An insurance claim triage tool that classifies loss type, scores documentation completeness, and flags fraud indicators in a first-notice-of-loss narrative, routing claims without denying any or setting a payout.
A native Brazilian Portuguese typed-decisions benchmark: seven choice and noul tasks over legislative, legal, scientific, and fact-checking text, with a tuned 4B model at 72.9% mean.
Grounded information extraction library: deterministic code proposes candidate spans with exact offsets and Jev answers one Choice per span plus sentence-level Nouls, so every extraction is a substring of the source. Ships a bilingual benchmark and paper.
Graph RAG framework over Milvus vector search with no graph database required: Jev Noul judgments select graph relations and rerank full source passages alongside vector-search candidates, reaching 86.07% average Recall@5 on 1,000 MuSiQue and 1,000 2Wiki queries in the two-stage evaluation.
TypeSafe's data processing addendum, last updated 24 April 2026. For customer personal data, the customer is the controller and TypeSafe is the processor.
The index of TypeSafe's worked examples, from parallel questions and reranking to guardrails, date extraction, and self-consistency, each with datasets and measured results.
Two browser games against a live Jev opponent: a first-person quick-scope arena and a neon survival challenge, where Jev picks tactics and authored code handles movement and physics.
An autonomous Pac-Man recreation steered by Jev: before each junction the game sends the maze, actors, and legal directions as structured data, and Jev returns one direction with a probability per option.
Pokémon Red under PyBoy where ordinary Python owns the route and arithmetic and Jev is asked only at branches, with each battle turn's faint prediction scored by Brier.
A TypeScript library that walks a taxonomy and calls Jev once per level, so a choice can cover thousands of leaves despite Jev's 255-option limit per question.
A browser Snake game where code computes the legal moves, food distance, and reachable space and Jev picks one move per tick with a single Choice question; a late answer leaves the snake going straight.
A Next.js playground that benchmarks Jev against GPT-5.6 Luna, Claude Haiku 4.5, Gemini 3.5 Flash Lite, and GPT-6 Astra in tic-tac-toe and Connect Four, giving every model the same legal-move Choice options through the Vercel AI Gateway.
A single shared chess game where visitors play against Jev, which weighs every legal move at once as one Choice question and shades the board with its probabilities.
A map of documented Jev use cases, routing, guardrail verification, citation checking, reranking, rubric scoring, bulk classification, and semantic linting in CI, each with a confidence policy.
A single shared chess game where visitors play against Jev, which weighs every legal move at once as one Choice question and shades the board with its probabilities.
A Score is a weighted position, not a ruler. Why the number can land between levels, why levels are judged separately, and how to combine several atomic scores in code instead of one crowded rubric.
Jev bills input tokens and gives output away, and questions in one call run in parallel. Why the state is where the money is, and how to size a request before you ship it.
Retrieval is a sequence of bounded judgments. How a decision model slots into passage filtering, semantic find, reranking, and citation checks, and what stays in deterministic code.
Jev Cookbooks#retrieval#reranking#rag
Read article
Watch: Jev in action
Video walkthroughs and demos from the community.
YouTube: One terminal, three subscriptions, and Jev picking the model for Claude Code, Codex, and Cursor.
YouTube: A podcast segment on TypeSafe AI, its $40M raise, and the Jevons paradox behind the name.
YouTube: Gary Explains the catch: Jev understands natural language but answers with structured values and confidence.
Discussions about Jev
Threads from Reddit and Hacker News about building with Jev.
Hacker News: Show HN for the decide CLI: brew install, file context, and multi-question shell scripts.