Costs and Limits

Agent Mode can make several paid model calls for one goal. Paid planner calls use your Nyquest wallet, including when your normal chat is set to BYOK; Agent Mode does not use your saved provider key.

Before each call

Nyquest reserves a conservative amount based on the complete encoded planner context, tool definitions, the model’s published prices, and its output limit. The same check applies to final synthesis. A new call is refused if the reservation exceeds available wallet funds or the run’s remaining dollar budget.

The reservation temporarily reduces the available balance. When usage is known, settlement releases the reservation and applies the actual user charge and any eligible compression credit in one transaction. The call’s user charge cannot exceed its reservation. If the provider outcome is unknown, the reservation remains until reconciliation; do not assume the run was free or immediately restart it. Contact support with the run ID if it remains pending.

Dollar ceiling

• New users and users with no priced completed runs: $1 per run.
• Otherwise: 5× the last 30 days’ average completed-run cost, with a $1 minimum.
• Settled charges plus outstanding reservations count toward the ceiling. The check can stop a run before its displayed settled cost reaches the ceiling.
• A partial synthesis is attempted only if its own reservation fits. If it does not, the run stops without another paid call; the existing timeline remains available.

Historical runs from before the September 5, 2026 accounting correction retain their recorded values. They have not been retroactively charged or repriced.

Context and step limits

• Default: 6 planner steps; per-run setting 1–20.
• A lower project step cap takes precedence. 20 is the absolute loop limit.
• A separate final synthesis may be attempted after a stop or the step limit, subject to the same budget checks.
• Each planner response is limited to 2,000 output tokens.
• The complete encoded planner request is limited to 32 KB. Large goals or accumulated observations can stop a run; use smaller inputs or split the task.

Default planner models

The model selector or project preference can override the default. A planner must have an available catalog entry with published prices. Its actual served usage determines settlement; rates and availability can change. Free account access does not mean every planner model is free.

Reading the cost

The timeline shows cumulative settled cost, including sub-cent values. The run detail API includes token totals, net user cost, compression savings in dollars, and billing.pending_calls / billing.held_usd. The Usage page and wallet ledger record paid calls. Pending settlement can update the stored totals after a run ends.

Runtime depends on model latency, tools, and step count; there is no guaranteed time or typical dollar quote. Review the model’s input/output prices and begin with a small goal and a low step limit.

Project controls

Projects can restrict allowed tools, lower the maximum steps, and set a preferred model. These controls shape the work; the wallet and per-run dollar checks still apply.