Lesson 6 · Ship it

A key, a curl, a response hands-on

Goal: make one real API call and recognise every part of the response.

Keys live at console.typesafe.ai/keys. The endpoint is a single POST (docs: API). Everything from the earlier lessons goes in the body unchanged:

curl -s https://api.typesafe.ai/v1/systemone \
  -H "Authorization: Bearer $TYPESAFE_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "state": "Hi, I have been trying to connect my Stripe account for 3 days and it keeps failing. Please help ASAP.",
    "model": "jev-latest",
    "questions": {
      "department": {"type": "choice", "instructions": "Which team should handle this",
                     "criteria": {"billing": "Payment or subscription issues", "technical": "Bugs or integration problems", "sales": "Pricing or account questions"}},
      "is_urgent":  {"type": "noul", "instructions": "The message conveys urgency or time-sensitivity"}
    }
  }'

Python and JavaScript in ten lines code

Goal: the same request as typed objects, with retries handled.

Python (3.10+): pip install typesafe-sdk or uv add typesafe-sdk. The client reads TYPESAFE_API_KEY from the environment (docs: Python SDK).

from typesafe_sdk import TypeSafeClient, Choice, Score, Noul

client = TypeSafeClient()          # AsyncTypeSafeClient for asyncio; RetryPolicy to tune backoff

r = client.system_one(
    state=transcript,
    questions={
        "outcome": Choice("What the caller wants from this call",
                          {"sales_lead": "Wants a quote, purchase or consultation",
                           "support": "Existing customer with a technical or billing problem",
                           "vendor_or_spam": "Selling something to the company",
                           "unclear": "Too little information to tell"}),
        "caller_mood": Score("How the caller comes across",
                             ["Friendly or neutral", "Impatient or worried", "Angry or hostile"]),
        "callback_requested": Noul("The caller asks to be called back or agrees to an appointment"),
    },
)
outcome = r.answers["outcome"]          # .choice, .probabilities, .confidence
mood    = r.answers["caller_mood"].score
cb      = r.answers["callback_requested"].noul
print(r.model, r.usage.input_tokens)

JavaScript / TypeScript (Node 20+): npm install @typesafe-ai/sdk (docs: JavaScript SDK).

import { TypeSafeClient, choice, score, noul } from "@typesafe-ai/sdk";

const client = new TypeSafeClient();   // TYPESAFE_API_KEY from env

const r = await client.systemOne({
  state: transcript,
  questions: {
    outcome: choice("What the caller wants from this call", {
      sales_lead: "Wants a quote, purchase or consultation",
      support: "Existing customer with a technical or billing problem",
      vendor_or_spam: "Selling something to the company",
      unclear: "Too little information to tell",
    }),
    caller_mood: score("How the caller comes across", ["Friendly or neutral", "Impatient or worried", "Angry or hostile"]),
    callback_requested: noul("The caller asks to be called back or agrees to an appointment"),
  },
});
const { choice: outcome, confidence } = r.answers.outcome;   // typed: outcome is one of the four keys

No code: one HTTP node and one IF no-code

Goal: wire Jev into an n8n or Make flow without writing a line.

There is no dedicated n8n node as of 2026-09-19; none is needed. The generic HTTP Request node does everything:

Then the routing from lesson 4 as n8n nodes:

What it costs, how far it scales, which version answers facts

Goal: estimate a month's bill in your head and know when to pin a version.

claude plugin marketplace add typesafe-ai/skills
claude plugin install typesafe@typesafe-ai
# other agents:
npx skills add typesafe-ai/skills --skill typesafe-ai

Shadow it before you trust it method

Goal: run a two-week evaluation that produces thresholds instead of opinions.

The community benchmark's recommendation was not "switch to Jev". It was: run a focused shadow evaluation on your own traffic first (benchmark brief, recommendation). Here is that evaluation as a procedure:

Final quiz, then ship the real thing quiz

Goal: leave with a wired decision and a plan to evaluate it.