cost-cutter

Enforce a hard cap and a drift-check when a cheap executor model consults an expensive advisor model, and compute the effective cost from actual call counts instead of a benchmark's assumed rate. Use when the user adopts the advisor or orchestrator pattern, pairs a cheap model with an expensive reviewer, or quotes a benchmark discount like "63% of the price".

Before swapping any workload to a cheaper model, declare the cases the premium model still dominates and prove they keep routing to premium. Use when the user wants to "just switch" to a cheaper model (text, image, or video), or asks whether a cheap model is good enough to replace an expensive one.

Cut agent token spend by shrinking what enters the context window. Index the repo or corpus once and query it instead of re-reading files on every question. Use when the user complains their coding agent burns tokens, the context fills up fast, the same files get read repeatedly, or the bill scales with conversation length.

Route high-volume, low-stakes triage (reading piles, inbox summaries, needs-reply flags) to a free model with a strict output schema. Use when the user wants one-line summaries of many items cheaply, asks to triage email, articles, or reports with AI, or wants to decide what's worth reading without paying premium rates for it.

Size a big one-time batch job against a free tier's rate limit and token budget BEFORE starting it, with a proven wall-clock ETA. Use when the user wants to label a dataset, summarize an archive, or process a large backlog for free (or on a tiny rate limit), or asks "will this finish overnight?"

Choose a model with evidence by running the few prompts that actually matter across candidate models inside a free tier's caps, picking by a stated criterion. Use when the user asks "which model should I use", wants to evaluate a newly launched model, or is choosing between models for a specific task.

Stop paying for deep reasoning on easy turns by setting a modest default reasoning effort and escalating per task. Use when the user runs a reasoning model in an agent and the bill is dominated by output or thinking tokens, or asks about reasoning_effort, thinking budgets, or why a cheap model is still expensive to run.

Cut LLM spend by routing bulk work to a cheap model and escalating only the hard turns to a premium one. Use when the user says their AI bill is too high, asks to "use a cheaper model", or wants two-tier model routing without losing quality on the hard tasks.

Pin an open-weights fallback model with a tested-on date and real smoke prompts, so a pulled or deprecated model is a two-minute config swap instead of a lost week. Use when the user worries a model could vanish or be deprecated, builds anything important on one model, or asks about model failover and resilience.

Attribute AI usage by tokens AND dollars so a high-volume cheap model is never mistaken for the expensive one. Use when the user asks where their AI spend actually goes, why the bill is high, which model is costing the most, or wants a usage audit across agents and models.