17 guides
Learn Jev
Short, sourced guides to TypeSafe's decision model. Each one links to the docs it cites and to real projects from the directory.
01What is Jev?Jev is TypeSafe AI's first System One model. It takes text and typed questions and returns structured answers with probabilities, not generated text.Read →02Jev pricing and API costJev costs $0.042 per million input tokens ($42 per billion) and output tokens are free. What you pay for, worked examples, limits, and what is not public.Read →03Jev vs LLMs: when to use whichUse Jev when code needs a fast decision from answers you define; use an LLM for text, code, or long reasoning. How they differ and how to combine them.Read →04How to use Jev: a quickstartGet a TypeSafe API key, POST a state and typed questions to /v1/systemone, and branch on the answers. Playground, cURL, Python and JavaScript steps.Read →05Jev and MCP: using Jev as a Model Context Protocol toolIs there a Jev MCP server? TypeSafe has no official one. Here is how community MCP servers wrap Jev, what to check, and which projects to try.Read →06How to use Jev with Claude CodeJev cannot replace Claude Code's model, but the official TypeSafe skill and community hooks let Claude Code use Jev for fast, typed decisions.Read →07Jev in an agentic harness: where it fits in an agent loopJev is not an agent, but it fits the decision points in an agent harness: picking the next action, skill or tool with calibrated confidence.Read →08Jev for orchestration: routing requests, models and skillsHow to use Jev as a router: classify each request, read the confidence, and send it to code, an LLM, a model tier or a person. With real projects.Read →09Jev as a judge: evaluating LLM and agent outputsHow to use Jev as an evaluator: typed Choice, Score and Noul questions with probabilities instead of a written verdict. Patterns, limits and projects.Read →10Jev for SEO and GEOUsing Jev for SEO and GEO: classify search intent, grade pages against a rubric and filter crawl data, with rules kept in code. Patterns and projects.Read →11Jev for ads and ad reviewUsing Jev for advertising: check ad copy against your policies, tag creatives by hook and funnel stage, and send unsure cases to a person.Read →12Jev for marketing and lead scoringUsing Jev for marketing: score leads against your ICP, prioritise outreach and check content, with weights and thresholds kept in your code.Read →13Is there an open source Jev?Jev is not open source or open weights; it runs on TypeSafe's hosted API. What TypeSafe does publish, and the community's local Jev alternatives.Read →14Can you fine-tune Jev?No. TypeSafe does not fine-tune or LoRA-adapt Jev per customer. How to adapt Jev without training, and how community projects train Jev-like models.Read →15Designing questions for JevWhat is Jev engineering? A community term for designing the state, questions and confidence thresholds Jev works with, using TypeSafe's own guidance.Read →16What is RLCD?A plain-English guide to Reinforcement Learning for Calibrated Decisions, the training method behind Jev.Read →17Jev vs. classifiersAn independent benchmark of Jev against classical classifiers on standard datasets.Read →18Jev statisticsHow many Jev projects exist, by use case, language and day, from this directory’s data.Read →