# eu/jev EU-hosted System One-style inference API for structured choices, probabilities, and scores. - Findings24: - Site web: - Tarifs: Freemium - Catégories: Plateformes d’inférence IA, API de LLM ## Qu'est-ce que eu/jev ? eu/jev is a Bevel inference API for converting context into structured choices, probabilities, and scores. Applications send shared state and one or more named questions to `POST /v1/systemone`, then receive answers designed for direct use in software. Supported question types are `choice`, `noul`, and `score`. The service runs inference on a Hetzner server in Germany and provides a playground for testing workflows such as support routing, sentiment analysis, and content moderation. ## Que peut faire eu/jev ? ### Structured decision primitives Create choice questions with two to ten options, noul true/false questions, or ordered score questions. Responses include the relevant answer fields and, where applicable, probabilities and confidence. ### Jev-style HTTP endpoint Send Bearer-authenticated JSON requests to \`POST /v1/systemone\` using a shared state value and named questions. The documented payload structure is compatible with Jev-style request patterns, while eu/jev uses its own supported model IDs and validation rules. ### Multiple questions in one request A request can include up to eight questions against the same context. State may be supplied as a string, JSON object, or array, allowing applications to submit different forms of structured input. ### Application-ready response metadata The API returns answers keyed to the submitted question names, along with model information, input-token usage, request ID, and server-side latency metadata. ### EU-based processing and data controls Inference runs on Hetzner infrastructure in Germany. The site states that prompts and answers are not stored in the usage database or logs, and provides privacy, DPA, terms, and subprocessor documents. ## Cas d'utilisation ### Support ticket routing Classify an incoming customer issue against operational teams such as billing or technical support, then use the returned choice to direct the ticket to the appropriate queue. ### Content and message classification Evaluate text for categories such as moderation outcomes or sentiment-related labels. The playground can be used to test a question and its criteria before API integration. ### Option and policy evaluation Use choice or noul questions to assess alternatives against descriptions written in natural language, producing a selected option or a true/false result for downstream logic. ### Ordered scoring workflows Use score questions with descriptions arranged from low to high when an application needs an expected position on a defined scale rather than a categorical label. ## Questions fréquentes ### How do I make an API request? Create an account to obtain an API key, then send a Bearer-authenticated POST request to \`/v1/systemone\` with JSON containing \`state\`, \`questions\`, and optionally \`model\`. The documented current model is \`jeff-latest\`; the default key is shown once at signup. ### What does the API return? Answers are keyed to the names of your submitted questions. Choice responses include the selected option, probabilities, and confidence; noul returns a probability; and score returns an expected position with probabilities, confidence, and a legend. Responses also include service metadata. ### How much does eu/jev cost? The published rate is €37 per billion input tokens, and output tokens are free. New accounts receive €5 in starting credits. There is no automatic billing when credits run out, and extra usage can be requested through the service’s request form. ### Where is data processed and stored? Inference runs on a Hetzner server in Germany. The site says prompts and answers are not stored in its usage database or logs; its privacy policy, DPA, and subprocessor list provide additional details about processing and service providers. ### What are the main request limits? The documented limits include eight questions per request, ten options per question, a 256 KiB JSON body, 1,000 characters per instruction, and 500 characters per criterion after serialization. The service also documents a 60-request-per-minute limit across account keys and signed-in playground activity. ## Informations rapides - **Product type**: Inference API for structured decisions - **Provider**: Bevel - **Processing location**: Hetzner server in Germany - **API endpoint**: POST /v1/systemone - **Question types**: choice, noul, and score - **Published pricing**: €37 per billion input tokens; output tokens are free