BRING YOUR OWN KEY

Add a Cerebras, OpenAI or Anthropic API key to continue

Held in memory for this tab. Sent only to this API and your provider.

AN UNOFFICIAL MOCK · STILL ON THE WAITLIST

You’re not called Ben?
Still on the waitlist?

Try a TypeSafe-like API for people who can’t get past their waiting list.

This is an unofficial mock for those of us not lucky enough to be in the typesafe.ai beta yet. It imitates the API with general-purpose models. Expect much poorer results than TypeSafe—not a preview of their actual model or quality.

Try it with a Cerebras key. Its gpt-oss-120b model is the recommended way to get a feel for this mock’s performance. These timings are not a benchmark of TypeSafe.

UNOFFICIAL MOCKOFFICIAL SDKSBRING YOUR OWN KEY
INPUT / 001
CUSTOMER MESSAGE

I was charged twice.
Can you help?

{ }
MOCK RESPONSE
{
"department": "billing",
"confidence": 0.97
}
ILLUSTRATIVE OUTPUT · NOT TYPESAFE INFERENCE++
POWERED BY YOUR KEYCerebras gpt-oss-120b · recommendedOpenAI GPT-5.6 LunaAnthropic Claude Haiku 4.5Built with AI SDK 7 ↗
01 / THE PLAYGROUND

A little language.
A useful decision.

Pick an example. See what comes back.
Connect a key to make it your own.

YOUR INPUT
QUESTIONS Three questions. One request.
department

Which team should handle this request?

choice
urgent

Does the customer need action today?

noul
frustration

How frustrated is the customer?

score
Your provider key. Your live result.
03 / PUT IT IN YOUR CODE

Their SDKs.
This mock’s URL.

Use the current official TypeSafe JavaScript or Python SDK unchanged. Set baseURL (JavaScript) or base_url (Python) to https://typesafe.severin-marcombes.com and pass your provider key as apiKey or api_key. Your systemOne / system_one calls stay the same.

npm install @typesafe-ai/sdk
Explore the original API reference ↗

Code follows your selected playground example.
Cerebras is recommended. The examples use CEREBRAS_API_KEY; you can also supply an OpenAI or Anthropic key.

curl "https://typesafe.severin-marcombes.com/v1/systemone" \
  -H "Authorization: Bearer $CEREBRAS_API_KEY" \
  -H 'Content-Type: application/json' \
  --data '{
  "model": "jev-latest",
  "state": "I was charged twice for my subscription. Please refund the extra payment — I need this sorted out today.",
  "questions": {
    "department": {
      "type": "choice",
      "instructions": "Which team should handle this request?",
      "criteria": {
        "billing": "Payments, invoices and refunds",
        "technical": "Bugs, outages and integrations",
        "sales": "Pricing and new purchases"
      }
    },
    "urgent": {
      "type": "noul",
      "instructions": "Does the customer need action today?"
    },
    "frustration": {
      "type": "score",
      "instructions": "How frustrated is the customer?",
      "criteria": [
        "Calm",
        "Frustrated",
        "Very angry"
      ]
    }
  }
}'
02 / SEE IT DO SOMETHINGINTERACTIVE DEMO

“Make yourself
at home.”

Lights, temperature, music. Just say what you want. Watch one request become a set of decisions—and a home that responds.

Step inside the smart home

Recreated from TypeSafe’s smart home demo.
Virtual devices. Mock API decisions.

HOME, SWEET SMART HOME
Living room
Lights on
Kitchen
22°
Bedroom
Turn off all the lights in the house

Choice

Pick from possibilities.
Route to the right next step.

"choice": "billing"

Noul

A yes-or-no question.
With room for uncertainty.

"noul": 0.94

Score

Describe your scale.
Get a nuanced position on it.

"score": 1.08
KEEP CODE IN CONTROL

Models judge.
Your code decides.

Probability estimates and confidence come from your chosen model. This mock does not reproduce TypeSafe’s model, quality or calibration.