EvidenceVideos checked , 19:42–19:57 UTC
Jev (TypeSafe AI) demo videos that show it working, with timestamps
These are videos in which the creator runs TypeSafe's Jev on a concrete task: sorting email, routing tickets, moderating posts, routing between models, playing a game. Each link opens the video at the second where the creator's own chapter says the demonstration starts. Overview, news, review and compilation videos are left out.
Short answer
46 videos passed the rule (48 entries, because two videos appear in two groups), across 13 use-case groups. Start with the 16 picks below: one or two of the strongest original demonstrations per group. Every start time comes from the creator's chapter markers; the pool did not watch the videos.
How the timestamps were checked. Each timestamp is the creator's own chapter marker, also listed in the video description. The pool has not watched the frames or read transcripts. Figures quoted in videos are the creators' own and were not reproduced.
YouTube showed a login wall to the pool's client, so no frames or transcripts were available. That Jev is being called on screen at the linked second is the creator's labelling, not the pool's observation. Reported for every row.
Best original demos: one or two per use case
- Email and inboxMeet Jev: The AI Built to Make DecisionsClassifying the creator's own emails (category, priority, spam, reply) in batches of 100 then 1,000.Demo at 0:54
- Email and inboxJEV in pratica (con TypeScript)Email classification, then a helpdesk demo; compared with BERT and an LLM.Demo at 8:22
- Support ticketsJev Explained for Python DevelopersPython support-ticket classification, then Choice/Score/Noul and multi-question calls.Demo at 0:00
- Support ticketsJev: The AI Model That's Breaking The Internet (Full Tutorial)Support-ticket triage: route, auto-reply or send to a human (Build 2); also LLM router 8:56 and inbox classifier 12:06.Demo at 10:34
- ModerationJev Full Tutorial for Beginners: How to Use & Set up Jev (Typesafe)Content moderation and policy-violation assessment with borderline-content handling (Build 3).Demo at 12:06
- Routing and agent harnessHow To Build A Harness With Jev | A LangChain x TypeSafe ConversationLive demo: safety classification and model routing in an agent harness.Vendor participant (TypeSafe's Allie Laabs)Demo at 19:57
- Routing and agent harnessBuild a Jev router that picks your AI modelBuild a model router in Claude Code where Jev picks which Claude model gets each task; gates and human review.Demo at 5:11
- PythonI Tried Jev AI in Python: Honest First LookReplacing brittle yes/no parsing with one Noul question.Chapter and description give conflicting first-answer figures; quote neitherDemo at 10:38
- PythonJev: Typed AI Decisions in Python — Build a Classifier + Route Coding Standards with Claude CodeFirst API call, a support-triage classifier, then routing 54 coding conventions across a 14-module app; packaged as a Claude Code skill at 10:32.Demo at 3:42
- TypeScript and AI SDKWhat is Jev and How to Use it?Frustration detection and classification of customer messages, Playground first then TypeScript SDK.Demo at 12:54
- n8n and no-codeHow To Setup JEV With n8n: I Let It Sort My Gmail Leads (n8n Setup)n8n workflow sorting Gmail leads into Book call / Nurture / Not a fit / Client and drafting replies.Demo at 9:02
- Claude CodeHow to Use Jev Instantly in Claude Code with OpenRouter (No Waitlist)Test Jev with curl, then install and customise the TypeSafe skill for Claude Code over OpenRouter.Demo at 2:21
- Evaluation and financeIs Jev actually good? We tested TypeSafe's classifier on 3,080 bank support messagesBanking77 evaluation: 3,080 test messages, auto-accept at a 2% error target vs three LLMs, two local classifiers and Laya.Demo at 5:53
- Trading (paper)I Tested Jev on 12 Real Use Cases. My Honest Thoughts.Real-time paper-trading prototype; X feed classifier Chrome extension at 10:00.Paper tradingDemo at 14:13
- Games and real-timeJev is the FIRST of a Whole New Class of AI Models (Here's How to Actually Use It)Jev plays the creator's indie game at three decisions a second; PR triage 8:54; model routing 10:42.Sponsored segment 6:58–8:54Demo at 5:32
- Browser and scrapingJev: The New AI Model That's Breaking The Internet (Full Tutorial)Voice-controlled browser; memory retrieval at 11:33; YouTube idea scoring at 17:27.Demo at 5:59
All kept videos by use case
Email and inbox classification (4)
| Video (author · date · length · language) | Use case | What it shows | Demo start |
|---|---|---|---|
| Meet Jev: The AI Built to Make Decisions vogel · 15 Sep 2026 · 4:07 · English | Classifying the creator's own emails (category, priority, spam, reply) in batches of 100 then 1,000 | Creator's own batch run; route/SDK not stated in description; live call per description ('I put Jev to work on my own emails') Creator-reported: average response ~200 ms; cost shown at 3:13 (not quoted in description) | Demo at 0:54 Chapter: “0:54 Testing Jev on my own emails” |
| Jev is HERE. How to use it Greg Isenberg · 18 Sep 2026 · 28:25 · English | Email triage demo by guest Ryan Vogel (1,700 emails); later browser flight pick at 25:27 | Podcast with on-screen demo segment; route: description links Vercel AI Gateway; live vs recorded run not stated Creator-reported: '1,700 emails for 18 cents total'; '~200 milliseconds' per query | Demo at 4:32 Chapter: “4:32 Email Triage Demo” |
| JEV in pratica (con TypeScript) Giuseppe Funicello · 28 Sep 2026 · 23:38 · Italian | Email classification, then a helpdesk demo; compared with BERT and an LLM | TypeScript with Vercel AI SDK (description: 'codice TypeScript con AI SDK'); code at 10:38; linked repo giuppidev/jev-exam... Benchmark Jev vs BERT vs LLM at 9:42 is creator-run | Demo at 8:22 Chapter: “8:22 Demo: classificare le email” |
| I’m using Jev more than Opus 5.5 or GPT-6. Here’s why. How I AI · 28 Sep 2026 · 26:25 · English | Gmail triage with Jev scoring and LLM follow-up (use case 3 of 5) | Creator's own projects; also PR categorization at 7:39 and an audience-signal dashboard demo at 18:17; route not stated Description: '4 cents per million input tokens with no output charge' (vendor price, repeated by creator) | Demo at 13:00 Chapter: “13:00 Use case 3: Gmail triage with Jev scoring and LLM follow-up” |
What five email and lead sorters do to your inbox when Jev fails, with mailbox permissions and cost per 1,000 emails, read in source: email and lead sorting. The "1,700 emails" figure from the Isenberg and Vogel video is set beside the pool's own cost arithmetic there.
Support tickets and customer messages (5)
What five ticket-triage projects do when Jev is unsure or fails, read in source: support-ticket triage.
| Video (author · date · length · language) | Use case | What it shows | Demo start |
|---|---|---|---|
| Jev Explained for Python Developers Dave Ebbelaar · 21 Sep 2026 · 16:51 · English | Python support-ticket classification, then Choice/Score/Noul and multi-question calls | Python; official docs linked; follow-along repo daveebbelaar/ai-co...; latency compared with Claude models at 10:13 Creator-run latency comparison vs Claude Haiku/Opus 5/Fable 5.1 | Demo at 0:00 Chapter: “0:00 Jev and your first Python request” |
| Jev Full Tutorial for Beginners: How to Use & Set up Jev (Typesafe) Mikey No Code · 23 Sep 2026 · 17:29 · English | Support-ticket triage and department routing in the TypeSafe console Playground (Build 2) | TypeSafe console Playground (no code); also resume screening at 6:49 and moderation at 12:06 | Demo at 9:24 Chapter: “9:24 Build 2: Support Ticket Triage & Automated Department Routing” |
| Jev: The AI Model That's Breaking The Internet (Full Tutorial) Mayank Aggarwal · 20 Sep 2026 · 17:42 · English | Support-ticket triage: route, auto-reply or send to a human (Build 2); also LLM router 8:56 and inbox classifier 12:06 | Creator's demo app, repo mayank953/Jev; compared with Kimi and Claude Creator-reported: ticket triage 'about 7x faster and 16x cheaper than a large LLM'; 25 emails 'about 4 seconds with Jev vs 18 seconds' | Demo at 10:34 Chapter: “10:34 Build 2: Support Ticket Triage” |
| What is Jev AI? Explained with Live demo by Mohit. DrHimanshiChaudhary · 27 Sep 2026 · 11:48 · English (title/description) | Support-ticket decision vs Google Gemini structured output; agent routing demo at 10:20 | Flask app code walkthrough at 4:13; live latency/confidence test Creator-run latency comparison vs Gemini | Demo at 9:15 Chapter: “9:15 Live Test: Jev AI Speed & Confidence Scores” |
| Jev AI in 6 Minutes: How to Get In Without the Waitlist (and My Real Test) The AI Stack · 28 Sep 2026 · 6:13 · English | Refund-request test and the creator's bill | Access via OpenRouter / Vercel AI Gateway / Cloudflare described; which route was used for the test is not stated in the description Cost shown at 4:48 ('What it cost me'); not quoted in description | Demo at 2:59 Chapter: “2:59 What I found: the refund test” |
Content moderation (2)
What six moderation projects do when Jev fails or is unsure, and who can undo a removal, read in source: content moderation. Mikey No Code's moderation build starts at 12:06.
| Video (author · date · length · language) | Use case | What it shows | Demo start |
|---|---|---|---|
| Jev Full Tutorial for Beginners: How to Use & Set up Jev (Typesafe) Mikey No Code · 23 Sep 2026 · 17:29 · English | Content moderation and policy-violation assessment with borderline-content handling (Build 3) | TypeSafe console Playground; multi-question policy check, severity and action; borderline case at 13:17 | Demo at 12:06 Chapter: “12:06 Build 3: Content Moderation & Policy Violation Assessment” |
| Jev: This is what the AI model can really do (10 use cases) Julian Ivanov | KI-Automatisierung · 22 Sep 2026 · 24:54 · German | Guardrail and model router in a support bot | Creator's own apps built with Claude Code; access via OpenRouter (typesafe/jev-1.13) per description | Demo at 12:47 Chapter: “12:47 Guardrail und Model-Router im Support-Bot” |
Model routing, tool selection and agent harness (7)
Where six self-hosted routers send a request when Jev fails, and what OpenRouter documents about its Jev Router, read in source and docs: model routing.
| Video (author · date · length · language) | Use case | What it shows | Demo start |
|---|---|---|---|
| How To Build A Harness With Jev | A LangChain x TypeSafe Conversation LangChain · 22 Sep 2026 · 48:04 · English | Live demo: safety classification and model routing in an agent harness | LangChain x TypeSafe live session; TypeSafe's Allie Laabs participates (vendor-involved) Vendor participant (TypeSafe's Allie Laabs) | Demo at 19:57 Chapter: “19:57 Live demo: safety classification and model routing” |
| Building a Harness with Jev LangChain · 21 Sep 2026 · 9:15 · English | Jev vs an LLM head to head; model routing, risky-tool-call auto mode, Jev-as-judge | LangChain channel; Jev inside LangChain Description repeats vendor claim 'up to 200x faster and 400x cheaper' | Demo at 3:27 Chapter: “3:27 Demo: Jev vs an LLM head to head” |
| Build a Jev router that picks your AI model NextWork · 28 Sep 2026 · 18:01 · English | Build a model router in Claude Code where Jev picks which Claude model gets each task; gates and human review | OpenRouter key + TypeSafe plugin in Claude Code; first call 5:11; router part 1 at 8:31 | Demo at 5:11 Chapter: “5:11 your first jev call” |
| Jev and Herdr: The Best Way to Route Models (Must-Try Workflow) Nidhi Singh · 17 Sep 2026 · 17:11 · English | Model-router CLI (herdr + Jev) routing a research task to Grok 4.6; plan/implement split at 11:48 | Creator's own router CLI; API key in Keychain; not open-sourced at upload | Demo at 10:50 Chapter: “10:50 Demo 1: a research task gets routed to Grok 4.6” |
| Jev with LangChain: Smarter AI Agent Tool Selection | Reduce Tokens & Cost Mohamed Naji Aboo · 23 Sep 2026 · 8:54 · English | Tool selection before a LangChain agent runs (10 tools to 1) | Python + LangChain; the 10 tools are mocked dummies (chapter 1:45); basic classifier test at 0:43 Creator-run token/cost benchmark at 4:08 Tools are mocked | Demo at 2:30 Chapter: “2:30 Dynamic Tool Selection with TypeSafe Classifier” |
| How to Build Things with Jev & OpenJevs Sam Witteveen · 21 Sep 2026 · 19:19 · English | Model router built with the hosted Jev API and with SemIf (open alternative) | Router demo; code 'up in the next day or so' per description Stats at 16:46 | Demo at 5:59 Chapter: “5:59 Model Router Demo” |
| JEV Explained in Plain English — Build a Real AI Model Router cloud-monk - cloud in plain english · 21 Sep 2026 · 10:34 · English | Model router that picks which Gemini model handles each request | Chapter is titled only 'Real World Application'; the description says this is where the router is built | Demo at 3:59 Chapter: “3:59 Real World Application” |
Python (3)
| Video (author · date · length · language) | Use case | What it shows | Demo start |
|---|---|---|---|
| I Tried Jev AI in Python: Honest First Look Real Python · 22 Sep 2026 · 17:03 · English | Replacing brittle yes/no parsing with one Noul question | Python, typesafe-sdk package pointed at OpenRouter; first run at 10:38, first answer at 11:07; thresholds at 14:33 Chapter and description give conflicting first-answer figures; quote neither | Demo at 10:38 Chapter: “10:38 Running with --env-file” |
| Jev: Typed AI Decisions in Python — Build a Classifier + Route Coding Standards with Claude Code Omar Kamal · 21 Sep 2026 · 12:20 · English | First API call, a support-triage classifier, then routing 54 coding conventions across a 14-module app; packaged as a Claude Code skill at 10:32 | Python venv + agent skill; 'Every number in this video was measured live against the Jev API' (creator) Creator-reported: 14/14 module types, 0.87 precision / 0.97 recall, 'under 3 cents in total' | Demo at 3:42 Chapter: “3:42 Your first API call” |
| Jev - The Ultimate Classification Model? Sam Witteveen · 18 Sep 2026 · 16:19 · English | Choice, Score and Noul demos, then a 'practical demo' at 9:18 | Language/notebook not stated in description; demo chapters 4:56-12:49 | Demo at 4:56 Chapter: “4:56 Demo” |
TypeScript and Vercel AI SDK (3)
To choose between the paths shown in these videos (official SDK, AI SDK, fetch, Workers), see Call Jev from TypeScript.
| Video (author · date · length · language) | Use case | What it shows | Demo start |
|---|---|---|---|
| What is Jev and How to Use it? Codevolution · 19 Sep 2026 · 22:12 · English | Frustration detection and classification of customer messages, Playground first then TypeScript SDK | TypeScript SDK; Playground at 11:20; multiple questions at 16:17 | Demo at 12:54 Chapter: “12:54 Using Jev with TypeScript” |
| I tried TypeSafe's System One Model: Jev Joe Maddalone · 18 Sep 2026 · 7:48 · English | TypeScript application using and testing Jev | TypeScript; code repo joemaddalone/jev-t... (truncated) Description repeats vendor '193x speed and 444x lower cost' claim | Demo at 2:11 Chapter: “2:11 Typescript Application Using and Testing Jev” |
| Is Jev Actually Useful for E-Commerce? 306 Technologies · 28 Sep 2026 · 7:30 · English | E-commerce warehouse (fulfilment location) selection for an order | MedusaJS + Vercel AI SDK setup at 2:24; priority change test at 5:52 | Demo at 4:01 Chapter: “4:01 Testing Warehouse Selection” |
n8n and no-code workflows (3)
| Video (author · date · length · language) | Use case | What it shows | Demo start |
|---|---|---|---|
| How To Setup JEV With n8n: I Let It Sort My Gmail Leads (n8n Setup) Stephanie Nyarko · 24 Sep 2026 · 18:43 · English | n8n workflow sorting Gmail leads into Book call / Nurture / Not a fit / Client and drafting replies | n8n via OpenRouter (access 'when TypeSafe's sign-ups were full'); community node failure at 7:08; bug at 12:45 Creator notes vendor '194x faster and 445x cheaper' are 'their numbers, not my tests'; 6-email test | Demo at 9:02 Chapter: “9:02 The Ask Jev node: three questions” |
| jev + n8n: turn a probability into a decision Perseo n8n Automations · 22 Sep 2026 · 2:39 · English | n8n Jev Gate community node: 6 support messages, confidence gate, Calibrate recommending 0.7 | n8n-nodes-jev-gate (MIT); jev-1.13 through OpenRouter Creator-reported: 4 of 6 through Confident, 2 to Review; Calibrate recommends 0.7 | Demo at 1:25 Chapter: “1:25 Running it in n8n” |
| #51 - JEV: THE AI THAT DECIDES IN MILLISECONDS (4 practical cases in N8N) - Complete N8N Course Hiperautomação · 27 Sep 2026 · 13:08 · Portuguese | Four n8n cases: service triage with Switch, 'intelligent IF', Jev as judge of AI replies, batch review analysis | n8n HTTP Request node through OpenRouter | Demo at 4:57 Chapter: “4:57 Caso 1: triagem de atendimento” |
Which n8n node is TypeSafe's, and what each of five nodes does when the Jev call fails: Jev n8n node. The Gmail-leads workflow in Stephanie Nyarko's video was not read as a project; email sorters read in source are on email and lead sorting.
Claude Code and coding agents (3)
Before installing a Jev hook, plugin or MCP server like those in these videos, check its failure behaviour on Claude Code and MCP: which integration fails open.
| Video (author · date · length · language) | Use case | What it shows | Demo start |
|---|---|---|---|
| How to Use Jev Instantly in Claude Code with OpenRouter (No Waitlist) Philipp Acsany · Practical Claude Code · 22 Sep 2026 · 8:14 · English | Test Jev with curl, then install and customise the TypeSafe skill for Claude Code over OpenRouter | curl at 2:21; skill install 3:52; OpenRouter instead of the official SDK | Demo at 2:21 Chapter: “2:21 Test Jev with curl” |
| 10 Levels of Jev For Agentic Engineers IndyDevDan · 28 Sep 2026 · 35:18 · English | Ten levels from a 'smart if' to guardrail hooks (14:31), auto-compact and an agent that calls Jev itself | Coding-agent hooks; repo disler/ten-levels-... | Demo at 0:57 Chapter: “0:57 Smart If Statement Jev” |
| How to add Jev AI to Claude Code in 6 minutes (step-by-step) Kev Builds Apps · 19 Sep 2026 · 6:08 · English | Jev + Claude Code CLI demo, then step-by-step setup | Claude Code CLI; creator promotes own product | Demo at 1:21 Chapter: “1:21 Jev + Claude Code CLI Demo” |
Evaluation and finance checks (3)
Fraud and trading checks in code: trading and fraud gates. Jev as an eval judge in code: Jev as a judge.
| Video (author · date · length · language) | Use case | What it shows | Demo start |
|---|---|---|---|
| Is Jev actually good? We tested TypeSafe's classifier on 3,080 bank support messages M37 · 26 Sep 2026 · 23:27 · English | Banking77 evaluation: 3,080 test messages, auto-accept at a 2% error target vs three LLMs, two local classifiers and Laya | Results walk-through (not a live coding session); cutoffs fixed on 573 dev messages Creator-reported: Jev accepted 84.5% at cutoff 0.99 with 2.04% disagreement (95% CI 1.56-2.65%) | Demo at 5:53 Chapter: “5:53 Jev's result” |
| JEV против CLAUDE: прогнал 200 счетов, дешевле в 50 раз INCUBEAI · 28 Sep 2026 · 12:41 · Russian | Checking 200 supplier invoices (15 with traps: changed bank details, duplicates, new suppliers) with code, Jev, Claude and a human | OpenRouter; first invoice at 3:00; 83 invoices auto-approved on attempt 2 Creator-reported: 134/200 first attempt; 181 correct after two changes; whole run 13 cents | Demo at 3:00 Chapter: “3:00 Первый счёт: «сменили банк, оплатите сегодня»” |
| TypeSafe AI Jev: 50 Financial Jobs, Verifiable Demos & Honest Limits The Fintech Builder · 22 Sep 2026 · 2:01:28 · English | 50 finance workflows (ledger, reconciliation, fraud, AML, trading) with baselines and failures | Saved model outputs replayed, not live calls (description: 'We show the saved model outputs'); code and records open-source Replayed saved outputs, not live calls | Demo at 7:35 Chapter: “7:35 01 Ledger integrity review” |
Creator-run tests in videos (M37's Banking77 run above, Nate Herk's "12 Real Use Cases" and Giuseppe Funicello's comparison with BERT and an LLM) are not graded yet. On the benchmarks ledger they are listed as leads, treated as grade D until reviewed. Published written studies of the same tasks, such as Banking77, are graded there. Reported
Trading (paper) (2)
Whether a Jev failure still places the order, for four trading bots and three fraud or payment gates read in source: trading and fraud gates. The videos show demos; the pool did not check any trade or return shown in them.
| Video (author · date · length · language) | Use case | What it shows | Demo start |
|---|---|---|---|
| I Tested Jev on 12 Real Use Cases. My Honest Thoughts. Nate Herk | AI Automation · 19 Sep 2026 · 16:08 · English | Real-time paper-trading prototype; X feed classifier Chrome extension at 10:00 | Description: 'live builds including an X feed classifier and a real-time paper-trading prototype' Email classification benchmark at 5:28 (creator-run) Paper trading | Demo at 14:13 Chapter: “14:13 Real-Time Trading & Handoffs” |
| Build a 24/7 AI Trading Bot With JEV Lewis Jackson · 21 Sep 2026 · 17:16 · English | 24/7 trading-bot loop: configure API keys, run the initial test | Claude Code-built loop; API keys configured at 8:36. Creator's linked page (part-timequant.com/jev, 'The Jev loop from the video') says it 'Runs on Alpaca paper money' and without a Jev key 'runs on a labelled mock' Paper or live not stated; creator's page mentions running mocked without a key | Demo at 12:39 Chapter: “12:39 Running the initial test” |
Games and real-time control (4)
| Video (author · date · length · language) | Use case | What it shows | Demo start |
|---|---|---|---|
| Jev is the FIRST of a Whole New Class of AI Models (Here's How to Actually Use It) Cole Medin · 21 Sep 2026 · 17:14 · English | Jev plays the creator's indie game at three decisions a second; PR triage 8:54; model routing 10:42 | Creator's own game and repo; sponsor segment 6:58-8:54 Creator-reported: three decisions a second Sponsored segment 6:58–8:54 | Demo at 5:32 Chapter: “5:32 Jev Plays My Game (Insane)” |
| Jev Explained for Beginners with Demo KodeKloud · 23 Sep 2026 · 21:51 · English | 'Papers, Please'-style approval-queue game: Jev vs GPT-5.5 vs DeepSeek | Built with Claude Code on KodeKey (KodeKloud's own access route; sponsored link); results at 21:15 Sponsored access route (KodeKey) | Demo at 14:54 Chapter: “14:54 Building the game on KodeKey with Claude Code” |
| 爆火的 TypeSafe AI Jev 模型到底怎么玩？海量社区案例带你入门，附两个实战 Demo NiceKate AI · 17 Sep 2026 · 16:28 · Chinese | Two demos: a rabbit-adventure game (action choice) and an 'alien entry' approval check | Community cases first (2:54-9:27 are other people's demos); own demos from 9:27 | Demo at 9:27 Chapter: “9:27 实战一：兔子冒险” |
| This new AI model isn't an LLM. It's 20x cheaper and plays Doom in real time (Jev explained) Alex Volkov from ThursdAI · 18 Sep 2026 · 37:32 · English | Live tweet classification at 12 tweets a second; Doom demo at 26:40 | ThursdAI livestream with TypeSafe's Allie Laabs (vendor participant) Creator-reported: 12 tweets a second Vendor participant (TypeSafe's Allie Laabs) | Demo at 2:48 Chapter: “2:48 Live demo: classifying 12 tweets a second” |
Browser, scraping and data extraction (3)
| Video (author · date · length · language) | Use case | What it shows | Demo start |
|---|---|---|---|
| Jev: The New AI Model That's Breaking The Internet (Full Tutorial) Moritz | AI Systems · 17 Sep 2026 · 22:41 · English | Voice-controlled browser; memory retrieval at 11:33; YouTube idea scoring at 17:27 | API key setup at 4:06; three prototypes | Demo at 5:59 Chapter: “5:59 Building a voice-controlled browser” |
| AI Web Scraping with Jev + Playwright is Mind-Blowing! ByteGrad · 28 Sep 2026 · 24:53 · English | Classifying scraped data with Jev, then full demo with CSV export (21:29 onward) | Playwright scraping pipeline; first 21 minutes are scraping without Jev; sponsored (Oxylabs) Sponsored (Oxylabs); first 21 min are scraping without Jev | Demo at 21:29 Chapter: “21:29 Classifying data with Jev” |
| Jeev in Practice: I Created a Tool That Finds Ads That Sell Nate IA | Vibe Coding Elite · 27 Sep 2026 · 24:24 · Portuguese | Chrome extension reading the Meta Ad Library and flagging ads likely to sell | OpenRouter key in the browser; extension repo vibecodingelite-ai/... Costs at 20:46 | Demo at 14:22 Chapter: “14:22 JEV Funcionando na prática.” |
What browser and computer-use agents do when Jev errors, is unsure or has no key, read in source for seven agents on 5 Oct 2026: Jev browser use and computer use.
Other concrete demos (6)
| Video (author · date · length · language) | Use case | What it shows | Demo start |
|---|---|---|---|
| Jev: This is what the AI model can really do (10 use cases) Julian Ivanov | KI-Automatisierung · 22 Sep 2026 · 24:54 · German | Two apps the creator built (Demo 1 and Demo 2), then support-bot guardrail/router at 12:47 | Apps built with Claude Code; OpenRouter typesafe/jev-1.13 | Demo at 2:17 Chapter: “2:17 Demo 1” |
| Jev by TypeSafe AI | What is a System-1 Decision Model | CampusX CampusX · 24 Sep 2026 · 1:29:58 · English title/description; spoken language not verified | Hands-on multi-attribute e-commerce review classifier; SDK deep dive at 51:00 | Long lecture; hands-on section 45:00-55:20; repo campusx-official/j... | Demo at 45:00 Chapter: “45:00 Hands-on Project: Multi-Attribute E-Commerce Review Classifier” |
| Understand Jev (the only video you need) ZazenCodes · 22 Sep 2026 · 20:58 · English | Noul, Choice, Score examples and an autonomous triage loop (12:04) | Source code linked (zazencodes); language not stated | Demo at 4:14 Chapter: “4:14 Noul: Yes/No Probabilities” |
| JEV / TypeSafe AI API & Rust - Let make a rust lb Jeremy Chone · 19 Sep 2026 · 2:44:41 · English | Rust client: cURL API deep dive (9:12), live API run (38:45), crate v0.0.3 published at 2:29:45 | Rust livestream; creator's sysone library | Demo at 38:45 Chapter: “38:45 Live API Run: Testing Fast Classification & Multi-language Prompting” |
| Jev入門 話題の判断専用AI、Jevの組み込みを試してみました ともすた | たにぐち まこと · 21 Sep 2026 · 16:15 · Japanese | Using Jev (0:31) then embedding it in an app built with AI coding (7:33); inquiry-email classification per description | Japanese tutorial | Demo at 0:31 Chapter: “0:31 Jevを使ってみよう” |
| Jev Explained: Demos and Use Cases Syntax · 17 Sep 2026 · 17:51 · English | Creator's own 'jev chat' bot with no LLM; Wikipedia Q&A 14:27; smart-home control 15:40 | Own repo github.com/w3cj/jev-chat (exists; created 2026-09-18T21:09Z per GitHub API). Earlier demo chapters (5:36-11:26) link to other people's X posts and are not the creator's demos Use from 12:10 only; earlier demo chapters are other people's clips | Demo at 12:10 Chapter: “12:10 demo: jev chat bot with no LLM” |
Inclusion rule and what was left out
A video is listed only if the creator's chapters or description show Jev being run on a concrete task (in the Playground, an SDK, curl, n8n or an app) and a chapter marks where that starts. The link points to the start of that chapter. Videos with a TypeSafe participant are listed but flagged, because they are not independent.
| Left out | Videos |
|---|---|
| Overview, news, review or interview | 10 |
| Talks about Jev without showing it working | 4 |
| Compilation of other people's demos | 2 |
| Shows an alternative model only (Laya or Kev), not Jev | 2 |
| Mock responses, no Jev call (creator's own disclosure) | 1 |
| Excluded in total | 19 |
| Held: a demo is claimed, but the start could not be verified from chapters or description, or the author or date could not be established | 13 |
78 distinct videos were inspected on 28 September 2026 (76 on YouTube, 1 on X, 1 on Loom). Excluded and held videos are not linked. Channels that re-upload or farm other people's content, named in the 28 September audience research, stayed excluded and were not re-inspected. The two Laya and Kev comparison videos are alternatives material, not Jev demos; see Alternatives. Graded tests that ran Jev beside Laya or Kev, and newer "jev vs laya" videos (titles only, not reviewed), are on Jev vs Laya vs Kev.
What was not verified
- No frames or transcripts were viewed for any video.
- The route or SDK is not stated for several videos (for example
9oWxrsRo4d8,X117w2Rark8,EAh1h1GYxDM), and whether4mTLpuQpB80shows a live or recorded run is not stated. - Whether the trading-bot video
8DgRqDukf-Uused a real key is not established; the creator's page says the system runs on Alpaca paper money and has a labelled mock mode. - The spoken language of the CampusX lecture (
0zFfcEr1e9U) was not verified; other languages are taken from titles and descriptions. - Repository links that YouTube truncated in descriptions were not resolved.