Explore the Jev ecosystem.
Learn, build, and play with Jev — from deep technical guides to creative experiments.
New browser game Play Jev Dispatch Route invented support messages with Jev’s Choice, Noul, and Score answers, and see what Jev can decide in your app. PlayVideo guides
Learn it on screen.
New videos
by Instantly
Instantly shows how Jev can check whether companies fit a sales team’s criteria, recognize buying signals and sort contact data. The presenter reports scoring 1,000 businesses in 6.8 seconds for about $0.045; that demonstration is not an independent performance test.
Duration: 21 minutes 22 seconds.
by AnassKartit
Anass Kartit compares sixteen hosted decision models and four local setups on his own 82-question test. Perplexity’s Decider leads his results; Jev scores 79 out of 82 and Microsoft Decision-1 scores 75. He also examines malicious instructions, reordered choices and differences between local runtimes.
Duration: 5 minutes 51 seconds.
by Derek Cheung | AI Agents Automation
Derek Cheung demonstrates how to build an automated speed-to-lead system in n8n using Jev and OpenRouter. The workflow qualifies leads by analyzing text derived from user-submitted photos and voice notes, routing qualified prospects to receive generated media while pausing uncertain cases for human review in Slack.
Duration: 12 minutes 56 seconds.
by Simon Scrapes
Simon Scrapes demonstrates how to combine Jev's rapid classification capabilities with Claude's reasoning to build efficient browser agents, highlighting both the architectural benefits and common failure points like iframes and false positives.
Duration: 12 minutes 2 seconds.
by MEGA DEV
John Lindquist demonstrates using Jev to turn messy inputs into typed decisions for agent workflows, highlighting its speed, cost-efficiency, and limitations like menu size constraints.
Duration: 94 minutes 21 seconds.
by Tech With Tim
Tech With Tim evaluates the Jev decision model against traditional LLMs across four tests, including support ticket routing and shell command guardrails, highlighting its speed, cost-efficiency, and structured outputs.
Duration: 22 minutes 4 seconds.
by Kontext FM
Kai Pervölz and Trutz Fries discuss the mechanics and benefits of decision models like Jev. They explain how providing predefined options yields fast, structured probability outputs without generating text, using support ticket routing as a practical example.
Duration: 46 minutes 25 seconds.
by DSwithBappy
DSwithBappy builds a Python customer-support triage demo using FastAPI. Jev selects a department, scores urgency, and checks refund and human-review conditions, while optional LangChain and Groq generate a suggested reply.
Duration: 50 minutes 32 seconds.
by Reşit Abdullah Yavuzkol
Reşit Abdullah Yavuzkol demonstrates how to set up the Jev model in Python using the TypeSafe SDK and Vercel AI Gateway. The tutorial covers basic configuration and runs a support ticket triage script using Choice, Score, and Noul questions to categorize requests.
Duration: 4 minutes 57 seconds.
by The TWIML AI Podcast with Sam Charrington
Diogo Almeida explains why TypeSafe built Jev for machine-to-machine decisions rather than prose, and describes its probability outputs and training approach.
Duration: 89 minutes 40 seconds.Featured videos
by Codevolution
Codevolution takes a customer message from the TypeSafe playground into TypeScript, explaining Noul, Choice, and Score responses before combining them in one request. The closing examples make the boundary clear: use Jev for bounded judgments, a text model for writing, and ordinary code for exact calculations.
Duration: 22 minutes 12 seconds.
by codebasics
Codebasics and AI research engineer Siddhant Pandey demonstrate Jev's API and LangChain integration. They explain how Jev functions as a decision and classification model using primitive types like Noul, score, and choice, testing it on insurance processing, model routing, and safety guardrails.
Duration: 17 minutes 6 seconds.
by KodeKloud
KodeKloud explains Jev through support-ticket routing, then builds a Papers, Please-inspired approval game with Claude Code and KodeKey. The most useful section shows how several independent ticket questions can guide escalation even when the department choice remains uncertain.
Duration: 21 minutes 51 seconds.
by Dave Ebbelaar
Dave Ebbelaar builds a Python support-ticket flow, inspects category probabilities, and combines category, frustration, and refund questions in one call. He also explains why sending client data to a new provider needs a separate privacy decision, even when the API is easy to use.
Duration: 16 minutes 50 seconds.
by Telusko Alien Coach
This Python tutorial uses Noul, Choice, and Score questions to route support tickets. It demonstrates request configuration, confidence gates, composite priority scores, and parallel questions that inform a handoff to application code, an LLM, or a person.
Duration: 40 minutes 17 seconds.
by ZazenCodes
ZazenCodes demonstrates how to query TypeSafe's Jev model using the Python SDK. The tutorial covers Jev's core decision primitives—Noul, Choice, and Score—along with parallel multi-question evaluation and an automated support ticket triage loop.
Duration: 20 minutes 58 seconds.
by Moritz | AI Systems
Moritz demonstrates how Jev differs from text-generating LLMs by outputting fast probabilities over predefined options. Using Cursor and the TypeSafe API, he implements three practical prototypes: a real-time voice-controlled browser, an efficient semantic memory lookup system, and a hybrid YouTube title scoring and ranking pipeline.
Duration: 22 minutes 41 seconds.
by Ray Amjad
Ray Amjad explains TypeSafe's Jev model as a fast System 1 probability classifier. He demonstrates its three primitives—Noul, choice, and score—in the web console before exploring agentic patterns like parallel browser testing, dynamic skill selection, qualitative linting, and multi-angle PR review paired with System 2 frontier models.
Duration: 27 minutes 28 seconds.
by LangChain
LangChain product manager Sydney introduces Jev, a System 1 decision model from TypeSafe AI. The video covers how Jev handles parallel typed queries (choice, score, boolean) directly on structured state, contrasting it with LLMs and demonstrating harness use cases like model routing, risky tool call gating, and judge-based evals.
Duration: 9 minutes 14 seconds.
by Steve (Builder.io)
Steve from Builder.io tests Jev on automated browser and OS tasks, revealing significant failure rates when operating on raw text representations without visual input. He demonstrates where Jev fails standalone, how a hybrid fallback with multimodal LLMs works, and practical integration patterns like tool selection and priority classification.
Duration: 5 minutes 30 seconds.Written guides.
Understand Jev, one idea at a time.
New written guides
by JP Caparas
JP Caparas tests OpenAI’s Decisions API against Jev on text decisions, then explores photos and long documents. He explains where speed, accuracy and the cost of asking many questions change the choice.
Hamel Husain and Antaripa Saha show why the default 0.50 cut-off can make a good AI checker look bad
by Hamel Husain & Antaripa Saha
Hamel Husain and Antaripa Saha use six decision models to check an apartment-leasing AI for mistakes, and show how to set, compare and trust their scores.
by Lambert Leong
Lambert Leong tests seven open-source projects that pick from a list the way Jev does against a small text-writing model, on routing banking customer messages.
by Evan Fenster / Channel3
Evan Fenster explains where Jev helped Channel3 sort a huge product catalogue into categories, and where it cost too much or saved little.
by kinfey
This community tutorial follows a small Qwen model from training through local deployment, while leaving permissions and safety rules to ordinary code.
by Prasanth Jayachandran / Isotopes AI
Prasanth Jayachandran reports Isotopes AI’s comparison of four models on follow-up questions inside its own data-analysis app.
by scottgal / mostlylucid
scottgal's twelve-case support-chat probe checks whether a reply answers the question, matches the evidence and explains itself clearly, keeping missing answers separate from wrong ones.
by Kevin Kelleher / HiddenLayer Research
Kevin Kelleher describes HiddenLayer’s tests showing that returning a fixed choice does not make Jev immune to misleading page text or altered spellings.
by Artificial Analysis
Artificial Analysis gives four kinds of decision-making equal weight, then compares model fees, response times and how far confidence exceeds accuracy.
by Frederic Lardinois / The New Stack
Frederic Lardinois reports a Workday press demonstration in which Sana uses Jev to choose already-designed paycheck and time-off components.
Featured written guides
by Movez
This guide explains how to separate simple choices from writing tasks. It shows how to use an AI tool that picks from options to help software organize tasks, ask for human review, and run safely.
by Dhravya Shah
Dhravya Shah tests ways to help AI assistants sort and save information. Some choices improve search results, but leaving out text can also mean losing details the assistant needs later.
by Jev Guide
This guide explains how to use an AI tool to sort incoming customer messages into departments like billing or sales. You will learn how the software handles clear requests and what happens when a message is confusing.
by Daria Cupareanu
Daria Cupareanu explains how to connect Jev to an AI assistant through Amplifiers, with fifteen examples of sorting information and choosing next steps.
by Jev Guide
This guide explains how to get a private access code and send your first question to Jev. You will learn how to check your connection, format a yes-or-no question, and understand the AI's mathematical answer.
by Jev Guide
You will learn how an AI tool selects the best action from a list of options. This guide explains why checking for close ties is safer than letting the software act on its first guess.
by Sydney Runkle and Hunter Lovell
This guide explains how to use an AI tool called Jev to make quick decisions for your AI assistant. You will learn how to sort tasks and check for risky actions before the software carries them out.
by TypeSafe AI
This guide explains a method to test whether an AI assistant's answers match the original documents. It shows how to spot made-up quotes, false claims, and statements that lack evidence.
by TypeSafe AI
This guide explains how to fix broken sentences and identify headings in a damaged text file. The approach labels the text parts instead of asking software to rewrite the whole document.
by TypeSafe AI
This guide explains how to use an AI assistant to write questions about wine reviews. The answers become numbers so a prediction program can guess the human rating for each written review.
Experiments.
See what builders made with Jev.
New experiments
by JP Caparas
Street photos, tiny checkboxes and hidden clues in Middlemarch put Jev 1.14 to work in JP Caparas's code-backed experiments.
vLLM Jev
by mode-io (Ziyi Chen)
vLLM Jev runs open Jev-style decision models on your own hardware: one command downloads a Hugging Face checkpoint and serves Choice, Noul and Score questions over HTTP.
Honing
by Haoze He (HectorHHZ)
Honing puts small open decision models into Red Alert 2 and a bouncing-shell tank game, then feeds what happened back into their training and the code around them.
by Parallel Web Systems
Parallel's hosted decision API looks things up on the live web, then answers yes/no, multiple-choice and score questions with the sources it read.
by Firecrawl
Firecrawl's developer guide feeds scraped pages and search results to Jev, which checks a page for a fact, keeps only relevant results and sorts company sites into leads or a review pile.
by Superdesign / Jason Zhou
treg's people-search skill lets a coding agent find companies, the people at them and verified work emails, using Jev to drop companies outside your ideal customer profile before paying for lookups.
by Discourse / Sam Saffron
Discourse AI adds decision model integration to Discourse Workflows, letting site administrators evaluate community posts against custom classification questions for automated moderation, notifications, and tagging.
Verathos Decide
by Verathos
Verathos Decide serves specialized AI decision models on the Bittensor decentralized network, returning structured probability scores across predefined options for classification and moderation tasks.
jevcrawl
by Rayen Mlayeh
jevcrawl is an open-source web crawler and scraper for AI assistants, using a small decision model to score links and passages so agents receive exact source excerpts rather than entire sites.
HiRoute
by HiRoute / Higress
HiRoute is a local routing engine that evaluates each new agent turn to direct tasks across different AI models based on complexity and stage.
Featured experiments
by Nader Dabit
A playful collection of decision demos spanning dispatch, moderation, search, desktop control and a 32-agent arena.
by Unattributed
Jev places Tetris pieces by considering holes, stack height and the current board.
by jkudish
An MCP server that lets agents navigate a browser with Jev decisions.
by Nikhil Hooda, Utkarsh and Xirui Huang
Jev controls rabbits and wolves in a pixel-art habitat where food, weather and inherited traits shape survival.
by RatioDaemon (AI operator)
Change a robot’s sorting rule, take over a game, or inspect Jev’s choices in DriftLoom’s eight-demo playground.
by openroboto-ai
Compare Jev and two language models as they control the same simulated xArm7 apple-placement task.
by adammichaelwood
Solve four-part harmony quizzes, then switch to an endless Rhodes piano whose chords are chosen by Jev.
by Keitark
Read or autoplay an illustrated gamebook while Jev chooses among the exploration and combat actions currently available.
by muratcanberber
Swim through a shared aquarium while Jev-controlled fish decide when to flee, hunt, eat or wander.
by Adriaan Senden-Dennis
A game about linking two seemingly unrelated things in as few steps as possible.
The community, in numbers.
- page views
- —
- visits
- —
- countries recorded
- —
Sources unavailable
It’s a once-a-day snapshot rather than a live counter.
A little of everything.
Written and video guides alongside a recent pick from each registry category. Explore a section for the full collection.
Real-time research powered by
Same ideas.
Different paths.
19 starting points · guides link to our notes first
- Guide · ArticleOpenAI’s Decisions API: fast choices, compared with JevJP Caparas tests OpenAI’s Decisions API against Jev on text decisions, then explores photos and long documents. He explains where speed, accuracy and the cost of asking many questions change the choice.
- Experiment · Benchmarks & researchJev 1.14 can look at photos, point at things and read long booksStreet photos, tiny checkboxes and hidden clues in Middlemarch put Jev 1.14 to work in JP Caparas's code-backed experiments.
- Video guide · ClassificationJev Just Changed Lead Generation Forever! (Tutorial)Instantly shows how Jev can check whether companies fit a sales team’s criteria, recognize buying signals and sort contact data. The presenter reports scoring 1,000 businesses in 6.8 seconds for about $0.045; that demonstration is not an independent performance test.
- Repository · SDKs & clientsvLLM JevvLLM Jev runs open Jev-style decision models on your own hardware: one command downloads a Hugging Face checkpoint and serves Choice, Noul and Score questions over HTTP.
- Demo · Playable demos & botsHoningHoning puts small open decision models into Red Alert 2 and a bouncing-shell tank game, then feeds what happened back into their training and the code around them.
- Guide · guideTune the cut-off before you judge a decision modelHamel Husain and Antaripa Saha use six decision models to check an apartment-leasing AI for mistakes, and show how to set, compare and trust their scores.
- Experiment · IntegrationsJev + Firecrawl company dataFirecrawl's developer guide feeds scraped pages and search results to Jev, which checks a page for a fact, keeps only relevant results and sorts company sites into leads or a review pile.
- Video guide · EvaluationMicrosoft Decision-1 vs JEV vs Perplexity and EVERY Decision Model (I Tested 20)Anass Kartit compares sixteen hosted decision models and four local setups on his own 82-question test. Perplexity’s Decider leads his results; Jev scores 79 out of 82 and Microsoft Decision-1 scores 75. He also examines malicious instructions, reordered choices and differences between local runtimes.
- Experiment · Agent toolingParallel Decisions APIParallel's hosted decision API looks things up on the live web, then answers yes/no, multiple-choice and score questions with the sources it read.
- Experiment · Apps & data pipelinesd1-ultrad1-ultra is Liquid AI’s hosted text-and-image decision model, separate from its earlier Open d1 release.
- Guide · guideSeven open decision models try to route banking requestsLambert Leong tests seven open-source projects that pick from a list the way Jev does against a small text-writing model, on routing banking customer messages.
- Experiment · Humor & parodyThe People v. Your DrawingThe People v. Your Drawing is a browser sketch game that challenges you to illustrate an assigned prompt so an automated judge can pick it out of four options.
- Video guide · Agent workflowsJev Can SERIOUSLY Supercharge Your AI Second Brain (Here's How)Cole Medin shows how Jev screens incoming email, decides which messages deserve a draft, filters news and checks duplicates. He keeps a richer model for judgments needing more personal context.
- Experiment · Browser & computer useCodeceptJS Decision AssertionsAlongside exact assertions, CodeceptJS offers I.decide and I.decideVisually to check whether a page meets a requirement written in ordinary language.
- Collection · Community listsAwesome Jev (tanxarx)A curated ecosystem directory tracking integrations, open-source reproductions, and agent-architecture patterns for the Jev decision model.
- Guide · guideUse Jev to choose from a short list, not to search a huge oneEvan Fenster explains where Jev helped Channel3 sort a huge product catalogue into categories, and where it cost too much or saved little.
- Resource · Official TypeSafeOfficial TypeSafe AI n8n nodeTypeSafe’s own n8n node asks Jev typed questions and routes incoming items to workflow branches, with an optional fallback for uncertainty.
- Video guide · Agent workflowsHow to Build a Speed-to-Lead System in n8n (Jev + OpenRouter)Derek Cheung demonstrates how to build an automated speed-to-lead system in n8n using Jev and OpenRouter. The workflow qualifies leads by analyzing text derived from user-submitted photos and voice notes, routing qualified prospects to receive generated media while pausing uncertain cases for human review in Slack.
- Writeup · Articles & threadsUsing Jev to convert conversational learning signals into tabular featuresA short concept thread proposes converting learner and model language into numerical features using Jev, then fitting a tabular model to learner outcomes.