Skip to content
KlusAI Docs

Credits and billing

razorBridge is prepaid. An account holds a balance of credits denominated in euros — one credit is one euro — and every GPU blade and every inference call draws that balance down. There is one balance and one ledger per account: no separate bill for compute and inference, and no invoice to reconcile afterwards.

Where credits come from

Section titled “Where credits come from”
  • Signup credit. An account is created the first time you sign in to the web app or approve a CLI login, and it receives €10 immediately. The entry shows up in the ledger as Signup grant.
  • Grants. Additional credit is added by KlusAI on request. Email hello@razorbridge.eu with what you need it for; grants for courses and competitions are the normal route today.
  • Organization accounts. If you belong to an organization’s account, whoever manages its billing controls its balance; your blades draw on it.

Credits do not expire.

How GPU runtime is charged

Section titled “How GPU runtime is charged”

Each GPU type has a published rate in euros per GPU-hour. The rate is read from the live pricing catalogue, which is what the pricing page and GET /v1/gpu-types/ both show. It is per blade, so an event using ten blades bills ten times the rate.

The charge is calculated from the blade’s exact elapsed wall-clock time, converted to fractional hours and multiplied by the rate. There is no rounding up to the minute or the hour and no minimum charge — a 7-minute blade costs about 7/60 of an hour.

Timing matters when you read your balance:

  • While a blade runs, its accumulated cost is recomputed every few minutes and shown on the session row. It is not yet in the ledger.
  • The ledger entry is written once, when the blade stops. For an event, one entry covers all of that event’s blades.

So a balance that looks untouched mid-session is normal, and the in-flight cost is still counted against your monthly cap and against the balance check that stops a session when it would run out of money.

How inference is charged

Section titled “How inference is charged”

Inference gate calls are billed per token, based on the model used, and converted to euros. Each call writes a usage record and, if it costs anything, deducts from the same balance with the source inference. Charges appear as soon as the call completes — unlike GPU runtime, there is no wait until teardown.

See Inference gates for per-key restrictions and budget caps.

Reading your ledger

Section titled “Reading your ledger”
Terminal window
rb credits
rb --json credits    # machine-readable

Or open Credits in the web app for the same balance and history. Every entry has a type (credit or debit), an amount in euros, the balance after it, a description and a source that tells you what kind of movement it was:

Source What it means
grant Credit granted, including the €10 signup credit
manual An adjustment made by KlusAI
invoice Credit against an invoice
voucher Credit from a voucher
stripe Credit from a card payment (not in use yet)
consumption A debit for GPU blade runtime
inference A debit for inference gate usage
refund Credit returned to the account

Ledger entries are immutable: a correction is a new entry, never an edit. A historical entry always reflects the rate that was billed at the time, even if the published rate has changed since.

Running out

Section titled “Running out”

Low balance is enforced in three places, in increasing severity:

  1. A warning email goes out — at most once a day — when your effective balance is worth less than about an hour of your current burn rate.
  2. New work is refused. Starting a session that your balance cannot cover fails up front with the amount needed, and an inference call with an empty balance returns HTTP 402 Insufficient credits.
  3. A running blade is stopped. If its accrued cost reaches your available balance, the blade is torn down at the next metering cycle rather than being allowed to run up a debt.

Account guardrails

Section titled “Account guardrails”

Beyond the balance itself, each account carries limits that cap accidental spend — by default 2 concurrent self-serve blades, an 8-hour maximum session, and €50 of spend per calendar month. Hitting the monthly cap produces an explicit error naming the cap and how much of it you have used. Ask KlusAI to raise them if your course or project needs more; see GPU blades and sessions.

Section titled “Related pages”