Billing
What an approved fix costs
Unlimited on your own inference key. $1.49 each on Novix’s, uncapped on Pro and capped at $98 on Solo.
The workspace subscription is flat, and on Solo it is nothing at all. The one variable left is whose account pays for the inference, and it decides whether an approved fix bills.
| Running on | What an approved fix costs |
|---|---|
| Your own inference key | Nothing. Unlimited, and nothing is metered. |
| Novix’s key, on Pro | $1.49 each, from the first one, and unlimited each billing period. |
| Novix’s key, on Solo | $1.49 each, and the spend stops at $98 in a billing period. |
A card before the first approval
A workspace running on Novix’s key needs a card on file before anyone approves a fix, and Solo reaches that point with no subscription behind it. Adding the card charges nothing: it opens Stripe’s own checkout, starts no monthly plan, and the first charge is the first fix somebody approves. Until one is added, the approve button refuses and says so.
On your own key the tokens already bill to your provider account, so Novix has no per-fix cost to recover. On Novix’s key the per-fix charge is what recovers it. There is no included pool and no setting in front of it: per-fix billing is how the plan works when Novix is paying for the tokens.
What never bills
- A diagnosis. Novix can read a hundred reports and find a hundred root causes and none of it is metered.
- A drafted fix your team rejects.
- A question back, a report that turns out not to be a defect, or a run that errored.
Only an approved fix bills, and it is metered once wherever the approval happened, including when you merged the pull request yourself.
The cap
The cap belongs to Solo alone. On Solo the spend stops at $98 in a billing period, which at $1.49 a fix is 65 approved fixes, and it is a hard stop rather than a warning. On Pro there is no cap at all: a Pro workspace on Novix’s key approves as many fixes as it likes and each one bills $1.49. The calendar month is used before a subscription period exists.
Bringing your own key
Point Novix at your own Anthropic, Kimi, OpenAI or OpenRouter account in settings, on the Pro plan. It is a cost, compliance and procurement choice rather than a quality one.
- Anthropic-shaped endpoints keep every engine feature, including prompt caching and the effort dial.
- OpenAI-shaped endpoints go through an adapter that has neither, so a run costs you more and the stages are not tuned for it.
The numbers on this page are also on the pricing page, in the same words.