User repository on GitHub
Daily demo applying TypeSafe's Jev model to Spain's official gazette (BOE).
- Forks
- 0
- Last pushed
- Updated Oct 2026
Repo stats from the GitHub API, cached Oct 2026.Homepage
CommunityJev Demos & Experiments
Jev Boe Demo
Spanish-language demo running Jev over Spain's official gazette (BOE) daily. One call per disposition scores relevance 0-3, probes 13 policy categories with Noul, and picks the paragraph explaining the change, so the summary is selected rather than written. Astro on GitHub Pages, JSON data, no login.
- Stars
- 1
- Language
- TypeScript
- Published by
- Community
- Author
- Tatuck
- Last verified
- Oct 2026
About this project
Demo de Jev, el modelo System One de TypeSafe, aplicada al Boletín Oficial del Estado. Cada mañana un pipeline descarga el BOE del día (secciones I, II-B y III), lee cada disposición y le hace a Jev, en una sola llamada por texto, un conjunto de preguntas cerradas:
Quoted from the project's README. Read the full README.
Works forJev for Classification
Used inJev for Software & DevTools
- Languages
- TypeScript
Highlights
- One Jev call per disposition: a 0-3 relevance Score, 13 category Nouls assigned above 0.6, and a Choice returning the literal BOE paragraph.
- No text is generated, so the summary shown on the site is an original paragraph the model picked out of the provision, not prose it wrote.
- A full day, 50 to 80 provisions and roughly 300k tokens, is analyzed in 5 to 15 seconds with 6 calls in parallel for about $0.01, timing each call.
- Raw probabilities are stored in the per-day JSON, so thresholds can be retuned and the site rebuilt without paying for another Jev call.
Watch out
The classification cutoffs are hand-set constants: relevance at 2 or above counts as relevant, a category is assigned above 0.6 probability, and LOW_CONFIDENCE sits alongside both in pipeline/analyze.ts. Parsing is BOE-specific, reading only sections I, II-B and III, skipping the index and truncating each disposition at 20k characters, so a long provision is judged on a prefix. The taxonomy and question wording live in pipeline/questions.ts and are written in English against Spanish state, which is a deliberate choice worth revisiting for other corpora. Calibration is by eye: pnpm calibrate --limit 15 prints a table for adjusting questions and thresholds. Unattended operation needs TYPESAFE_API_KEY as an Actions secret and costs roughly $0.01 per gazette day. The README also states the code was written with Claude Code. Copy reviewed against the repository README on 2026-10-04, not from running it.
Questions about this entry
Short answers drawn from the linked project or page.
No, and that is the design. Jev generates no text, so the pipeline asks a Choice question over the provision's paragraphs and the summary displayed is the literal BOE paragraph the model selected. Everything else on a disposition is a score, a probability or a category label.
The README reports 50 to 80 provisions, around 300k tokens, analyzed in 5 to 15 seconds with 6 calls in parallel at about $0.01, with real per-call latency and daily totals recorded. It is one API call per disposition, which is what keeps the whole day inside a scheduled job.
Yes. The per-day JSON keeps the raw probabilities, so CATEGORY_THRESHOLD, RELEVANT_SCORE and LOW_CONFIDENCE in pipeline/analyze.ts can change and the Astro site rebuilds from stored data alone. Categories and question wording are kept separately in pipeline/questions.ts, and pnpm calibrate prints a readable table for eyeballing both before you commit.
The parsing. pipeline/boe.ts reads the BOE XML and sumario shape, skips the index, truncates each disposition at 20k characters and analyzes only sections I, II-B and III. Questions go to Jev in English with the state in Spanish, and pnpm calibrate exists precisely because the taxonomy and thresholds need checking by eye. The README also flags that the code was written with Claude Code.
Run pnpm install, copy .env.example to .env and set TYPESAFE_API_KEY, then pnpm pipeline --today or --date 2026-09-17, with --force to reanalyse and --backfill 7 for several days. Adding --dry-run --verbose analyses without writing to data/ and prints one line per disposición as it arrives.
One call per disposición asks a 0 to 3 Score for how much it changes life for ordinary people, where 2 or above is relevant; thirteen Noul questions for categories such as housing, employment and taxes, assigned above 0.6; and a Choice over the paragraphs for the one that best explains what changes and for whom.
The site deploys to GitHub Pages. Create a TYPESAFE_API_KEY secret under Actions, set Pages source to GitHub Actions, and run the BOE diario workflow manually or wait for the 07:15 UTC cron from Monday to Saturday. It analyses the day, commits data/, builds the static Astro site and publishes it.
More Jev resources like this
98GitHub stars
Self-hosted Docker service and web UI that packages Laya structured-decision inference behind a compatible System One endpoint.
Jev Demos & Experiments#community#jevbest#catalog-import
Community
1Panel-dev/laya-serverPython
1Panel-dev/laya-serverPython20GitHub stars
General-purpose System One agent where Jev chooses tools and arguments while Jot executes them and loops until it can reply.
Jev Demos & Experiments#community#awesomejev#catalog-import
Community
runta-dev/jotTypeScript
runta-dev/jotTypeScript23GitHub stars
Browser drone-fleet lab where TypeSafe Jev steers each drone as fast reflexes and an optional planner offers one-use strategy advice when confidence drops, without pausing control.
Jev Demos & Experiments#community#jevbest#catalog-import
23GitHub stars
Chrome extension that sends each X post to Jev with six typed questions in one call near the viewport, then renders a behavior chip under the post plus a session cost and latency panel.
Jev Demos & Experiments#community#jevbest#catalog-import
Community
oso95/x-scannerTypeScript
oso95/x-scannerTypeScript14GitHub stars
Manifest V3 Chrome extension in TypeScript that sends each LinkedIn post to Jev and collapses bait, self-promo, off-topic, and sponsored posts into a one-line bar with reasons and scores.
Jev Demos & Experiments#community#jevbest#catalog-import
Community
stefw/lkcleanTypeScript
stefw/lkcleanTypeScript11GitHub stars
Standalone Flue example that routes support chat through Cloudflare Workers AI, using gpt-4o-mini for chat and Jev for typed intent and urgency routing.
Jev Demos & Experiments#community#awesomejev#catalog-import
Community
matthewp/flue-jev-demoTypeScript
matthewp/flue-jev-demoTypeScriptFrom the Jev guides
Original write-ups on what this entry is for.
A game tick is a state plus a legal action list. How Clash Royale, Minecraft, and the earlier Mario, Pokémon, StarCraft, and drone loops ask Jev one step at a time.
Jev Demos & Experiments#gaming#agents#control
Read article
From the Jev community
Posts from builders shipping with Jev right now.