What people are shipping with Jev.

https://github.com/vishesh-baghel/typesafe

@moritzkremb Less jaw-dropping but equally useful, is this native Postgres extension for Jev :) https://t.co/UJ1eRWtaGm

Dumping 50+ tool schemas wastes 90% context. Tool search fixes bloat, but costs an extra LLM round-trip (+2s) and fails on 7B models. tool-prune does it pre-flight in 0.4ms. h

@LakeAustinBlvd Jev has the potential of reducing a lot of "agentic" megaprojects to a bunch of ifs with curl, and I am totally here for it. Am experimenting with jevlike in https:

stupid questions are a pretty good model test playing with @TypeSafeAI (jev) and this is more fun than it should be napoleon's white horse → 87% white car wash 50m away 52% dri

Board: https://t.co/bQMPwlAInH Climber: @GravitateCRE Built by: @mattyp @poteto @roshan_s / @thursdayarena Jev: @typesafeai @CompleteSkeptic Also: @grok @xai

@typesafeai built an AI that can’t write you a single sentence but is also one of the most intelligent, cheapest, and fastest models ever. In Thinking, Fast and Slow, @kahneman_da

Continuing the legacy is-jeven: check if a number is even using @typesafeai Jev https://t.co/sKF7MMgu0K https://t.co/2hqSVF2kil

just found another cool project trying to make a jev like model but runs in your browser https://t.co/l6Pgymhgko

Jev from @typesafeai reframes MCP tool calling for anything customer-facing. Right now every tool-call is a full LLM round trip: 2-8s, a token bill, and a nonzero chance the model

diegosouzapw/OmniRoute · issue · open · 2 comments | Problem / Use Case TypeSafe AI (https://typesafe.ai) ships "System One" models. Its flagship, Jev, does not generate text. You

So I'm thinking spin up containers that run E2E tests using Jev and have each run recorded using @cap and add a "Human Review" status in Linear. Just need to get the sandbox envir

@spacedevin Not a product. Just someting cool I built using Jev https://t.co/k3UbpkkD85

@c4chaos @typesafeai Jev for MFK would be perfect: feed it three names, get instant calibrated probs for marry/fuck/kill, no rambling essays. Sub-second decisions, zero hallucinati

Oaklight/llm-rosetta · issue · open · 0 comments | Summary TypeSafe.ai ships a new class of AI model called Jev — a "System One" model that returns typed, probabilistic decisions i

Interesting behavior of Jev from @typesafeai. I wanted to see if it could use the probability field to answer numerical questions, but it usually can't! DiffusionGemma or modified

@typesafeai Since Jev can't synthesize text or numbers, extracting a value took a different trick: it doesn't generate an answer, it judges — word by word — whether each token in y

@moritzkremb Has anyone tried using Jev inside a speculative decoding pipeline — e.g. generate many candidate continuations cheaply, use Jev to rank them in parallel, and only have

https://t.co/3FN9fy008i openevals with jev cheaper and faster

@o_kwasniewski @typesafeai nice. the failure mode id watch for on agent-e2e is treating the clicker and the oracle as one model. keep actuation (playwright/agent) separate from ju