Jev Use Cases: What 600+ Real Builds Say People Actually Do With It

The fastest way to understand a new model is to ignore the launch thread and look at what people ship with it in week one. This site catalogs those ships: 600+ builds and counting, each with the cost and latency numbers its author reported. This page is the guided tour, organized by what the builds are actually for.

If you want the one-line version first: Jev answers small structured questions instantly for almost nothing, so every use case below is some flavor of "a decision that used to be too expensive or too slow to automate, automated."

1. Triage and routing: the boring goldmine

The single most economically obvious use case. Every company has streams of items needing a fast decision each: emails, tickets, leads, bug reports, meeting requests. The receipts:

• 500 emails classified for 3.5 cents; a separate builder ran ~1,500 of his own inbox through it (build, build)
• Fraud detection at 96/100 accuracy for ~$0.07, with Jev handling the bulk and a bigger model taking the uncertain slice (build)
• Ticket severity grading, churn-risk tiers, lead-quality scoring, crash-report routing (category)

We think support ticket triage is the most under-exploited item on this entire page: every company has the pile, almost nobody has pointed a decision model at it yet.

2. Judging, evals, and moderation

The pattern with its own name: LLM as a judge. Grade outputs, verify agent claims, screen reviews for authenticity, issue safety verdicts on user posts. One builder ran 8 judge questions across 3,282 posts (~26,000 verdicts) for $0.13 (build). At that price, "judge everything" stops being rhetorical.

3. Agents and browsers

Agents spend most of their wall-clock time waiting for a model to choose the next action. Swap a frontier model out of that loop and the whole thing changes character:

• Flight search in ~7 seconds for $0.004 with Browser Use (build)
• ~$0.001 per task driving a remote browser via Stagehand (build)
• Computer use with no screenshots at all: on-device OCR reads the screen, Jev only ever sees text (build)

Deep dive: Jev and browser agents. Category: agents and browsers.

4. Real-time and games

The showoff category, and secretly the most important one: it proves the latency claim in public. Doom at ~10 decisions/second, Mario, Slay the Spire 2 at 0.7 seconds per move, a semantic typing racer (category). The serious translation: any product loop that needs a model decision inside the frame budget (games, live UIs, robotics) just became buildable. Full tour: real-time games on Jev.

5. Content and growth

Scoring drafts before posting (61 questions per draft, ~1 second, $0.0004), tearing down 724 competitor ads in 40 seconds for 9 cents, filtering reply-guy comments, rebuilding the X algorithm to simulate virality (category). The common shape: marketing decisions that were vibes are becoming measurements.

6. Trading and markets

Yes, people wired it to real money within 48 hours, including a $10,000 live-trading experiment and an on-chain bot placing orders every 300ms block (category). We catalog these as engineering artifacts, not endorsements: a fast model is not an edge, and the market does not care about your token costs.

7. Tools, apps, and the genuinely weird

The biggest category, and where the surprises live. The standout: a PostgreSQL extension that turns natural language into a query filter — WHERE jev(people, 'could work from home') — judging 129 rows in about a second (build). Also: instant context compaction for coding agents, PR-merge verdicts from a pasted link, and 200+ more (category).

8. Robotics and devices

Smallest category, longest fuse. A decision model fast enough for game loops is fast enough for control loops, and the first on-device experiments are trickling in (category).

The pattern behind all of it

Read enough entries and one sentence covers the whole directory: Jev makes decisions cheap enough to put everywhere. Not smarter decisions than a frontier model. More decisions, in more places, inside tighter loops, at costs that round to zero. The builds that impress us most aren't the flashiest; they're the ones like the fraud pipeline, where a cheap judge does 90 percent of the work and an expensive model handles only the hard slice. That architecture is going to be everywhere by December.

Built something? Submit it. The directory runs on receipts.

By role

The same catalog, cut by who is building: Jev for developers, Jev for marketers, Jev for ecommerce, Jev trading bots, and Jev for founders, the solo operating stack.

Frequently asked questions

What is Jev best used for?

High-volume structured decisions: classification, routing, scoring, judging, and agent action-selection. Anywhere you need thousands of small verdicts fast and cheap, it fits; anywhere you need essays or deep reasoning, it doesn't.

What has actually been built with Jev so far?

600+ cataloged builds across eight categories: email and ticket triage, browser agents, real-time games, eval judges, content scoring, trading experiments, database tools, and early robotics. Every entry carries its author-reported numbers.

Can Jev replace GPT or Claude in my product?

For the decision-shaped parts, often yes, at a fraction of the cost. For generation and complex reasoning, no. The strongest production builds combine them; see Jev vs GPT.

What's the cheapest real Jev use case reported so far?

Post scoring at $0.0004 for 61 questions per draft, and browser tasks at ~$0.001 each. The running list lives in what builds actually cost.

How do I get my project listed here?

Ship it, then submit the link with your numbers. Curation is human; receipts required; sponsor slots exist but listings are free.