Authentication¶

Every call needs a key. Sign in at japanfold.aiand.com/account/token and the page shows one. A new account starts with $100.00, once.

The catalog is the one exception, because a client has to read it before it can ask for anything:

curl -s https://api.japanfold.aiand.com/v1/models

Your API key¶

Sign in at japanfold.aiand.com/account/token with your email address. A six-digit code arrives by mail, and the page makes a key the moment you are in.
Copy it. It is shown once, because only its hash is kept; if you lose it, revoke it on the same page and make another. Make one per machine or agent and name it after where it lives, so you know which one to revoke.
Send it on every call as Authorization: Bearer (or X-API-Key). The examples in these docs read it from $JAPANFOLD_API_KEY:

export JAPANFOLD_API_KEY=jf_live_your_key_here
curl -s https://api.japanfold.aiand.com/v1/predictions \
  -H "Authorization: Bearer $JAPANFOLD_API_KEY" \
  -H 'Content-Type: application/json' \
  -d '{"model":"boltz2","sequence":"MKTAYIAKQRQISFVKSHFSRQLEERLGLIEVQ"}'

A job belongs to the account that submitted it. Poll, download and delete it with any key on that account; another account's id returns 404, and so does an id that does not exist. A key that is present but invalid or revoked is rejected with 401.

Billing¶

Work is priced in US dollars for the time a chip actually holds your job: $0.26 per processor-hour. Nothing is quoted or reserved before a run, so what you are charged is what the meter measured, and a job that fails costs nothing.

When the balance reaches zero the API answers 402 with out_of_credit: true and a link to top up. A job already queued still runs. How billing works in full: https://api.japanfold.aiand.com/billing. Your balance, keys and usage are on your account.

What you spent is GET /v1/usage. Its summary covers the whole period you ask for (from/to as dates, or since/until), not just the page of rows: amount is the total in dollars, currency is usd, and by_model splits it per model. Every *_micros field in the API is the same money in millionths of a dollar, exact and safe to add up; amount beside it is ready to read.

curl -s "https://api.japanfold.aiand.com/v1/usage?from=2026-10-01&to=2026-10-07&limit=1" \
  -H "Authorization: Bearer $JAPANFOLD_API_KEY" | jq .summary
# {"amount": 1.84, "currency": "usd", "jobs": 37, "billable_s": 25476.9,
#  "by_model": {"boltz2": {"jobs": 30, "billable_s": 21012.3, "amount": 1.52, ...}, ...}, ...}

The CLI prints the same: japanfold usage --days 7.

Ownership and limits¶

Your balance is the limit. There is no cap on how many jobs you may queue, how many designs you may ask for or how much you may download, because every one of those costs money and the balance already bounds it.

What is still refused is what the hardware cannot fold: the per-model residue ceilings on Models & limits, live at GET /v1/models. Over one of those you get 400 (see Errors).