Official surface Choice · Score · Noul
TypeSafe API + official SDKs
TypeSafe
POST /v1/systemone. Python `typesafe-sdk`, JS `@typesafe-ai/sdk`, official agent skill. Pin `jev-1.13.0`.
Public builds · labeled
Where people are actually putting Jev
The 101 use-case lesson maps official TypeSafe themes. This page is the harvest of launch-week platforms and community projects. Every row has a URL. Speed and cost numbers belong to the person who posted them.
Third-party = a named post, changelog, or blog.Teaching = official surface restated for beginners. We do not invent logos, win rates, or “customers.”
Where you can call Jev
Waitlist is still the official path. Platforms below are labeled as platforms — their features are not TypeSafe defaults.
Third-party evaluate API
Vercel AI Gateway
Vercel changelog
Model `typesafe-ai/jev` through AI SDK 7 `evaluate`. Free-on-Gateway until Sep 25 was a Vercel Developers promo.
Third-party hosted System One
OpenRouter (beta)
OpenRouter + official handshake
Hosted path without a new TypeSafe vendor setup. OpenRouter: no JSON prompting, parsing, or validation layer.
Third-party env.AI.run
Cloudflare Workers AI
Cloudflare docs
Model `typesafe/jev`. Documented samples: department routing, refund review, account-risk Score + escalate Noul.
Third-party evaluate API
Vercel AI Gateway model card
Vercel
Hosted typesafe-ai/jev. Lists $0.042/1M input and max output tokens 0. Example uses type: boolean — Gateway field name, not TypeSafe’s noul.
Third-party hosted System One
AI/ML API
aimlapi.com
Hosted typesafe/jev via POST /v1/decisions. Lists $0.0577668 / 1M and 32K — platform card, not TypeSafe defaults.
Third-party hosted API
Venice API (beta)
Venice, quoted by TypeSafe
“It answers, it doesn’t write.” Send state and typed questions; get a probability, option, or score.
Agent harnesses & routers
The launch-week pattern: an LLM writes; Jev picks the model, the tool, or whether the tool is safe to run.
Third-party Choice + Noul
LangChain TypeSafeClassifier
LangChain (Sep 17)
`langchain-typesafe`: invoke state + questions, get classifications. Experimental ModelRouterMiddleware and AutoModeMiddleware.
Third-party Choice
Model router (community + LangChain)
Duncan / LangChain middleware
Jev chooses the cheapest model that can finish the task. Same official theme as TypeSafe’s harness map.
Third-party Choice
Computer-use / browser-use
aaron (@awlevin); LangChain names Browserbase
Jev picks the next OS or DOM action; a small LLM types only when needed. Builder speed/cost claims.
Third-party Choice fan-out
Jev Ultrafast (Browser Use)
browser-use/jev-ultrafast · via DEV.to / Awesome Jev
One request picks operation + DOM target; a small LLM types only when needed. Author: Zürich–London Google Flights in 7.1s / $0.0039. Builder claim.
Evals, review, guardrails
Builders put Jev on traces, PRs, and command reviewers. Numbers stay with the person who published them.
Third-party Noul + Score + Choice
Langfuse eval scoring
Langfuse (Sep 18)
Score traces with atomic Noul/Choice/Score questions (e.g. user disagreement). Write the probability back as a Langfuse score. Noul has no extra confidence field.
Third-party Noul / Choice
Vercel fx command safety reviewer
Guillermo Rauch, quoted officially
Every command reviewed. Luna today; Jev reported up to 18× faster (p95) and more accurate on that reviewer.
Third-party Choice
Vercel internal classifier eval
Malte Ubl, quoted officially
Replaced Gemini 2.5 Flash Lite on an existing classifier eval: saturated quality, 6× speed. One eval.
Third-party Score / Noul
Good Start Labs rubric bench
Relayed by Langfuse
6,003 rubric checks. Jev agreed with Claude Fable 5.1 on 91.5% of verdicts at a much lower quoted cost. Third-party bench.
Third-party review gate
Greenwash PR reviewer
Ayush Gupta on X
GitHub app that reviews PRs after CI, aimed at catching silent agent pass-through. Community project.
Product decisions
Inbox, news, search, and chat rooms — classify or score, then let product code act.
Third-party Noul + Choice
Moderated chat room
Mustapha (@stoufax)
Every message judged before the room sees it. Plain-English rules, no keyword lists; uncertain messages held for a human.
Third-party Score
Priority inbox scoring
Ian Nuttall
Classifies content and writes a “most likely to read” score from consumption history. Engagement lift not claimed as measured.
Third-party Choice
SlidesAI classifiers
Anurag Bhagsain
Reported ~20× faster on icons and ~60× on an image classifier, 98% latency drop. One product’s numbers.
Third-party Score fan-out
HN front-page rerank
Vishesh Baghel; Mukesh Utmani
Personal sliders over stories; another post claimed 2,000 typed judgments over 250 HN stories in 16 seconds for $0.0095. Community measurements.
Third-party Score + Choice
Crypto tweet sentiment → trade
brainstormity
Ingests live tweets per asset, scores bullish vs bearish, combines into BUY/SELL/HOLD. Open-sourced the terminal. Not investment advice, not a TypeSafe result.
Third-party Noul fan-out
Tab Bouncer
Manish Sharma
Browser extension: judge many open tabs in one call and close the unused ones. Builder latency/cost claims.
Third-party Noul / Choice
Podcast ad detection
@TTLequals0
Sep 20 post: using Jev to detect ads in podcasts. Builder demo; no accuracy table on the tweet.
Third-party Noul / Choice
Chrome X-feed filter
@suvsahoo05
Extension hides AI slop, repeats, and personal irrelevance. Author: probabilities → code → verdict.
Map-reduce over corpora
Many independent questions over rows or papers, then aggregate in code. Numbers stay with the person who posted them.
Third-party Choice
1kpapers classification
Hassan El Mghari · 1kpapers.com
1,018 paper summaries classified into 24 topics. Author: summaries $3.99, Jev $0.08, median 256 ms. He says he is still eval-ing before replacing the old labels.
Games & real-time loops
Company Doom / Wikiracing demos plus community games. Fun proofs of latency — not product case studies.
Third-party Choice
jev-trader
jarrodwatts
One buy/sell per Monad block on Kuru MON-USDC. Author: ~81 ms model latency, dry-run mode. Not investment advice.
Third-party Choice + Score + Noul
jev-drone
RomanSlack
Tactical judgment at ~2.5 Hz. README line relayed by DEV.to: Jev cannot be the perception layer and cannot run at control rate. Advisory only.
Third-party Choice over legal moves
jev-chess
qibinlou · via HN
Play White; Jev picks among legal Black moves through Vercel evaluate. Author: ~300 ms, ~$0.00004/move. One demo.
Third-party Score
Wordshift typing racer
Marcel Pociot
Semantic typing game: the bigger the meaning gap, the farther the car moves. Community real-time demo.
Third-party Choice
Tetris / Snake / Doom-class games
Marcus Lowe; Raihan Khan; official Doom
Community Tetris and two-Jev Snake plus TypeSafe’s own Doom demo. Proof that ~100 ms decisions fit a game loop.
Independent tutorials
Other 101s on the public web. We map them so jev.page stays complete; they are not TypeSafe and they are not this site.
Official surface interface clone
openjev
TheoLeeCJ
Reads typed option probabilities off a 4B open model’s logits. Explicit: reproduces the interface pattern, not Jev’s model or training.
Official surface essay
Flavio Copes · Jev deep dive
Flavio Copes
Independent long-form 101 (Sep 17). Author: console access, not in production. Early-demo survey stays labeled.
Third-party HTTP API
jevaiguide.com API notes
jevaiguide.com
Live-call log Sep 19: pin jev-1.13.0; jev-1.13 is unknown; error table. Their tests, not TypeSafe’s.
Official surface first call
jevtypesafeai.com how-to-use
jevtypesafeai.com
Independent first-call curl page. Overlaps official quickstart. Prefer docs.typesafe.ai when they disagree.
Official surface essay
DEV.to · How to Use Jev (Valyu)
valyuai
Independent practical 101 plus a first-48-hours survey. Builder numbers stay labeled.
Third-party press
InfoWorld launch writeup
Anirban Ghoshal
Secondary press restating TypeSafe’s launch positioning and list price. Not a TypeSafe table.
Official surface essay
Refix · Python tutorial
Neil Agarwal
Sync vs async TypeSafeClient. Side effects stay in Python. Teaching page, not TypeSafe.
Official surface essay
Refix · TypeScript quickstart
Neil Agarwal
Inferred answer types and confidence-aware routing in TS. Teaching page.
Official surface first call
MrJev getting-started
mrjev.com
Independent first-call (Playground, both SDKs, pin the dotted model ID). Prefer official docs on conflict.
Third-party index
Jevable catalog
jevable.com
Launch-week demo wall with builder quotes and view counts. Catalog, not a TypeSafe scoreboard.
How to read this as a beginner
- Start with one official decision slot from the 101 — a ticket, a gate, a score.
- Copy the shape of a community build (router, reviewer, rerank), not their metrics.
- Threshold in your code. A typed answer can still be wrong.
- Keep an LLM for writing. Put Jev in the decision.
Back to the guide