What a credit buys
What a plori credit actually buys: measured run costs from production, what never draws credits, how to see each run's cost, and why billing is credits instead of hours.
Your wallet is one balance, held in US dollars, that covers everything a plori agent does: model thinking, run time past your included minutes, and any disk you keep beyond your plan. You watch one number, never tokens. Model usage bills at what the model call actually costs, at cost with no markup, so a light job stays cheap on any plan.
Here is what real runs cost on production, measured August 2026 across the first week of per-run metering (all real accounts, our own test traffic excluded):
| Measured on production, August 2026 | Cost |
|---|---|
| Half of all runs cost | $0.05 or less |
| Three in four runs cost | $0.20 or less |
| The heaviest single run | $0.77 |
A run is one turn. You send a message. The agent thinks, runs commands, edits files, and replies. The median run is a question answered or a small edit landed. A task the agent takes a few turns over, like a bug fix, a small feature, or a research pass, typically lands between $0.20 and $0.40, thinking and run time included. These figures refresh as the data grows.
How do I see what a run cost?
plori runs shows each run's net cost in its SPEND column. The dashboard billing
page shows the net spend for recent runs. The REST run endpoint and the MCP results
from invoke_agent and get_run_result also report the run cost.
Use the REST run endpoint or the MCP result for gross spend and token counts by role: executor, advisor, and reviewer. These results also report the advisor calls used and allowed. They include the advisor spend cap when you set one.
Plori automatically refunds the charges for a run that stops because of a platform
fault, so its net charge is zero. Automatic refunds have a per-account rolling
24-hour limit. The REST and MCP results report the net amount as credits,
consumption before the refund as gross_micro_usd, and the returned amount as
refunded_micro_usd. refund_reason records why the run qualified for a refund.
If the account reached its limit, refund_withheld_reason records why the charge
remains.
The dashboard billing ledger shows the refund as a separate entry.
What never draws credits?
- Run time, for real use. Every registered account includes 2,000 minutes of run time each month. Past that, run time meters by the minute. The anonymous trial is the exception and bills run time from the first minute.
- Sleep. An idle agent scales to zero and spends nothing on compute.
- Disk within your plan. A sleeping agent inside its included disk costs nothing.
- Almost every tool call. The web tools carry small metered rates, shown live on the pricing page.
- Model usage on your own key. Bring your own key and the model line disappears from your plori bill.
What does a month of usage cover?
Free ($5 a month) covers a dozen or two typical tasks. Pro ($20) covers a real task or two every day, all month. Power ($120) keeps several agents busy daily. Plans and the full rate card live on the pricing page.
Why dollars instead of hours?
Because what a run costs is mostly the model's own bill, not the clock. Two runs of the same length can differ by an order of magnitude depending on which model worked and how hard it thought. Hour pricing would make the slow run the expensive one and reward nobody for being fast. Your balance tracks the real cost, and the per-run receipts keep that claim checkable.