Software Factory fundamentals

Factory costs

Connect token and compute spend to completed tasks and workflow choices.

Factory cost reporting shows the token and compute spend recorded for tasks that closed in the selected period. Use the total view to understand the mix of spend and the per-task view to see whether the cost of completing work is changing.

The Factory Costs view showing total cost split between model tokens and compute across the reporting period. — Total cost separates managed model usage from execution compute.

Understand the cost categories

Tokens: Recorded model input and output usage for supported managed models.
Compute: Recorded machine time used by supported managed execution.
Total cost: The token and compute amounts attributed to the completed-task view.

Connected model, cloud or continuous integration providers may charge outside SuperPlane. Their charges are not automatically represented by these charts. Use provider budgets alongside the Factory report.

Read per-task cost as a trend

Per-task cost shows the median spend for a task that closed, grouped by day. It is more useful than the total when task volume changes between periods.

The Per task costs chart showing median token and compute spend for tasks closed each day. — Per-task cost makes a rising cost visible even when only a few tasks close.

A higher value is not automatically a problem. The period may contain broader tasks or a stronger verification policy. Compare tasks with similar scope and completion rules before changing the workflow.

Trace a cost increase to its cause

Review the affected tasks and their automation runs. Look for:

repeated agent attempts with unchanged context
a larger model used for task classification or summaries
prompts that include unrelated repository or run history
long machine sessions before a failure is reported
verification loops caused by the same missing requirement
tasks that closed without an accepted result

The automation table can show which workflow has high average or total cost. The task record then shows whether the spend came from useful work, recovery or avoidable repetition.

Reduce avoidable spend

Start with workflow design before weakening model quality or tests.

Filter unsupported tasks before an agent runs.
Match the model to the stage and task complexity.
Use deterministic components for routing, status changes and notifications.
Pass focused context between correction attempts.
Stop a retry loop when the next step requires a person.
Set concurrency that avoids duplicated or conflicting repository work.

Review Costs and budgets for billing behavior, allowances and external controls.