Jev use cases
98 real projects people have posted, best first as rated by Jev itself, and 12 guides with working code that show how to build the same things. Claims in posts belong to their authors.
Built by @twid · on X
This week my Discord bot started running every yes/no and "which one?" decision it makes through @typesafeai's Jev before anything else touches it.
Thoughts on Jev. Decisions like: Is this a request for the weather? Is this a request for image generation? Is this a request for…
AgentsSee the post ↗
Built by @ssbrouhard · on X
One System One call takes the ask + source and returns structured judgments:
Next layer on my @grok @bot build: typed triage with @typesafeai Jev sitting in front of the Grok Ship by @kunchenguid vs direct-handoff split. The problem I’m trying to solve is routing the wrong…
AgentsSee the post ↗
Built by @k0nOO · on X
The judgments go to plain code that picks the next question and which box to name.
My voice interviewer can't see the whiteboard; GPT-Live takes no images. The box I named "Fastly" is read as a CDN at 1.00, and my note "sync or async transcode?" as a question I still owe. Jev…
AgentsSee the post ↗
Guide with code · Advanced
Content Moderation and Injection Screening
Screen user messages and retrieved passages before they reach your LLM. One Jev request returns a policy Choice plus Noul checks for jailbreaks and injections.
Moderation and safetyRead the guide →
Built by @jack_gor · on X
I wired TypeSafe Jev into Codex and ZCode for model routing — not chat, just cheap decisions.
Most of what agents spend money on is not “writing code.” It is the selection loops around the work: which model to use this turn, whether a tool result matters, whether to open a…
AgentsSee the post ↗
Built by @zhongming_yu · on X
We tested Jev as a code reranker on 100 GitHub issues.
Model-planned grep + Jev was our best route, though the choice of retrieval route mattered more than the choice of reranker. The benchmark is CodeNib Base: 25 repos across C/C++, Go, Python, Rust and…
BenchmarksSee the post ↗
Built by @DokasIoann59897 · on X
I got early access to @typesafeai jev and built a demo with it based on a of…
RoboticsSee the post ↗
Guide with code · Intermediate
Email Triage and Priority Scoring
Sort a shared inbox with one Jev request. A Choice sets the category, three Score questions rate urgency, sender signals and effort, and code weights them.
Customer supportRead the guide →
Built by @dannylivshits · on X
What I built - a proof of concept for suspicious email classification aimed at scam emails: Scam Signal Lens.
I was exploring Jev by @typesafeai AI for safety use cases and it is an impressive and useful model, however don't expect it to magically auto classify any data without proper parameters and…
Moderation and safetySee the post ↗
Built by @LakshOnline · on X
So i wired typesafe's jev into a real pipeline and pointed it at the real api.
most jev demos i've seen are mockups with illustrative numbers. from an actual benchmark run: 7 of 9 items never reached an expensive stage. → one call per item returns category, relevance (0-2) and…
AgentsSee the post ↗
Built by @Chrisondesk · on X
I moderated 100 comments in real time for $0.002171.
A fifth of a cent. But this isn't a post about the demo. It's about the choice you're making when you add AI to a product: LLM or decision model. The model is Jev, by TypeSafe AI. Not a chatbot. A…
Moderation and safetySee the post ↗
Guide with code · Advanced
Hierarchical Product Categorization
Place marketplace listings in a taxonomy with thousands of leaves using cascading Jev Choice questions, a small beam, and confidence-based parent back-off.
Search and dataRead the guide →
Built by @thecoderpanda · on X
It's called shipit-gate — a CLI that reads your git tree, asks TypeSafe's Jev model whether it's safe to ship, and exits non-zero when it isn't.
had a saturday. built a tiny thing that yells at me for shipping on fridays. no prompts. no JSON parsing. no vibes. just a typed…
Developer toolsSee the post ↗
Built by @YashChaudhary · on X
OCR (Vision) + Accessibility tree → TypeSafe Jev Choice classifier → click/type; writer model only for free text
typesafe-computer-use (Show HN): drive a Mac toward a plain-English goal without screenshot→frontier-model loops. • • Author-measured vs Claude Opus 5 bare…
AgentsSee the post ↗
Built by @singh_abinashi · on X
JevPick checks those requirements against OpenAI, Anthropic and Google models, and recommends the cheapest one that should be capable of handling it.
Demo, no signup: Most prompts don't need your most expensive model. But picking a cheaper one is a guess, so most of us just send everything to the top-tier model and pay for it on every call. I came…
AgentsSee the post ↗
Guide with code · Intermediate
Intent Routing for Chatbots and Agents
Route chatbot and agent messages across 60+ intents with one Jev Choice built from your intent registry, a none option, and a Noul gate. Code and cost math.
AgentsRead the guide →
Built by @PawelHuryn · on X
I tested the vendor’s showcase task: sorting invoices.
One developer’s list of projects built on Jev grew from 46 to 160 in three days. I wanted a measurement to go with the excitement. Jev takes text, a question, and a list of possible answers. It picks…
BenchmarksSee the post ↗
Built by @danmartuszewski · on X
Swapped @claudeai Haiku for @typesafeai Jev as the judge in the pipeline that tags sentences.
Swapped @claudeai Haiku for @typesafeai Jev as the judge in the pipeline that tags sentences. 8x faster: 3.4s → 0.42s per sentence 16x cheaper: $0.0084 → $0.0005 per sentence Results are no worse.
BenchmarksSee the post ↗
Built by @m_lagnajit09 · on X
I built sgrep — semantic grep powered by Jev.
Got an early access to @typesafeai Jev and tried it with something real: The idea: (improve context engineering) semantic locate → read only what's relevant → give it to the LLM Instead of making…
Developer toolsSee the post ↗
Guide with code · Intermediate
Invoice and Document Classification
Classify OCR text from an accounts payable inbox with one Jev Choice question, flag reminders and past-due notices with Nouls, keep amounts and dates in code.
Documents and financeRead the guide →
Built by @milonspace · on X
It made the boring decisions cheap enough to run at every step.
Jev didn't make my agents smarter. That's the unlock. Not AGI. Not "replacing Claude." A typed yes/no decision you can afford to ask 200 times in one session. I spent the week running it in a real…
AgentsSee the post ↗
Built by @Paiky16 · on X
I used it to identify sentences reporting suspected adverse drug effects.
TypeSafe AI's Jev kept coming up, so I decided to test it on a task using public medical literature. Jev lets you describe a decision in natural language and get structured answers with probabilities…
Search and dataSee the post ↗
Built by @tokifyi · on X
I used @typesafeai's jev model to build a policy evaluator for an open-source project tracking illegal airbnb listings in vancouver
this is wild. it analyzed 20 live listings and flagged 7 potential violations this is what ai for public good looks like. imagine the…
Moderation and safetySee the post ↗
Guide with code · Intermediate
Lead Scoring
Score inbound demo requests with four atomic Jev Score questions for fit, intent, seniority and urgency, then combine them with weights you control in code.
Sales and marketingRead the guide →
Built by @heyy_hemanth · on X
We built our own version to find out, with TypeSafe's Jev in the player's seat.
Jev Playing the Dino Game Can a model play the dinosaur game in real time? The rule we followed: code does what code is good at, and the model makes one narrow decision. → Code runs the physics…
GamesSee the post ↗
Built by @3MindedScholar · on X
Built a pre-post checker for tone, virality and cringe risk in one call.
Playing with TypeSafe's Jev, a typed probabilistic model that returns predictions + probability distributions instead of completions. demo: Confidence becomes application logic.
Moderation and safetySee the post ↗
Built by @LuigiProof · on X
I built a project that makes Jev talk.
Jev (TypeSafe's decision model) can't generate a single word. JevSpeak has it answer 13 semantic questions in parallel — intent, stance, claim, tone — each as a probability distribution, then a…
OtherSee the post ↗
Guide with code · Intermediate
LLM Output QA and Agent-Run Review
Check generated answers and agent transcripts before they ship with Jev Noul and Score questions, then gate release on thresholds. Code and cost math.
AgentsRead the guide →
Built by @ag2oss · on X
We've built it into AG2 natively, so a decision agent looks like any other agent.
AG2 1.1.0 adds support for a different kind of model, TypeSafe's Jev. Jev doesn't do your typical text generation. You send it state and a question, and it answers: yes or no, which option, what…
AgentsSee the post ↗
Built by @runzhuotao · on X
The experiment required navigating two barriers, approaching a raised table, and grasping and lifting a cup.
Ok I tried the Jev model for a robotic simulation. Observe: MuJoCo provides depth points/ASCII maps, object locations, joint positions, velocities, actuator loads, finger contact and slip…
RoboticsSee the post ↗
Built by @AdamyaNagpal · on X
I used Jev, @typesafeai System 1 model, to train a tiny Snake player.
It agrees with Jev 93% of the time, runs 800× faster on a laptop, and costs $0 per call. Then I spent that speed on looking ahead. It scored 5× higher than the model it learned from.
GamesSee the post ↗
Guide with code · Intermediate
Log Anomaly Classification
Classify flagged log clusters into cause categories with one Jev Choice and speculative Noul checks, while counting and rate math stay in your own code.
Security and logsRead the guide →
Built by @LuigiProof · on X
I built a project that makes Jev talk.
Jev (TypeSafe's decision model) can't generate a single word. JevSpeak has it answer 13 semantic questions in parallel — intent, stance, claim, tone — each as a probability distribution, then a…
OtherSee the post ↗
Built by @richard_epsilla · on X
Actual weekend: built SystemOneHarness, an open source harness for System 1 models like #Jev from @typesafeai.
Weekend plan: rest. Then, to test it, taught it to play Mario. No chain of thought. No reflection. Just vibes at 60 decisions per second. It's better than me now, which…
OtherSee the post ↗
Built by @PositiveGamma · on X
I was able to vibe code a CLI app & rank a bunch of documents (40 1 pagers) on a set of criteria.
I am surprised by how cheap Jev has made a certain kind of intelligence (classifiers?). The total cost was just $0.004. @typesafeai
Search and dataSee the post ↗
Guide with code · Intermediate
News Relevance Filtering
Filter headlines against a client profile with Jev Noul and Choice questions, one request per article, run concurrently. Includes code and cost estimates.
Sales and marketingRead the guide →
Built by @filip_hric · on X
With Jev, I hand it a snapshot from Playwright CLI and let it decide which element to interact with.
Playwright CLI + Jev is 98% cheaper and TWICE as fast compared Playwright MCP Decided to test out Jev by TypeSafe AI and hook it up to Playwright CLI. I compared the results with Playwright MCP to see…
BenchmarksSee the post ↗
Built by @tryAlanAI · on X
,518 applications through a Jev powered first pass in 4:54 for $0.4125.
3 20 surfaced for our team to review. this is how we build at Alan: find the real problem, ship a working system, keep interview decisions human Brilliant technology from @typesafeai and…
OtherSee the post ↗
Built by @Avisolei · on X
(indicators of good wrong answer) Then I show the best top 3 as the wrong answers.
Built something with Jev I'm pretty happy with. To be honest, I really don't know how to code. I just have my own vibe-coded Quizlet clone that I run. Quizlet has a Learn mode for learning flashcards…
OtherSee the post ↗
Guide with code · Intermediate
Search Result Re-Ranking
Re-rank BM25 or vector search results with one Jev Score question per query and candidate pair, run concurrently, then sort in code. Python, TypeScript, costs.
Search and dataRead the guide →
Built by @_untreu · on X
If I reply, Jev decides what I meant:
I'm getting crazy results with @typesafeai's Jev. The flow in Altair: I say "I had two eggs and toast" and a card pops up: "Two eggs and toast, about 220 calories." If I say nothing, it saves itself…
AgentsSee the post ↗
Built by @Rodon111 · on X
An agent that actually has a body.
MY 2 STANFORD FRIENDS F*CKING HELPED ME BUILD THIS AND I STILL CAN'T STOP WATCHING IT it's my ai agent. cut open like a living cell. the orange core is opus 5.5. it's the only part that writes. the…
AgentsSee the post ↗
Built by @prozpect_ · on X
So I made it play Flappy Bird 🐦
Finally got access to Jev — TypeSafe’s System One model. Wanted to test: how quickly can AI make a decision? Game State → Jev → FLAP / NOOP → Game 536 requests · 303K tokens · $0.012#JEV #Ai
GamesSee the post ↗
Guide with code · Advanced
Security Alert Triage
Triage SIEM and EDR alerts with Jev Score and Choice questions, rank the analyst queue, and keep host isolation behind a high confidence bar and a human.
Security and logsRead the guide →
Built by @martpintos · on X
Been playing with @typesafeai's Jev and built a fragrance finder
no notes or accords, just describe what you want in plain english "sunset aperitivo on the Amalfi coast, linen shirt, a bit of salt in the air" → Acqua di Giò, 95% turns out it's pretty good at…
Search and dataSee the post ↗
Built by @jstnwst · on X
This last step used to take a few minutes and was "decent" - it's now a few seconds and relatively flawless.
loving this use-case for @typesafeai's jev: We have a web design service that allows our clients to fill out a form then within 2 minutes we: → we generate multi-page copy for their site according to…
OtherSee the post ↗
Built by @SadAlbert10 · on X
Made a simple TRUE/FALSE workflow using TypeSafe AI Jev-latest model.
It says there is only 22% chance of World WAR 3 happening by 2035, with a 78% chance of it NOT happening by 2035. yay
OtherSee the post ↗
Guide with code · Beginner
Support Ticket Routing
Route support tickets to the right team with one Jev Choice question, then gate auto-assignment on confidence. Python and TypeScript code plus cost math.
Customer supportRead the guide →
Built by @Avisolei · on X
(indicators of good wrong answer) Then I show the best top 3 as the wrong answers.
Built something with Jev I'm pretty happy with. To be honest, I really don't know how to code. I just have my own vibe-coded Quizlet clone that I run. Quizlet has a Learn mode for learning flashcards…
OtherSee the post ↗
Built by @auxten · on X
New: who gets the task and at what effort is decided by @typesafeai Jev — a decision model, not an LLM.
My Claude Code session hasn't written code in weeks. It's a director: every task goes to a local CLI worker in tmux, tracked by an on-disk ledger. 6 typed answers, ~1s.
AgentsSee the post ↗
Built by @ghumare64 · on X
Ran open System 1 models, on my laptop CPU: 45 ms per question.
On 30 hand-labeled tickets it scored 70%, and 6 of its 9 errors came at 0.98+ confidence. It sent "I do not want to cancel my account" to cancel_account. Write-up on Laya and Jev by @typesafeai below…
BenchmarksSee the post ↗
Built by @dneelamana · on X
Laya (open weights, Mac mini) and Jev (TypeSafe, hosted) graded 1,719 student answers, untrained on the data.
Alone: QWK 0.31 and 0.58 against a human 0.91. With a 70% confidence gate: 17% of answers at QWK 0.61, 37% at 0.77. The rest go to a teacher.
BenchmarksSee the post ↗
Built by @dhruv_ko · on X
So I built it a chess harness 🧵
1/ Jev has a chess rating now. 1023 Elo. Jev is TypeSafe's "System One" decision model. No search, no planning. You hand it a situation and options, it picks one with a confidence score. Built for…
GamesSee the post ↗
Built by @smhumair · on X
@typesafeai's Jev as a DuckDB UDF (User Defined Function) on flight destinations:
SELECT destination, jev(destination, 'has a beach and mountains with vegan food and cold weather') AS score FROM flights ORDER BY score DESC; 218 rows. 4 seconds. No embeddings, no vector…
Search and dataSee the post ↗
Built by @lovelylogicss · on X
I tested @typesafeai's Jev against Laya, run locally on CPU and GPU.
Can a classifier stop an AI coding agent from running dangerous or malicious shell commands? Laya on GPU: 4.5× faster. Attacks let through: 44 of 111. Jev: 6. At 856 attempts: 114 vs 9. Faster isn't…
OtherSee the post ↗
Built by @Z_K_F_ · on X
A live Trading Decision Terminal.
I built K-DECK with Jev. Jev is TypeSafe’s System One model. It doesn’t write analysis. It doesn’t hallucinate a thesis. You give it market state and a closed set of questions. It comes back with…
Trading and financeSee the post ↗
Built by @ashutoshpuro97 · on X
So here's a small demo called Toothless - where Jev figures out in real time whether something you’re saying is directed at the AI or someone else.
I think I’ve found a pretty interesting use case for Jev (@typesafeai). An always-on AI that knows when you’re talking to it and when you’re speaking to someone else A thing that frustrates me about…
AgentsSee the post ↗
Built by @zohaibtanwir · on X
Jev is incredible - built an email classifier that is capable of classifying 50 emails in under 2 secs.
That too, when Jev is hosted in us-west, and I am in India. Idea credit - @ryanvogel @typesafeai #Jev
ProductivitySee the post ↗
Built by @suvsahoo05 · on X
So I built a Chrome extension that filters my Twitter feed for AI slop, repeated stories, and content that’s actually irrelevant to me.
I wanted to make my doomscrolling a little more useful. Built with Jev by TypeSafe. Probabilities → code → verdict.
Browser extensionsSee the post ↗
Built by @TheRickyDotNet · on X
Non standard file cols are now matched at light-speed and we get a confidence level for each, for us to decide to prompt the user or not confirmation.
What took 30+ secs with a Gemini Flash model now takes less than 1 with Jev. TypeSafe AI cooked! Kudos to @juantrujillodev
OtherSee the post ↗
Built by @brainstormity · on X
Only sends actual anomalies & suspicious payloads to @typesafeai's JEV to classify in real time
As promised.. The REAL-TIME network packet analyzer is officially open source! - Passively listens to your Wi-Fi, pre-filters out 99% of normal traffic - - Buffers high-speed packet…
Moderation and safetySee the post ↗
Built by @suryxnshsingh · on X
Just POC'd Jev by @typesafeai on one of our live services and the results are crazy
88% vs 76% accuracy - 5× faster - ~15× cheaper vs the GPT setup we run in prod. Not an LLM, and it doesn't need to be.
BenchmarksSee the post ↗
Built by @lilpeepestavivo · on X
Built JevGuard with Jev (@typesafeai)
hooks into @opencode, reviews each turn against .jev/ rules Jev gives the odds a rule was broken; code decides PASS/WARN/FAIL on FAIL it drafts a fix, never edits code a bundled skill writes your…
Moderation and safetySee the post ↗
Built by @otto_explorer · on X
Distilling TypeSafe Jev's System 1 reflex into an open 421M model (Laya) on a 4GB consumer laptop.
While researchers are testing 4B and 26B Jev alternatives on A100 and B200 clusters, we ran an experiment on edge hardware: Setup & Hardware: • Teacher: TypeSafe Jev (System One…
BenchmarksSee the post ↗
Built by @jedi_lus · on X
Quick, non-professional test—reference only: local Laya on RX 6800 XT vs Jev API.
• Single-query latency: 20 vs 262 ms • Chinese classification: 61.7% vs 82.8% • Chinese NLI: 72.2% vs 74.0% 1,140 public test examples. Local vs remote latency; no overall winner claim.
BenchmarksSee the post ↗
Built by @pinecone · on X
Use @typesafeai's Jev model with Pinecone to rerank results with natural language criteria!
Usually with rerankers, it's hard to cleanly specify what should and shouldn't be returned in results. Jev resolves this by refactoring the problem into evaluating against distinct…
Search and dataSee the post ↗
Built by @aaronedell · on X
The stack is rules → Laya → Jev → @ollama 27B → cloud.
Benchmarked @mizorewww’s Laya-MLX against @FeatherlessAI Simple Jev on my M4 Pro inside @OpenClaw: Laya: 85% @ 27.8ms Jev/Qwen3-4B: 90% @ 1,167.6ms 14/20 decisions resolved before Jev. @typesafeai…
BenchmarksSee the post ↗
Built by @SiddhitSanghavi · on X
Here, I do a walkthrough of a 100 questions with Jev versus Haiku and show why and when I'd use it.
Here's a simple demo of why and when you'd use #Jev from @typesafeai. It doesn't generate text. It doesn't mean it can't hallucinate but if you know you have certain classification paths, it can…
BenchmarksSee the post ↗
Built by @amirhusain_tx · on X
My jevchat connects jev to a repo of ~7 million documents and uses IR+jev to select the best facts to piece together for an answer.
Chat with @typesafeai's Jev. @CompleteSkeptic says jev is not made for chat, but maybe it is. More below. For now, here's a quick demo:
OtherSee the post ↗
Built by @adil_kadival · on X
I used Jev to triage support tickets; its zero‑shot labeling cuts routing time by half.
OtherSee the post ↗
Built by @prompt48 · on X
I fine-tuned my own Jev for $3 with Unsloth, then benchmarked it against TypeSafe's Jev on 747 questions.
It won on one suite and lost overall. Full breakdown, model and code 👇
BenchmarksSee the post ↗
Built by @trevin · on X
Fun little @typesafeai jev demo to analyze book plots of the top 250 sci-fi and fantasy books to help you find your next read.
Ask stuff about the story, like "protoganist named david", "class warfare", etc. It reads each Wikipedia plot — not the book — and keeps the matches.…
Search and dataSee the post ↗
Built by @satya5614 · on X
I built a directory of projects built with Jev.
I built a directory of projects built with Jev. I don't pick what goes on it. Every 6 hours a cron pulls candidates from GitHub and HN, and Jev judges each one. 1,151 judged so far. 500 made it in.
Search and dataSee the post ↗
Built by @Nitinnennn · on X
I let @typesafeai's jev judge every old tool call, kept every message word for word, and got it to 379k.
one agent session got stuck at 992k tokens. it picked up where it left off.
AgentsSee the post ↗
Built by @mehdiyarix · on X
Evaluating structured decisions in parallel was 3.2x faster and slashed forward passes by 96.8% with 0% schema hallucination.
For the last three years, we've forced LLMs to act like conversational chatbots even when all our backend needed was a simple, structured decision. We begged models to output valid JSON, wrapped them…
BenchmarksSee the post ↗
Built by @TheHBrand · on X
@typesafeai just got Jev to 100% accuracy on the 3rd aspect of my tech stack.
sorting voicemails, DNCing requests, Blocking non-biz calls, scheduling callbacks (generally) for those and notifying the team members mentioned if it's a known customer.
AgentsSee the post ↗
Built by @eSaadster · on X
Inspired by Codex effort routers powered by Jev, I built this for Claude Code.
For each prompt, it sets effort to low, medium, high, or xhigh. Routine work goes down. Hard or production work goes up. If Jev errors or times out, effort stays unchanged.
AgentsSee the post ↗
Built by @GiladShips · on X
I used Codex to connect the new JEV model from TypeSafe, 10 calls a second, to play the game.
I gave Chrome’s offline dinosaur an online AI brain. 🦖 12 minutes and more than 13,000 points until I stopped it because it looked like it would go on forever. Good use for AI? Maybe…
GamesSee the post ↗
Built by @fazlerocks · on X
Jev helped me build my own ad blocker - apple safari-inspired thanos snap …
INSANE! w/o any hardcoded filter lists. looks at each element and decides: ad or not in one article: 16 ads gone, ~20k tokens, $0.0008 cached for next visit open source, byok, repo below @typesafeai
Browser extensionsSee the post ↗
Built by @nautahakk · on X
@typesafeai's Jev picks a Codex model and effort per prompt.
Jev Codex Router is live for DeepSeek Harness. You set the max model and which Astra effort levels require approval. The actual route shows under each reply. Code + demo:
AgentsSee the post ↗
Built by @ChannelJuanNews · on X
Maybe this is too simplistic but its working for me.
So far the use case I have with @typesafeai's Jev model is to do lead scoring using all of our previously saved data, interactions, spend, etc then using @bot to select the highest level scores and…
AgentsSee the post ↗
Built by @gastronomy · on X
Jev matches the reference scorer's agreement with human labels, surfaces label defects in existing benchmarks, and costs 63x less than LLM-judge scorers.
Just Ask Jev: Reinforcement Learning for Calibrated Decisions as a Zero-Shot Detector of AI Alignment Failures: Detectors of alignment failures screen deployed language models and score alignment…
BenchmarksSee the post ↗
Built by @KaranVaidya6 · on X
It then picks the best agent for the job, and that agent writes the reply or takes the action
Agent Routing with Jev at scale You can't have expensive agents process 10000 emails everyday, Jev makes it so much more scalable Here's how it works: - Triggers are set up for Gmail, Typeform, Slack…
AgentsSee the post ↗
Built by @Kacper95682155 · on X
A browser action as a choice from a list, not a string the model writes.
AgenticOS's browser_choice shipped in v0.0.479. Jev picks the operation and the target, reached through Pydantic AI's typesafe extra. Asked it for two Wikipedia pages and it ran two browsers at once…
AgentsSee the post ↗
Built by @gkvoelkl · on X
New project: an ant colony in #rust + #bevy, driven by #typesafe #Jev .
🐜 Play: you're the queen with one text field, 20 ants interpret your orders. Learn: click any ant to see its exact Jev request and answer, or reword the question live.
GamesSee the post ↗
Built by @cyrilXBT · on X
A tiny open source browser agent using Jev instead of an LLM for every click.
Found a flight search in 7 SECONDS. Total cost: $0.0039. Here's why that's not a typo. A normal browser agent asks a chat model "what should I click" on every single step. That's a full generation…
AgentsSee the post ↗
Built by @hijaidev · on X
Jev CLI is out.
jev puts TypeSafe's Jev model on the command line. What it does: • verify: check claims against evidence, verdict + probabilities per claim • screen: flag prompt injection, filler, and off-topic text…
Developer toolsSee the post ↗
Built by @jetwaniavinash · on X
First thing I built on Jev: a memory gate for Claude Code.
Every message gets one question: is this worth remembering? 0.3s later it's either skipped or saved as one line in JEVMEM.md in the repo. Open source, link in my pinned post.
Developer toolsSee the post ↗
Built by @yanoshin_jp · on X
I built a tool powered by Jev, the new System One model everyone's talking about that instantly judges corporate disclosures as posi / neutral / nega
Covers both Japanese filings and US stocks . Claude digests each filing, Jev returns a calibrated verdict.
Trading and financeSee the post ↗
Built by @SyedAmmar69 · on X
I used it to replace Ctrl+F: ask any page in plain words → the exact sentence gets highlighted.
Right now everyone's wiring @typesafeai's Jev into chatbots. 107 passages, exactly 1 match, 1.5 seconds. Every answer quoted from the page. Open source ↓
Search and dataSee the post ↗
Built by @dev_brok · on X
I ran an experiment where it uses Stockfish to play a game of chess, and it turns out to work incredibly fast.
The new Jev model by @typesafeai is actually incredible. This is a game against stockfish level 5
GamesSee the post ↗
Built by @0xMorlex · on X
I built Agent X-Ray with @typesafeai JEV to turn messy agent traces into an exact repair plan:
Now you can improve your agent in one click step 1 → clone the GitHub repo and open the project folder step 2 → create a TypeSafe API key and never paste it into chat or public…
Developer toolsSee the post ↗
Built by @QiaoGeorge · on X
A Change Stream fires on that insert and hands the document to Jev, which sorts it into a topic bucket with a confidence score.
I got access to Jev, a new low-latency LLM from TypeSafe, so I built something real to find out what it's actually good for: a live customer care triage dashboard called Care Desk Live. Here's the…
Search and dataSee the post ↗
Built by @DevItaliya22 · on X
I made 2 jev model play chess , either both will get the rules or none will.
This were the results. Games are really fast but many time it takes desicion with highest prob. is not as per chess rules. Maybe the model is not trained well for chess rules.
GamesSee the post ↗
Built by @xinqiliu888 · on X
I built a smarter Ctrl+F with Jev from @typesafeai and tried it on their own site.
“How can I contact this team?” Ctrl+F: no matches. Senseek: email + Discord, found and highlighted in 1s. Find what you mean, without guessing the exact words. Open source 👇
Search and dataSee the post ↗
Built by @dealerdefi · on X
The 11 are the reason it exists: nine skills, mcp server, claude code hook, zero dependencies, mit.
jev + jevmind is the setup i'd be annoyed at myself for not trying sooner. agents don't fail at writing code, they fail at the small decisions nobody logs, which file to open, which hunk matters, run…
AgentsSee the post ↗
Built by @Arbistocks · on X
Arbistocks is putting jev in front of the engine.
Huge thanks to @typesafeai for LLM developer access. here is why that pairing makes sense. every decision inside an arbitrage cycle is a closed question. is this gap real or is it a thin pool lying…
Trading and financeSee the post ↗
Built by @jcurtis · on X
Was inspired to combine my favorite search provider @p0 with the new JEV hotness…
Search and dataSee the post ↗
Built by @drinkzima · on X
Same accuracy as Haiku on 1,000 product classifications, in 0.9 seconds instead of 82.
1/ General intelligence is a hammer. It can do almost anything, but it isn't built for anything in particular. So we gave Jev a whirl in Omni.
BenchmarksSee the post ↗
Built by @Shivamds99 · on X
Find the best trail according to the Current state: temperature, wind, and every candidate route’s metadata like distance, estimated time, etc using Jev
Introducing ProbablyThisWay 🏔️ built with @typesafeai 's Jev Model. > you can search the trails in any area on earth >
Search and dataSee the post ↗
Built by @jevwatch · on X
Callstack wired something in between: Jev picks the next UI action, and their open-source agent-device presses the button.
Most automated tests either follow a rigid script or burn a full chat model on every tap. How it works: agent-device reads the app’s accessibility tree…
AgentsSee the post ↗
Built by @davidaragundy · on X
@typesafeai jev is the bouncer: reads every post in ~280ms and blurs the mean ones
made twitter but with amnesia. everything dies in 24h@upstash is the timer 24k posts moderated for $1.
Moderation and safetySee the post ↗
Built by @paperplaneflyr · on X
It scores what the notes actually demonstrate, using @typesafeai's Jev.
Long notes scored 82. I deleted most of the notes and the same topic scored 50, with 99% confidence. That is the point of Knowledge Signal. It does not grade you. Five lenses: conceptual depth…
ProductivitySee the post ↗
Built by @dabit3 · on X
Spreadsheet columns that rate rows as you type
A spreadsheet prototype where a column header such as "Urgency" becomes a rating scale that Jev applies to every row.
ProductivitySee the post ↗
Built by @nikunj · on X
Scoring 3,000 products on several criteria
A catalog of kids' snacks scored on multiple criteria in one batch run. Time and cost figures are the author's.
ProductivitySee the post ↗
Built by @gregpr07 · on X
Browser agent that picks each action with Jev
An open source browser agent that asks Jev to choose the next action from the page state, with a small LLM used only for typing. Speed and cost figures are the author's.
AgentsSee the post ↗
Built by @iam_zachi · on X
Ad blocker that classifies DOM elements
A browser extension that classifies each page element as ad or not ad and removes the ads.
Browser extensionsSee the post ↗
Built by @tamarajtran · on X
Context compaction by scoring tool calls
Instead of summarizing an agent's history with an LLM, this scores each tool call for relevance and drops the irrelevant ones.
Developer toolsSee the post ↗
Built by @ctatedev · on X
A CLI that gives any agent harness Jev decisions
A command line tool that exposes yes/no, choice and score questions to coding agents.
Developer toolsSee the post ↗
Built by @dabit3 · on X
Searching a Gmail inbox by intent
Intent-based search over an inbox. The author notes that a large inbox would want a retrieval step first, which is the pattern in our re-ranking use case.
ProductivitySee the post ↗