TypeSafe’s new Jev model is fast 🔥 and Agentgateway keeps up! In minutes, I had Jev running as a prompt guard through Agentgateway, with distributed tracing and cost tracking. Check out the example below Example: github.com/agentgateway/a… Intro to Jev: typesafe.ai/blog/introduci…
Evaluation & Training
Indirection
Questions about a property of a property, or with double negatives, cost accuracy; write instructions as directly as possible.
- Category
- Evaluation & Training
- Also known as
- —
- Related terms
- 4
- Directory entries
- 2
- Docs
- docs.typesafe.ai
- Added
- 2026-09-24
Definition
The jaggedness page groups double negatives and multi-hop questions as System Two tasks that Jev answers less reliably. Each hop away from the literal state costs accuracy.
The guardrail is to name the relevant parts of state explicitly with backticked paths, reduce hops, and ask the direct form of the question rather than a negated or nested one.
Tagslimitationsdesign
Related terms
Definitions that connect to this one.
JaggednessThe documented set of jev-1.13 failure modes — literal reading, counting, dates, indirection, context rot — each with a guardrail.Evaluation & TrainingBacktick path referenceNaming a state field inside instructions with backticked dot-and-index paths, such as `ticket.messages[0].text`.PrimitivesSystem Two taskWork that needs extended, multi-step reasoning; the docs treat it as a signal to decompose rather than a job for Jev.Core ConceptsLiteral readingJev answers the question you wrote, not the one you meant; scoping words, negations, and implied conditions are read at face value.Evaluation & Training
From the directory
A maintained list of jev-1.13's known failure modes, literal reading, unreliable counting, dates as text, indirection, context rot, and contradictory criteria, each with a guardrail.
Practices & PatternsDocs#official#docs#evaluation
Official
The three TypeSafe question types, the typed answers they return, how to choose between them, and how to ask several in a single call.
Practices & PatternsDocs#official#primitives#choice
Official
From the community
Posts from builders shipping with Jev right now.
Jev as an LLM guardrail inside Agentgateway
Jev reranks station suggestions in TrainLCD
提案駅のリランク判定と実測スパイクを追加する by TinyKitten · Pull Request #33 · TrainLCD/Functions github.com/TrainLCD/Funct… AIチャット機能の一部にもJevを潜ませてみるゾ
Vercel's fx safety reviewer, measured at 18× faster
@vercel 's fx uses Jev to review every command. @rauchg says it's up to 18x faster than GPT Luna and more accurate. @JoshARosen built Foreman, where Jev watches coding agents for drift, missed tests and getting stuck. @shantanugoel's Hermes agent installed an "ask Jev" skill on Show more
A game where you race Jev at spotting slop
A small game to show how fast TypeSafe's Jev actually is. Try to beat it at detecting a spam message beatjev2it.jlamberts86.workers.dev
Jev plays itself at 3D chess
got access to jev and instead of doing anything useful i made it play itself at 3D chess same model on both sides, slowly developing beef with itself you can watch jev vs jev or jump in and ruin its day chess-jev.loomens.com
@CompleteSkeptic
After co-inventing ChatGPT, I kept asking myself: why have superhuman chat models not led to AGI? I’ve spent the last 2 years in stealth building a new way to train models (RLCD), and a new type of frontier AI model that we are releasing today: Jev • 20-200x faster • 40-400x
dmx.to reshapes the timeline in plain language
the next version of dmx.to will let you fully customize the timeline both with dumb rules (how did i not think of these until now lol) and smart rules using Jev (BYOK also supports vercel gateway) 🤓 you can filter 3 lvls of usefulness, tweet types, topics, etc
