Typed answers, paid by the request.
Experimental access to TypeSafe’s Jev System One model through Baking Bad’s paid service. Send a state and the questions you want answered about it; get probabilities, choices and rubric scores back. No API key, no subscription — an agent pays for each call over x402 or MPP.
The service exists solely to demonstrate agent payments. It is provided as-is and is not intended or permitted for production use. Neither TypeSafe nor Baking Bad accepts responsibility or liability for its use, outputs, errors, availability, losses or resulting decisions.
Never let Jev outputs determine real-world decisions affecting health, life, safety, financial or legal outcomes, or other material risks. Send synthetic or non-sensitive data only: evaluation data is forwarded through Baking Bad to TypeSafe. An agent must read /SKILL.md and obtain the user’s explicit confirmation of these terms before calling the service.
Three kinds of question, one call
Name each question, describe it, and Jev answers all of them about the same state. Combining questions keeps one paid request instead of several.
One number between 0 and 1 for how likely the answer is yes, instead of a word you have to parse.
The selected option name, the full probability distribution over your options, and a confidence value.
The probability-weighted average of your zero-based rubric levels, returned with the rubric, the distribution over it and a confidence value.
{
"state": "Help! My payouts have been failing for 3 days.",
"model": "jev-latest",
"questions": {
"is_urgent": {
"type": "noul",
"instructions": "Does this convey urgency?",
"criteria": {
"true": "Explicitly time-sensitive",
"false": "No urgency expressed"
}
},
"department": {
"type": "choice",
"instructions": "Which team should handle this?",
"criteria": {
"billing": "Payments, invoicing, refunds",
"technical": "Bugs, outages, integrations",
"sales": "Pricing, upgrades, new accounts"
}
}
}
}
{
"model": "jev-1.13.0",
"answers": {
"is_urgent": {
"type": "noul",
"noul": 0.92
},
"department": {
"type": "choice",
"choice": "technical",
"probabilities": {
"billing": 0.08,
"technical": 0.85,
"sales": 0.07
},
"confidence": 0.82
}
},
"usage": {
"input_tokens": 312,
"output_tokens": 48
}
}
How a paid call works
The service answers an unpaid request with the price for that request. Nothing is charged until the same body comes back with a credential.
Read /openapi.json for the contract, /SKILL.md for the agent guide, and GET /v1/models for model names. None of them cost anything.
Send the evaluation to POST /v1/systemone with no credentials. The 402 carries the authoritative price and signing fields for that exact body.
Repeat the unchanged request with payment-signature for x402 or authorization: Payment for MPP, and read the answers.
$ curl -i -X POST $ORIGIN/v1/systemone \ -H 'content-type: application/json' --data-binary @request.json HTTP/1.1 402 Payment Required payment-required: <x402 challenge for this exact body> www-authenticate: Payment <mpp offers> $ curl -X POST $ORIGIN/v1/systemone \ -H 'content-type: application/json' \ -H "payment-signature: $CREDENTIAL" --data-binary @request.json HTTP/1.1 200 OK { "model": "jev-1.13.0", "answers": { ... } }
Metered pricing
Priced from the request, not from a flat per-call rate.
Each evaluation is priced from a token estimate of the parsed request body and the number of questions in it, then clamped into the bounds published for the payment option you picked. A larger state costs more; a small one costs the option’s floor.
x-payment-info on the operation publishes the range below, which is discovery metadata
rather than a quote. The live 402 challenge for your exact body carries the authoritative
amount. If the body changes, ask for a new challenge.
Discovery is free: /openapi.json, /SKILL.md, /health and
GET /v1/models cost nothing.
- Batch-settlement floor$0.00001
- Per-request floor$0.0002
- Context budget64k tokens
- Model discoveryFree
Payment rails
The checked-in configuration offers these alternatives; the live challenge decides what is actually available for your request. Verify the scheme, network, asset, recipient and amount against your own spending authorization before signing.
Channel and voucher client. Preferred for repeated calls — it amortizes settlement instead of paying it per request.
Any client that can sign the exact offer. One on-chain settlement per call.
Needs this service’s Permit2 witness; a generic Permit2 implementation will not do.
A Solana-capable client on the offered network.
An MPP client on the offered Tempo currency and chain. A charge is one payment, not a session.