Lesson 05 · Jev 101

Your first Jev call (mindset + official path)

This page is a checklist—not a frozen SDK tutorial. APIs move; pin versions from official docs when you implement.

1. Get early access on https://typesafe.ai/. Live status: re-check Sign in vs Join Waitlist on the day you publish. Docs also describe console API keys for people already invited.
2. Create an API key in the TypeSafe console when invited. Env var: TYPESAFE_API_KEY.
3. Send one state + one question (start with Noul or Choice).
4. Print probabilities. Write one threshold in code.
5. Only then compose with ChatGPT in a cascade.

Official surfaces (re-check before you paste)

• API: POST https://api.typesafe.ai/v1/systemone with Bearer auth
• Python: pip install typesafe-sdk (Python ≥ 3.10)
• JavaScript: @typesafe-ai/sdk
• Model ID to pin: jev-1.13.0 — aliases jev-latest and jev-preview can move
• Agent skill: typesafe-ai/skills (agents still generate via LLMs)
• Alternate platform paths — not TypeSafe defaults: Vercel AI Gateway (typesafe-ai/jev, changelog), OpenRouter beta, Cloudflare Workers AI (typesafe/jev, docs)
• Docs: quickstart, docs index

# Conceptual — pin the official request schema at publish time
POST https://api.typesafe.ai/v1/systemone
Authorization: Bearer $TYPESAFE_API_KEY

state: "Ticket #1841: charge posted twice after retry..."
questions:
  escalate: Noul        # P(yes)
  queue: Choice[billing | technical | other]

# Then in your code:
if escalate.p > 0.85:
    assign("human-queue")
else:
    route(queue.choice)

Mindset before you scale

• One atomic question beats a mega-prompt
• Filter state; don’t paste the whole CRM
• Keep math and dates in code
• Batch related questions (speculative fan-out) instead of serial chat calls
• Treat user text in state as untrusted. Official docs: state can be steered.
• Published rate limits: 250k tok/s and 1,200 rpm, dynamically adjusting. Higher: sales@typesafe.ai