Last updated: 2026-09-20

How to Get Access to Jev

Jev is in early access. The official waitlist works — 140,000 users were cleared within about 36 hours of launch — but if you'd rather start today, three providers already serve Jev with no waitlist. All four routes end at the same model.

Option 1: the official TypeSafe waitlist

1. Sign up at typesafe.ai with your email.
2. Wait for your invite — waves have been clearing fast since launch (early testers reported hours, not weeks; no guarantee).
3. On approval you get a key for console.typesafe.ai, which includes the web playground and API key management.

Option 2: Vercel AI Gateway (fastest, no waitlist)

Jev is live in the Vercel AI Gateway as typesafe-ai/jev since launch day. If you already build on Vercel, there's no separate signup — call it through your existing account. Pricing matches TypeSafe's native $0.042/1M rate.

Option 3: Cloudflare Workers AI

Cloudflare lists typesafe/jev with per-request edge billing. Note the context window here is 32K (not 64K) — chunk long states first.

Option 4: OpenRouter

OpenRouter serves typesafe/jev-1.13 at $0.042/1M input, $0 output, 64K context. Useful if you want one key for Jev plus your other models, and for A/B testing decision quality against chat models side by side.

The 5-minute quickstart

1. Try the playground (zero code)

Approved users can open console.typesafe.ai/playground: paste a text state, write a plain-language question, pick the answer type (Choice / Score / Noul), and run. Results come back sub-second with confidence scores. The official starter: paste “I've been trying to connect my Stripe account for 3 days and it keeps failing. I'm losing sales.” and ask “Does this message express urgency?”

2. Call the API

curl https://api.typesafe.ai/v1/systemone \
  -H "Authorization: Bearer $JEV_API_KEY" \
  -d '{
    "model": "jev-latest",
    "state": "The checkout page throws a 500 on Safari only.",
    "questions": {
      "category": { "type": "choice",
                    "options": ["billing", "technical", "sales"] }
    }
  }'

3. Use an SDK

Official Python and JavaScript/TypeScript SDKs are available, plus the langchain-typesafe package and the Vercel AI SDK integration (@ai-sdk/typesafe). The TypeScript evaluate pattern:

import { evaluate } from '@typesafe-ai/sdk';

const verdict = await evaluate({
  model: 'typesafe-ai/jev',
  state: ticketText,
  questions: {
    category: { type: 'choice', options: ['billing','technical','sales','spam'] },
  },
});

// Your code owns the policy:
if (verdict.category.value === 'spam' && verdict.confidence > 0.90) archive(ticketText);

Rate limits & production notes

• Early-access defaults: ~250,000 tokens/sec and ~1,200 requests/min; enterprise tiers available.
• Keep states lean — strip HTML and metadata. Fewer tokens, better accuracy.
• Ask atomic questions; combine probabilities in code.
• Regression-test calibration on 100–500 historical cases before automating.

Not sure what to build first? Start with ticket triage or email routing — low risk, immediate payoff. And check when Jev shouldn't be used at all.