# FrugalAI

> FrugalAI is a FinOps-first, OpenAI-compatible LLM gateway. It applies cost-aware model routing, scoped response caching, request-level metering, team budgets, and decision receipts. Monitor mode measures potential changes before enforcement.

## Primary pages
- [Product](https://frugalai.io/): Product definition, rollout model, pricing, and FAQs.
- [Research](https://frugalai.io/blog): Evidence-grounded LLM gateway and AI FinOps guides.
- [Pricing](https://frugalai.io/pricing.md): Machine-readable plan summary.
- [RSS](https://frugalai.io/feed.xml): Research feed.
- [Full corpus](https://frugalai.io/llms-full.txt): Full article text with sources.

## Research articles
- [What Is an LLM Gateway? Architecture, Controls, and Tradeoffs](https://frugalai.io/blog/what-is-an-llm-gateway): A practical explanation of the proxy layer between applications and model providers, including routing, caching, policy, and observability.
- [LLM Cost-Aware Routing: A Practical Implementation Guide](https://frugalai.io/blog/llm-cost-aware-routing-guide): How to route routine requests to lower-cost models without silently sacrificing output quality.
- [FrugalGPT Explained: What LLM Cascades Teach Cost Teams](https://frugalai.io/blog/frugalgpt-cascades-explained): A plain-language guide to the FrugalGPT research and how to translate cascades into production controls.
- [RouteLLM and the Quality-Cost Frontier for Model Routing](https://frugalai.io/blog/routellm-quality-cost-routing): What RouteLLM measures, what its results mean, and how to avoid copying benchmark claims into production forecasts.
- [Semantic Caching for LLM Apps: Savings Without Cross-Tenant Risk](https://frugalai.io/blog/semantic-caching-for-llm-apps): How semantic response caches work, when they save money, and the controls needed to prevent unsafe reuse.
- [Prompt Caching vs. Semantic Caching: What Actually Gets Reused](https://frugalai.io/blog/prompt-caching-vs-semantic-caching): A side-by-side guide to provider prefix caches and gateway response caches.
- [AI FinOps: An Operating Model for Variable Inference Spend](https://frugalai.io/blog/ai-finops-operating-model): How FinOps practices change when model selection, tokens, cache policy, and quality affect the bill in real time.
- [How to Allocate LLM Cost by Team, Product, and Feature](https://frugalai.io/blog/allocate-llm-cost-by-team): A durable tagging and ledger model for chargeback, showback, and AI unit economics.
- [LLM Budget Guardrails That Do Not Break Your Product](https://frugalai.io/blog/llm-budget-guardrails): Design progressive AI spend controls that preserve essential workflows while slowing noncritical usage.
- [Shadow Mode for LLM Routing: Prove Savings Before Enforcement](https://frugalai.io/blog/shadow-mode-llm-routing): A rollout method that compares alternative model decisions without changing production answers.
- [An Evaluation Framework for LLM Routing Quality](https://frugalai.io/blog/llm-routing-evaluation-framework): How to combine deterministic checks, golden sets, judges, human review, and rollback thresholds.
- [Migrating to an OpenAI-Compatible LLM Gateway](https://frugalai.io/blog/openai-compatible-llm-gateway-migration): A phased migration checklist for base URLs, keys, streaming, tools, retries, and rollback.
- [LLM Gateway vs. Observability Platform: Where Each Control Belongs](https://frugalai.io/blog/llm-gateway-vs-observability-platform): A clear boundary between request-path enforcement and post-request analysis.
- [Build vs. Buy an LLM Gateway: A Decision Framework](https://frugalai.io/blog/build-vs-buy-llm-gateway): How to compare internal gateway work with managed products using control, risk, and total operating cost.
- [LiteLLM Alternative for Teams That Need AI FinOps Controls](https://frugalai.io/blog/litellm-alternative-finops): When an OpenAI-compatible proxy is enough and when budget owners need deeper allocation, policy, and reconciliation.
- [Portkey Alternative for Budget-Owner AI Operations](https://frugalai.io/blog/portkey-alternative-ai-finops): A neutral framework for comparing an AI control plane with a FinOps-first request gateway.
- [Cloudflare AI Gateway Alternative for Cost-Aware Model Routing](https://frugalai.io/blog/cloudflare-ai-gateway-alternative): Compare edge proxy capabilities with a control plane built around model economics and team budgets.
- [Helicone Alternative for AI Cost Control and Budget Enforcement](https://frugalai.io/blog/helicone-alternative-finops-gateway): How to compare LLM observability with a gateway centered on financial policy and routing.
- [LLM Model Pricing Drift: Why Your Cost Catalog Needs Verification](https://frugalai.io/blog/llm-model-pricing-drift): A workflow for tracking provider price changes, cached-token rates, and their impact on real traffic.
- [From Cost per Token to Cost per Outcome in AI Products](https://frugalai.io/blog/llm-cost-per-outcome): A measurement model for connecting inference spend to resolved tickets, accepted drafts, and completed workflows.
- [LLM Provider Failover Without Hidden Cost or Retry Storms](https://frugalai.io/blog/llm-provider-failover-cost): Design provider resilience with bounded retries, equivalent capabilities, and honest effective-cost accounting.
- [Secure LLM Gateway Checklist: Keys, Tenants, Logs, and Policies](https://frugalai.io/blog/secure-llm-gateway-checklist): A security baseline for the request layer that sees model traffic across applications and providers.
- [How to Run an AI Spend Month Close](https://frugalai.io/blog/ai-spend-month-close): A repeatable close process for provider invoices, request ledgers, budgets, savings, and unattributed usage.
- [AI Gateway RFP Checklist: 40 Questions for Vendors](https://frugalai.io/blog/ai-gateway-rfp-checklist): A procurement checklist covering compatibility, routing, security, FinOps, evaluation, reliability, and exit.
- [How to Reduce LLM API Cost Without a Blind Model Downgrade](https://frugalai.io/blog/reduce-llm-api-cost): A prioritized savings plan across measurement, prompt design, caching, routing, batch work, and budgets.

## Important boundaries
- Published research savings are cited as research results, not guaranteed customer outcomes.
- FrugalAI starts in monitor mode and does not change production answers until a route is approved.
- Protected owner dashboard routes are not part of the public corpus.