3 AI tools you can build this weekend
Jev answers one question at a time in a fraction of a second for a fraction of a cent, which makes small, always-on helpers cheap enough to run on everything you read.
What the reel showed
Three things people built with Jev, TypeSafe's decision model, in its first two weeks. Jev does not write text. It reads something and picks an answer from options you give it, in about a third of a second.
An X feed filter that tags every post as relevant, AI slop, an ad or skip, and dims what you do not want. A slop detector that flags generic AI-written paragraphs in a document. And an inbox sorter that files every email as reply now, later or bin.
| Entrant | Tool | Time | Cost | Verdict |
|---|---|---|---|---|
| X feed filter | Jev on OpenRouter | 1.0s | $0.0002 | 12 posts, every one tagged as expected |
| Slop detector | Jev on OpenRouter | 0.47s | $0.0002 | 271 words, all 6 filler paragraphs caught |
| Inbox sorter | Jev on OpenRouter | 1m49 | $0.09 | 5,000 emails, 99.8% correct |
The feed filter and slop detector are recreations of community builds (sift, jev-x and Jev Detector), run live on sample content. The inbox sorter is Felix's own run on 5,000 emails.
What this means for your business
Most of the noise in a working day is sorting: which emails matter, which posts are worth reading, which leads deserve a call. Each of those is one small decision, repeated hundreds of times.
A model like Jev makes that decision for a fraction of a cent, fast enough that you do not notice it running. That is what makes it practical to put a filter in front of everything instead of sorting by hand.
What I’d use
Start with the inbox sorter. It is the easiest to check, you see the result the same day, and it saves real time. Build the feed filter next if you lose time on X or LinkedIn.
Try it yourself
Each tool below starts with one prompt. Paste it into Claude Code (or Codex, if you use ChatGPT) and it builds the tool and walks you through the rest. You need an OpenRouter account with a few dollars of credit; running these costs cents.
Create an OpenRouter account and a key
One account for Jev and every other model. Add a few dollars of credit, then create a key and keep it private, like a password.
Open Claude Code in an empty folder
Claude Code builds and runs the tool for you. It comes with Claude Pro and Max. Put your key in a file called .env as OPENROUTER_API_KEY=your-key.
Tool 1: an inbox sorter
Sorts every email into reply now, later or bin, with a reason you can check.
Build me an inbox sorter that uses Jev (model typesafe/jev-1.13 on OpenRouter). My OpenRouter key is in .env as OPENROUTER_API_KEY. Never print it. Read https://openrouter.ai/docs/guides/community/jev-tutorial first. Make a small web page on my computer where I paste emails (one per block). For each email, ask Jev one choice question with these options: - reply_now: someone is waiting on me or it involves money, a client or a deadline - later: useful but nothing is waiting on me - bin: newsletters, marketing and notifications Show each email with its label, colour-coded, and a total cost at the top. Start with 20 realistic sample emails so I can see it working, then explain how I would connect my real inbox.
Tool 2: an X feed filter
Tags every post as relevant, AI slop, an ad or skip, and fades out the ones you do not want.
Build me a Chrome extension that filters my X (Twitter) feed using Jev (model typesafe/jev-1.13 on OpenRouter). Read https://openrouter.ai/docs/guides/community/jev-tutorial first. For every post that appears as I scroll, ask Jev one choice question with these options: - relevant: a concrete, first-hand, useful point about my work, with real numbers or a specific tip - slop: generic AI-written engagement bait, buzzwords, hype, asks for likes or reposts - ad: a brand or seller promoting a product, trial, course or paid seats - skip: off-topic personal chit-chat Put a small coloured label on each post, fade out slop and ads, and show a running cost counter. Let me paste my OpenRouter key in the extension's settings page, stored only on my computer. Then walk me step by step through loading it into Chrome, as if I have never done it before.
Tool 3: a slop detector
Paste any document and see which paragraphs read like generic AI filler.
Build me a slop detector that uses Jev (model typesafe/jev-1.13 on OpenRouter). My OpenRouter key is in .env as OPENROUTER_API_KEY. Never print it. Read https://openrouter.ai/docs/guides/community/jev-tutorial first. Make a small web page on my computer where I paste a document. Split it into paragraphs and ask Jev, for every paragraph at the same time, one yes/no question: "Is this paragraph generic AI-written filler rather than a specific, human, first-hand statement?" Highlight the filler paragraphs in red with Jev's confidence, and show how many words it checked, how long it took and what it cost.
Links
- OpenRouterOne account, every model, pay per useopenrouter.ai
- Jev 1.13 on OpenRouterTypeSafe's decision modelopenrouter.ai
- Jev tutorialYour first decision callopenrouter.ai
- Claude CodeAnthropic's coding agentclaude.com
- awesome-jevWhat people are building with Jevgithub.com
- Made with JevShowcase of Jev buildsmadewithjev.com
Want one of these running on your own business?
An inbox, a lead form, a support queue. I set up the sorter, connect it to where the work actually happens, and add a safe fallback for the unsure cases.
