Question 1

In one sentence, what does FrugalAI do?

Accepted Answer

It sits between the apps your company builds and the AI companies you pay, sends each request to the cheapest model that can handle it, and gives you an itemized record of what everything cost and who spent it.

Question 2

My boss told me to cut our AI costs. Where do I start?

Accepted Answer

Start by finding out where the money actually goes, because most companies do not know. Point one app at FrugalAI and leave it in watch mode for a month. Nothing about that app changes. At the end you have a real breakdown of your spend by team and a number for what you would have saved on cheaper models. That number is what you take into the conversation about what to change.

Question 3

Will this make our AI worse?

Accepted Answer

That is the right thing to worry about, and it is why FrugalAI does not swap anything until you tell it to. A cheaper model is not always as good, and we will not pretend otherwise. What we do is check whether a cheaper model handles your specific work before it is allowed to take that work, keep spot-checking afterwards, and pull back automatically if quality drops below the line you set.

Question 4

How much work is this for my engineering team?

Accepted Answer

Small. Your apps already send AI requests to an address, and they would send them to us instead. That is usually a one-line change plus a new key. Your team should still test their own app before letting anything get swapped, which is normal for any change to a live product.

Question 5

Do we have to switch AI vendors?

Accepted Answer

No. You keep your existing accounts with OpenAI, Anthropic, Google and the rest, and you keep paying them directly. FrugalAI only uses the accounts you connect. What you pay us covers the software that does the routing, the record keeping, the budgets, and support.

Question 6

Which AI companies and models does it cover?

Accepted Answer

Thirty-one models across nine companies: OpenAI, Anthropic, Google, DeepSeek, Mistral, Groq, Together AI, Fireworks AI, and xAI. Your requests only ever go to the ones you have connected an account for.

Question 7

Our engineers will ask what is not supported yet. What do we tell them?

Accepted Answer

Apps written against OpenAI Chat Completions or Anthropic Messages can connect today, including tool use, images, and streaming. Not yet supported: connecting through Google Gemini's own format, the newer OpenAI Responses format, Amazon Bedrock, Google Vertex, and custom endpoints. Anthropic's server-managed tools return a clear unsupported message rather than failing quietly. We would rather you find that out on this page than three weeks into a build.

Question 8

Are the savings you show guaranteed?

Accepted Answer

No, and we keep the two kinds of savings visibly separate. While FrugalAI is only watching, the savings figure is an estimate of what a cheaper model would have cost. It is a forecast, not money in your account. Once you let it start swapping models, savings become real and we report them separately.

Question 9

How does paying once for repeat answers work?

Accepted Answer

If your customers ask the same question a hundred times, you normally pay for a hundred answers. FrugalAI can store the answer and reuse it, so you pay once. Stored answers are walled off to your company only, and anything sensitive or one-of-a-kind skips this entirely and goes to a live model.

Question 10

Can I stop a team from overspending?

Accepted Answer

Yes. Give each team a monthly budget and pick what happens as they approach it: send you a warning, quietly move that team to cheaper models, or stop the spend. The point is that you decide in advance, so nothing important breaks without warning.

Question 11

What does it cost?

Accepted Answer

The founding-partner beta is $5,000 for a managed 90-day pilot on one meaningful workload. If you continue after the pilot, your platform rate is locked at $1,500 per month for 12 months. Enterprise agreements start at $60,000 per year. What you pay model providers remains separate and goes directly to them.

Question 12

What exactly is guaranteed?

Accepted Answer

For a qualified pilot, we agree on the workload, baseline, and quality bar in writing. If the first 30 days of shadow measurement do not identify at least $15,000 in annualized savings at that quality bar, we refund the $5,000 pilot fee. That is a guarantee on the measured opportunity, not a promise that every projected dollar will be realized after rollout.

Question 13

Can I sign up right now?

Accepted Answer

The product is in a private beta limited to ten founding partners. Apply with one high-volume workload and a current monthly AI bill; if the fit is right, we send the pilot order form and secure Stripe payment link, then schedule setup.

Question 14

How do I log in once I have access?

Accepted Answer

We email you a sign-in link instead of asking you to keep another password. Click it and you are in. Signing in with a Google account is not turned on yet.

Questions you are probably about to ask