Question 1

What is FrugalAI?

Accepted Answer

FrugalAI is a FinOps-first, OpenAI-compatible LLM gateway. It meters each request, attributes spend to a team, evaluates eligible routes, applies cache and budget policy, and records the decision in a request ledger.

Question 2

How does rollout work?

Accepted Answer

Workloads move through three modes: monitor, assist, and enforce. Monitor leaves requests unchanged while building evidence. Assist can route auto or unpinned requests. Enforce applies the approved policy. Quality evaluation is a continuous guardrail inside assist and enforce, not a separate stage: sampled checks compare eligible alternatives, and a configured SLO breach rolls the workload back to monitor.

Question 3

Do we need to rewrite our application?

Accepted Answer

The gateway exposes OpenAI Chat Completions and Anthropic Messages endpoints, so the primary change is usually the base URL and credential. Provider-specific features and streaming behavior should still be tested against your workload before routing becomes active.

Question 4

Which providers and models are in the catalog?

Accepted Answer

The current catalog includes 31 chat models across OpenAI, Anthropic, Google, DeepSeek, Mistral, Groq, Together AI, Fireworks AI, and xAI. A production route only considers providers configured with your organization's keys.

Question 5

Does FrugalAI only work with OpenAI models?

Accepted Answer

No. "OpenAI-compatible" describes one supported request format, not the model list. OpenAI Chat Completions and Anthropic Messages clients can reach eligible catalog models from configured providers, including Anthropic and Google, with tool, image, and streaming translation on those native egress paths. Current boundaries, stated plainly: Gemini-native and OpenAI Responses ingress are planned, and Anthropic server-managed tools return an explicit unsupported response.

Question 6

How does FrugalAI protect output quality?

Accepted Answer

Routes are filtered by model capability, workload requirements, policy, and available evaluation evidence. Sampled evaluations and configured SLO checks can roll a workload back to monitor mode. FrugalAI does not claim that a cheaper model is always equivalent.

Question 7

How does caching work?

Accepted Answer

FrugalAI supports exact caching and optional semantic matching. Cache entries are scoped by tenant and policy, and sensitive or dynamic workloads can bypass caching. A cache hit avoids a provider call and is recorded in the ledger.

Question 8

Are the savings shown guaranteed?

Accepted Answer

No. Monitor-mode savings are counterfactual estimates based on an eligible route that was not taken. Savings become realized only when an actual routed request or cache hit costs less than its recorded baseline.

Question 9

Who pays the model providers?

Accepted Answer

Your organization keeps and pays its provider accounts directly. FrugalAI pricing covers the gateway control plane, request ledger, policies, caching, budgets, and support. Provider token charges are separate.

Question 10

What is included in Individual?

Accepted Answer

Individual is a planned $149 per month tier for one seat, one app, and up to 1 million requests per month. It will include routing, fallback policy, exact and optional semantic caching, a budget guard, and the request ledger. It is not purchasable yet; during the beta, access is arranged directly.

Question 11

Is public self-serve checkout live?

Accepted Answer

No. FrugalAI is in an owner-led production beta, so access and billing are arranged directly. The site does not claim that a public checkout, customer portal, or instant entitlement flow is live.

Question 12

How does dashboard login work?

Accepted Answer

Approved owner accounts receive a secure email sign-in link. The link can be received in a Gmail or Google Workspace inbox. One-click Google OAuth is not currently enabled.

Frequently asked questions