ai-cost-finops

What problem does it solve?

This Skill helps you reduce and forecast LLM costs by turning pricing, token economics, and provider features (prompt caching, Batch APIs, and model routing) into actionable optimization steps.

Core Features & Use Cases

• Cost projection with real price inputs: Uses provider-specific input/output token rates to estimate spend before you ship.
• Prompt caching optimization: Applies Anthropic and OpenAI caching mechanics (discount rates, TTL behavior, what to cache, and break-even reasoning).
• Batching and async execution strategy: Designs where Batch APIs fit to capture standardized discounts for non-real-time workloads.
• Model routing playbooks: Guides when to default to smaller models and when to escalate to larger/frontier models using audit-friendly rules.
• Token and output-length economics: Focuses on output discipline and structured outputs to control the dominant output token cost driver.
• RAG cost stack budgeting: Estimates spending across embeddings, reranking, and vector DB operations to find the biggest cost drivers in retrieval pipelines.
• FinOps observability and attribution: Recommends instrumentation patterns (Helicone, Langfuse, Phoenix, LangSmith, OpenLLMetry, Vantage) so you can measure per-feature waste and allocate spend accurately.

Quick Start

Ask the AI: "Project our 2026 monthly LLM cost for the 'customer support triage' feature, using your May 2026 pricing assumptions, then list the top 3 levers to reduce spend with expected impact."