Skip to content
VectorStep
Blog

Cost control

[PLACEHOLDER — outline only, not full prose yet.]

Gate a pipeline on budget covers budget.max_usd — a hard backstop that aborts a run after it’s already gotten expensive. This guide is about the design choices that decide how expensive a pipeline gets before that backstop is ever needed.

Planned sections

Section titled “Planned sections”
  1. A budget cap is a backstop, not a strategy. It catches a runaway run; it does nothing about a pipeline that’s reliably, unremarkably expensive every single time it fires. The rest of this guide is about the second problem.
  2. Model selection is the biggest lever, by far. Writing good agents’s “match the model to the job” principle is a cost argument as much as a quality one — a cheap default with model_fallbacks, an expensive model reserved for the one step (often a verifier) where it’s actually earning its keep.
  3. An accurate confidence threshold saves money, not just catches errors. A step that escalates early on genuinely low confidence avoids paying for however many expensive downstream steps would otherwise have run on a shaky foundation. Choosing confidence thresholds covers picking the number; this is the cost argument for why getting it right matters beyond correctness.
  4. Verifier mode is a cost/rigor trade-off. critic re-reads what the primary already produced; independent redoes the entire task from scratch, roughly doubling that step’s cost. Verifiers frames the choice around correlated vs uncorrelated errors — the same choice, priced.
  5. Grounding is cheap if the judge is. Writing your grounding judge’s model-choice section applies directly here — a grounding judge doesn’t need a frontier model, and shadow mode means you’re already paying for it on every run whether or not enforce: true is set.
  6. max_items on a fan-out is a direct cost cap. N branches means N LLM calls, concurrently — fan-out’s max_items guardrail (default 20) is as much a cost control as a safety one, worth setting deliberately rather than leaving at the default for a step whose list could plausibly be much longer.
  7. Team budgets tell you about a trend; max_usd stops one run. pricing.team_budgets is advisory, month-to-date, and never blocks anything — it’s for noticing a team’s spend is climbing before it’s a crisis. budget.max_usd is the emergency brake for one run. Neither substitutes for the other.
  8. Unpriced spend is invisible spend. A model with no pricing.models entry contributes NULL, not 0, to every rollup — cost control starts with the pricing table actually covering what’s really running, or opting into live/approximate OpenRouter pricing so an unpriced model at least shows up as an estimate instead of a gap.

Where next

Section titled “Where next”