Jev Lab
Recipes that run live against TypeSafe's Jev through OpenRouter. Each one is the code plus the result.
Support message triage
Five yes-or-no questions about every one of 95 messages, in about a second.
475 answers in 1.2 s
Jev · $0.0014 a run
Agent action approval
Four checks on each of 24 tool calls in one request. Risky steps stop for a person.
96 answers in 0.5 s
Jev · $0.0004 a run
Extraction without guessing
Jev picks each field from candidates found in the text, so it cannot invent a value.
12 fields in 0.6 s
Jev · $0.0002 a document
Prompt to questions
Paste a prompt. Claude rewrites it as Jev questions, then Jev runs them on every row.
380 answers in 2 s
Jev + Claude Haiku 4.5 · $0.0025 a run
Feed filter in plain words
Write a rule in plain words and Jev flags every matching post, in one request.
40 posts in 0.5 s
Jev · $0.0003 a run
Live chase opponent
You run, Jev chases. Every chaser move is one request, two to three times a second.
2 to 3 moves a second
Jev · $0.16 an hour
Checkout form that notices
A checkout form asks eight questions after every click and reacts to the answers.
8 answers in 300 ms
Jev · $0.0001 a check
Documentation
Guides and cookbooks for building with Jev on OpenRouter.
- Jev TutorialGuide
- Jev (TypeSafe)Guide
- Multimodal DecisionsGuide
- TypeSafe SDK (Jev)Guide
- Gate Agent Tool Calls with JevCookbook
- Auto-Approve Coding Agent Permission Prompts with JevCookbook
- Classify Reddit and YouTube Comments with Jev and ScrapeCreatorsCookbook
- Classify and Tag Text at Scale with JevCookbook
- Cut LLM Cost with a Jev-Verified CascadeCookbook
- How to Get Structured Output from Any LLM: JSON Schema, Tool Calls, or a Decision ModelBlog post
- How to Use Jev: Moderation with the Jev API in TypeScriptBlog post
- Is Jev as Accurate as Frontier Models at Classification?Blog post
- What Is Jev? TypeSafe's Decision Model Explained for DevelopersBlog post
- Jev vs LLM-as-a-JudgeBlog post
- Jev vs LLM: When to Use a Decision Model Instead of Generating TextBlog post