Points
In short
A point is Sapilon's fixed-price unit of AI usage with one published rate, so an interaction costs a predictable whole number instead of a token bill you can only understand afterwards.
AI work is priced in points: one published rate, whole numbers, quoted in terms you can reason about before you press send.
Why not tokens
Token pricing is accurate and useless at the moment of decision — nobody can estimate the token cost of “add cancellation to orders” beforehand. Points move the estimate to where the decision is, so the meter never becomes a reason to avoid asking.
What makes a request cost more
- Scope. More files read and changed, more points.
- Depth. Harder problems get more capable models and more deliberation, and cost correspondingly more. Routine work is routed to cheaper settings automatically.
- Retries. A vague brief that needs three attempts costs three attempts — which is the real argument for briefing well.
- Context. Very large conversations cost more to carry. Starting a new conversation for a new subject is usually cheaper than continuing an old one.
What does not cost points
Reading your own project, browsing versions and diffs, opening artifacts, and everything in the UI that does not invoke the agent.
Keeping the bill down
The single biggest lever is scope per turn: one outcome per request, reviewed and accepted, beats a sprawling multi-part instruction that has to be partly undone.
Next: What a turn costs · Wallet and top-ups · AI pricing