions, then returns values that code can use directly.<\/p>\n<p>Unlike a generative language model, Jev does not write responses, explanations, or code. Its role is to produce <strong>constrained results<\/strong>, accompanied by probabilities and, for certain question types, a confidence measure.<\/p>\n\n<h2>Origin of Jev<\/h2>\n<p>TypeSafe AI introduced Jev on September 15, 2026 as its first <strong>System One Model<\/strong>. The term describes a class of models designed to make fast, bounded judgments within a software workflow.<\/p>\n<p>The <strong>name System One<\/strong> refers to the fast and intuitive mode of thinking popularized by Daniel Kahneman. Jev, in turn, is named after economist William Stanley Jevons. The <a href=\"https:\/\/typesafe.ai\/blog\/introducing-system-one-models-and-jev\" target=\"_blank\" rel=\"noopener\">official Jev announcement<\/a> explains that the project separates probabilistic judgments from deterministic logic that an application can execute on its own.<\/p>\n<h2>How Jev works<\/h2>\n<p>An application sends Jev a <strong>state<\/strong>, which may contain text or structured data, together with specific questions. The model evaluates each question independently and returns its answers in a predefined structure.<\/p>\n<p>Requests are made through an <a href=\"https:\/\/www.arimetrics.com\/en\/digital-glossary\/api\">API<\/a> or one of the TypeSafe AI SDKs. Code retains <strong>workflow control<\/strong>: it prepares the context, defines valid options, interprets probabilities, and decides whether to perform an action, request review, or route the case to another system.<\/p>\n<p>Jev can evaluate several questions in a single request. By processing them <strong>in parallel<\/strong>, it prevents one answer from becoming hidden context for the others. This is useful when a complex decision can be decomposed into independent judgments.<\/p>\n<h2>Response types<\/h2>\n<p>TypeSafe AI calls its three available question types primitives. Each one determines the <strong>output shape<\/strong> that the model is allowed to return:<\/p>\n<ul>\n<li><strong>Choice:<\/strong> Selects an option from a defined set and returns the associated probabilities and a confidence measure.<\/li>\n<li><strong>Score:<\/strong> Places the state on an ordered scale, such as levels of urgency or suitability, and returns a score, probabilities, and confidence.<\/li>\n<li><strong>Noul:<\/strong> Evaluates a binary question and returns the probability that the answer is yes.<\/li>\n<\/ul>\n<p>The <a href=\"https:\/\/docs.typesafe.ai\/introduction\" target=\"_blank\" rel=\"noopener\">TypeSafe AI documentation<\/a> explains how these questions can be combined. In a <strong>practical example<\/strong>, an application analyzing a customer support request can separately ask whether the message requests a refund, how urgent it is, and which team should handle it.<\/p>\n<h2>Differences from LLMs<\/h2>\n<p>Large language models generate text token by token and can respond to open-ended instructions. Jev uses <strong>bounded answers<\/strong> that must conform to the types and options defined by the application. It does not act as a chatbot or provide a narrative explanation of its reasoning.<\/p>\n<p>This constraint removes the need to extract a decision from free-form prose, but it does not guarantee that every judgment is correct. Its probabilities are intended to express <strong>uncertainty<\/strong> across groups of predictions, not to certify an individual answer.<\/p>\n<p>Jev is not an <a href=\"https:\/\/www.arimetrics.com\/en\/digital-glossary\/ai-agent\">AI agent<\/a>, either. It is <strong>not autonomous<\/strong>: it does not choose its next action or use tools on its own initiative. It can be part of an agent or another system, but external code and architecture decide what to do with its output.<\/p>\n<h2>Applications of Jev<\/h2>\n<p>Jev is intended for points in a process where software needs to interpret unstructured information and turn it into an <strong>actionable decision<\/strong>. Its applications include:<\/p>\n<ul>\n<li><strong>Classification:<\/strong> Assigning messages, documents, or incidents to a known category.<\/li>\n<li><strong>Routing:<\/strong> Choosing which team, model, or process should receive a request.<\/li>\n<li><strong>Evaluation:<\/strong> Scoring urgency, relevance, quality, or suitability against defined criteria.<\/li>\n<li><strong>Verification:<\/strong> Checking whether a claim is supported by a source or whether a passage answers a query.<\/li>\n<li><strong>Control:<\/strong> Detecting content that requires review before another system continues the workflow.<\/li>\n<\/ul>\n<p>In marketing and customer service, these capabilities can classify inquiries, estimate intent or frustration, and route each case. Jev supplies the probabilistic judgment, while <strong>business rules<\/strong>, thresholds, and actions remain in software.<\/p>\n<h2>Limits of Jev<\/h2>\n<p>Jev works with text inputs and questions that have a <strong>bounded space<\/strong> of answers. It is not designed to generate content, solve extended reasoning tasks, or replace exact operations that software can compute deterministically.<\/p>\n<p>Quality depends on providing <strong>relevant context<\/strong> and making each question represent a clear judgment. Excess irrelevant information, contradictory instructions, and content written to influence the model can alter its answers.<\/p>\n<p>Probabilities and confidence can inform thresholds, but they require evaluation in the real use case. Low-confidence results may be sent to <strong>human review<\/strong> or a reasoning model, while sensitive actions should remain subject to controls defined by the application.<\/p>\n","protected":false},"excerpt":{"rendered":"<p>Jev is TypeSafe AI&#8217;s System One model for typed, probabilistic decisions inside software. Learn how it works, its applications, and its limits.<\/p>\n","protected":false},"author":6,"featured_media":81468,"template":"","encyclopedia-tag":[1462],"class_list":["post-81463","encyclopedia","type-encyclopedia","status-publish","has-post-thumbnail","hentry","encyclopedia-tag-ai-platforms-and-models"],"_links":{"self":[{"href":"https:\/\/www.arimetrics.com\/en\/wp-json\/wp\/v2\/encyclopedia\/81463","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/www.arimetrics.com\/en\/wp-json\/wp\/v2\/encyclopedia"}],"about":[{"href":"https:\/\/www.arimetrics.com\/en\/wp-json\/wp\/v2\/types\/encyclopedia"}],"author":[{"embeddable":true,"href":"https:\/\/www.arimetrics.com\/en\/wp-json\/wp\/v2\/users\/6"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/www.arimetrics.com\/en\/wp-json\/wp\/v2\/media\/81468"}],"wp:attachment":[{"href":"https:\/\/www.arimetrics.com\/en\/wp-json\/wp\/v2\/media?parent=81463"}],"wp:term":[{"taxonomy":"encyclopedia-tag","embeddable":true,"href":"https:\/\/www.arimetrics.com\/en\/wp-json\/wp\/v2\/encyclopedia-tag?post=81463"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}