Ask most AI products what they did on your behalf last Tuesday and you will get a chat transcript. A transcript is a story the system tells about itself after the fact. It is not evidence.
Reserve, then settle
Klaren AI works the other way round. Before an agent runs, it reserves credits against your balance — an entry naming the product, the job, and the ceiling. When it finishes, that reservation settles to what the work actually cost. Both entries are append-only. Nothing is edited, nothing is deleted, and a failed run settles to zero rather than vanishing.
- Work that cannot reserve does not start — the request is refused before anything runs.
- No balance goes negative, and nothing silently degrades to a cheaper model to fit.
- Every entry is attributed to a product and a job, readable as it happens.
- The same ledger backs the invoice, so what you are charged and what you can inspect cannot disagree.
If the only record of what an agent did is the agent's own account of it, you do not have a record.
Klaren AI platform team
Why this is a product decision
It would be cheaper to meter usage in aggregate and reconcile monthly. We built reservations because delegation only becomes reasonable when it is reviewable — the ledger is what turns handing work to an agent from a leap of faith into a decision you can audit afterwards.