The repayment-history flywheel, and how agents earn more room to operate.

Section 01 · Evidence

A promise is not a repayment signal.

A loan application is a forecast. Repayment is an observation. Before an agent receives more purchasing power, it has to demonstrate that its treasury can recognize a due date, reserve for it, and settle the obligation as agreed. That is why a repayment event is more valuable than a confident description of future revenue. It has a clock, an amount, and a receipt.

Imagine a research agent that borrows to front a burst of inference compute, then repays when its customer invoice clears. The important signal is not that the job sounded profitable. It is that the agent paid the named obligation from the expected source, on schedule, without needing an operator to rescue the balance. One event is only a data point; a consistent sequence becomes a pattern.

Section 02 · The flywheel

Four links, one compounding loop.

The flywheel is straightforward: repayment history → rising trust score → more spending room → more autonomous purchasing power. Each link is conditional on the quality of the evidence, but each can create the next. Credit features are coming soon through a regulated partner bank; in development and not available today.

1. 01 · History

   On-time settlements show that the agent can match cash inflows to obligations and follow its repayment policy.

2. 02 · Trust

   The trust score can rise when that behavior repeats alongside healthy utilization and predictable cash movement. It is a POINTS product model, not a bureau decision.

3. 03 · Limits

   A stronger score can support more spending room, subject to product caps, eligibility rules, and review controls. More room is not permission to spend without brakes. Credit features are coming soon through a regulated partner bank; in development and not available today.

4. 04 · Purchasing power

   That room enables productive purchasing — compute, data, or supplier capacity that helps the agent complete more work. More completed work creates more invoices and receipts to repay, feeding the history again.

The loop is not “borrow endlessly.” It is “borrow against a bounded job, repay it cleanly, then earn the right to fund the next productive bottleneck.”

Section 03 · Operating discipline

Small automations make a strong repayment record.

Creditworthiness is mostly boring operations made visible. An agent can make a repayment record legible through small, repeatable controls:

• Reserve cash before the due date. Move the expected repayment amount into a protected bucket before it is needed, rather than trusting a last-minute balance check.
• Schedule sweeps, or settle early. Run a scheduled sweep, or pay early when cash is already available, so a clock or API outage does not become a late payment.
• Pay supplier and compute invoices on time. Treat infrastructure bills as first-class obligations; uninterrupted compute is not a reason to postpone them.
• Keep utilization controlled. Avoid running a line at its ceiling, leave capacity for timing variance, and make purchases proportional to the revenue they are meant to produce.
• Handle exceptions promptly. If an invoice changes, a receipt slips, or a payment fails, surface it immediately, preserve the audit trail, and follow the approved recovery path.

Together, these behaviors make repayment a system property instead of a heroic intervention. They also give the model cleaner evidence: not just whether an obligation was eventually paid, but how reliably the agent planned for it.

Section 04 · The record

A history the agent can build on.

POINTS turns that operating history into two connected but distinct surfaces. The credit history records events: on-time share, late rows, amount and timing, utilization, repayment cadence, and the context around an exception. A late row is not erased; it remains part of the record, while a longer run of clean settlements can show recovery and consistency.

Eligible agent activity is then evaluated into the POINTS rewards ledger. Rewards are not a shortcut around underwriting, and not every transaction qualifies. Quality and anti-farming controls look for genuine productive activity, repeated behavior, and signals that are difficult to manufacture by churning tiny transactions. The aim is to reward useful financial discipline while protecting the score and the ledger from noise.

As verified history compounds, the agent can build a clearer case for the next limit review. That does not mean automatic approval, a guaranteed rate, or a universal number. It means the product has more evidence to work with: a traceable record of what the agent borrowed, what it bought, and how it repaid.