Cycling through all six. Tap any point to stop.

Measured on six things

JevModel lets you run the Jev model in your browser: 3 live runs without signup, then 100,000 free tokens. Get Choice, Score and Noul answers, build decision flows, run batches and call the Jev API.

Security

Probe shows HTTPS not enforced, no HSTS, no CSP, no frame protection, no X-Content-Type-Options, no Referrer-Policy. Only SPF and DMARC present (2/12 checklist). No published audits or bug bounty found.

Screenshots

1 / 1

JevModel

About this app

JevModel is a browser-based playground and API for the Jev structured decision model. It turns text or JSON into typed answers—Choice, Score, or Noul—with probabilities. Key features include visual decision flows, batch processing (CSV/TXT/JSONL up to 1,000 rows), and a shared token balance across playground, batch, and API. The free tier offers 3 daily runs without signup and 100,000 tokens after email verification. Token packs never expire and failed requests are not charged. Use cases cover moderation, spam filtering, support triage, routing, and prompt injection screening. It's an independent service calling the typesafe/jev-1.13 model via OpenRouter, ideal for developers and teams needing reliable, cost-effective decision automation.

Key features

1Typed decision outputs — Choice, Score, and Noul — with probabilities for each answer.
2Visual decision flows with branch-only billing: only the questions on the executed path are charged.
3Batch processing of CSV, TXT, or JSONL up to 1,000 rows, with CSV export.
4Shared token balance across playground, batch jobs, and API; token packs never expire.
5Free tier: 3 live runs per day without signup, 100,000 tokens after email verification.

Target customer profile

Drawn from the product itself, not from a survey.

Demographic

Software developers and API integrators

Pain points

Need reliable structured decision outputs without building custom models

Primary needs

Easy API integration, clear documentation, and predictable pricing

Demographic

Customer support and operations teams

Pain points

Manual ticket triage and routing slows response times

Primary needs

Automated intent detection and priority scoring

Demographic

Content moderators and trust & safety teams

Pain points

Scaling content review while maintaining accuracy

Primary needs

Fast, probabilistic classification for toxic or spam content

Scout Review

Written by AI from measured evidence, scored out of 100.

JevModel is a competent, well-scoped developer tool with a real idea behind it: typed decision outputs (Choice, Score, Noul) with probabilities, wrapped in a console, batch processor, and API that share one token balance. The onboarding is genuinely low-friction — 3 free runs without signup, 100k tokens after email verification — and the learning center is unusually deep for a product this young. The design is coherent and the accessibility baseline is strong. Where it stumbles: a 10-second mobile LCP that will lose impatient developers, a security posture that doesn't match the trust-and-safety use cases it's pitching, and a growth story that leans on workflow differentiation rather than a defensible moat. It's a solid early-stage product that needs to fix its performance and security fundamentals before it can credibly sell to teams.

Measure by measure

Usability

3 free runs without signup plus 100k tokens after email verification is a low-friction onboarding. Learning center has 15+ guides and the homepage demo runs live. Dense copy raises the learning curve.

Design

Dark developer-console aesthetic with embedded live demo, code snippets, and consistent branding across homepage, pricing, and guides. Dense but coherent; pricing grid is busy.

Speed

Lighthouse mobile 62/100 with a 10.0s LCP — well past the 2.5s 'good' line. Desktop 60/100. CLS is clean at 0 and TBT is fine at 150ms, so the bottleneck is a heavy hero asset, not JS.

Security

Accessibility

Lighthouse accessibility 97/100 with only one failing audit (color contrast). Language switcher present for multi-language. No published WCAG statement found, but the automated baseline is strong.

Growth

Real problem (structured decisions vs. prompt-and-parse), specific wedge (typed Choice/Score/Noul with branch-only billing), but the category is crowded with OpenAI and Google. Domain is ~0.1 years old.

Improvement areas

Mobile LCP of 10.0s — compress or lazy-load the hero asset and re-run Lighthouse.
Security headers: enforce HTTPS, add HSTS, CSP, X-Content-Type-Options, and Referrer-Policy (currently 2/12).
Homepage density: consolidate the live demo, illustrative demo, and mockups into a clearer hierarchy.
Pricing page: differentiate the six token packs with distinct value props instead of repeated bullets.

Design & Usability

The design language is coherent — a dark developer-console aesthetic that carries from homepage to pricing to guides, with a live playground embedded right in the hero. That's a real strength: you can try the model before you read a single word of marketing. The problem is density. The homepage stacks a live demo, an illustrative demo, a flow diagram, a batch mockup, a code block, and a use-case grid into one scroll. The pricing page then adds six token packs with identical bullet lists. Usability is genuinely good on the onboarding side — 3 free runs without signup, 100k tokens after email verification, and a learning center with 15+ articles. But the cognitive load of the marketing site undercuts the product's actual simplicity. A developer who just wants to see if the API works has to scroll past a lot of theater to find the quickstart.

Speed & Security

Speed is the weakest measured dimension. Lighthouse mobile lands at 62/100 with a 10.0s LCP — that's four times the 'good' threshold and the kind of number that makes a developer close the tab before the hero renders. Desktop is 60/100. The good news: CLS is 0 and TBT is only 150ms, so the bottleneck is a heavy hero asset (likely the embedded demo or a large image), not a bloated JS bundle. That's fixable. Security is more concerning: the HTTP probe found HTTPS not enforced, no HSTS, no CSP, no frame protection, no X-Content-Type-Options, no Referrer-Policy, and no Permissions-Policy — 2/12 checklist points, with only SPF and DMARC present. For a product that positions itself for moderation and trust-and-safety use cases, that's a gap between the pitch and the posture. No SOC 2, ISO, or bug bounty claims were found in the fetched content.

Accessibility & Growth

Accessibility is a genuine bright spot: Lighthouse scores 97/100 with only one failing audit (color contrast), and the site ships a language switcher for multi-language support. That's above-average for an indie developer tool. Growth is the harder call. The idea is real — structured decisions with typed outputs and probabilities solve a genuine pain point for anyone who's tried to parse free-form LLM output into a routing decision. The wedge is specific: branch-only billing on decision flows, token packs that never expire, failed requests not charged. But the category is crowded. OpenAI's structured outputs and Google's NL API are the named alternatives, and both have vastly more distribution. The differentiator here is workflow (visual flows, batch processing, one shared balance) rather than model capability — and that's a thin moat if TypeSafe ever ships a decent console. Domain age of ~0.1 years means there's no traction signal to lean on either way.

Conclusion

If you're a solo developer or small team who needs structured decisions without building a custom classifier, JevModel is worth a free-tier test — the 3-runs-no-signup path costs you nothing but a minute. If you're a trust-and-safety or support-ops team evaluating this for production, wait for the security posture to catch up to the pitch: enforce HTTPS, add HSTS and CSP, and publish something auditable. And fix the LCP — a 10-second mobile load is the kind of thing that quietly kills conversion before anyone reads a single feature bullet.

Suggestions for improvement

Enforce HTTPS and add HSTS, CSP, X-Content-Type-Options, and Referrer-Policy headers — the current 2/12 probe result undercuts the trust-and-safety pitch.
Fix the 10.0s mobile LCP: lazy-load or replace the hero demo asset, and re-run Lighthouse to confirm the improvement.
Consolidate the homepage — pick either the live demo or the illustrative demo, not both, and move the batch/flow mockups below the fold.
Differentiate the six token packs with distinct bullet points instead of repeating the same five lines on every card.

How we score

Alternatives comparison

Named competitors, point by point. Nobody paid to appear here or to be left out.

	JevModel	OpenAI API	Google Cloud Natural Language API
Structured output format	Typed Choice/Score/Noul with probabilities, purpose-built for decisions	JSON mode / structured outputs via schema, general-purpose	Sentiment, entity, and classification analysis; no decision-flow primitives
Visual decision flows	Canvas-based flows with branch-only billing	Not offered; chaining done in code	Not offered
Batch processing	CSV/TXT/JSONL up to 1,000 rows with CSV export	Batch API available with 24h turnaround	Batch via async API, no row-level UI
Pricing model	One-time token packs ($4.90–$99) that never expire; 100k free tokens	Pay-per-token, monthly billing, no expiry	Pay-per-unit, monthly billing
Free tier	3 runs/day without signup; 100k tokens after email verification	Limited free credits for new accounts	Free tier with monthly unit cap
Target use case	Moderation, spam, triage, routing, prompt-injection screening	General-purpose generation, reasoning, and classification	Sentiment, entity, and content classification