Which providers can CostCompass track?

A growing list of the AI and compute services developers build on — model providers like Anthropic's Claude, OpenAI, and Google, plus voice, GPU, hosting, and analytics services. The full current list, each with a short connect walkthrough, lives on the providers page. Connect the ones you use and they fold into the same total.

How can one number cover providers that bill so differently?

CostCompass reads each provider's own usage data and turns it into a cost — pricing metered usage at the provider's published rates, or using the provider's own reported figure where it gives one. Tokens for Claude and OpenAI, characters for ElevenLabs, GPU-seconds for RunPod, requests for Cloudflare Workers and gigabytes stored for R2. Each provider meters differently, but everything arrives in your total already expressed in money, so the amounts line up and add together.

Does it forecast, or just show what I've spent?

Both. You get a running month-to-date total of what you've spent so far, and a forecast that projects what next month will cost at your burn rate to date. Provider dashboards give you the history; the forecast is the part they leave out.

Will I know if my spend suddenly jumps?

Yes, when you pull your running total. CostCompass doesn't send alerts, so the jump won't reach out to you. Each time you click Refresh it re-reads every connected provider, and a sudden jump shows up in your month-to-date figure and pushes the forecast up. You catch a spike forming while there are still days left in the month to act on it.

Where do my provider API keys live?

Stored on our servers as ciphertext, and only you hold the password that decrypts it. Your keys are encrypted in your browser with your vault password before they're saved, so what we hold is ciphertext. The password stays in your browser. When CostCompass fetches your usage, the key is decrypted in your browser and forwarded to the provider through a broker built not to log it, keeping the plaintext out of our database and logs. The security page documents exactly how.

Why might CostCompass's total differ from a provider's invoice?

Where CostCompass prices usage itself, it uses each provider's published rates at the moment usage is recorded, so that part of the figure is reproducible and doesn't shift when a provider later changes prices. Discounts that don't appear in the raw usage (batch-API rates, automatic prompt caching, promotional credits, committed-use deals) can make the actual invoice a little lower, and a few providers report their own cost or are read from a draining balance, which carries its own small approximation. Treat the CostCompass total as a close, slightly conservative running estimate.

Why use CostCompass instead of checking each provider's own dashboard?

Each provider's dashboard shows only its own slice, and only what you've already spent. So answering "what is AI costing me this month" means logging into a dozen consoles and adding it up by hand, after the fact. CostCompass turns that into one running, forward-looking number — a month-to-date total, a forecast, and a per-provider breakdown, down to the model wherever a provider exposes it, across everything you connect.

AI cost guides

Understand your AI bill, then get ahead of it

Written for developers paying several providers out of one budget, with no finance team behind them and no appetite for another spreadsheet. Start wherever your problem currently is.

See all your spend in one place — start free

By Joubert BergerUpdated August 20, 2026

An antique almanac-style engraving of a line chart whose curve climbs steeply and runs off the top of the frame, the runaway spike picked out in copper. — The spike the invoice only names after it's already spent.

A developer ships a feature on a Friday. Over the weekend a retry loop hammers a large prompt far more often than anyone meant it to, and a traffic spike starts routing calls through a pricier model. The code works fine. The invoice arrives at the end of the month and it's several times what they expected, and by then the money is already spent. No single screen had shown the line bending upward while it was still bending.

That's the trap with AI and compute spend. The damage is quiet and it adds up daily, and the one document that finally shows you the total is the invoice.

Do provider dashboards warn you about a rising bill?

Some of them do, within their own walls. Both OpenAI and Claude let you set a monthly spend limit and mail you as you approach it, and OpenAI’s will refuse requests once you reach it. But a limit is a threshold you picked yourself, not a projection of where the month is heading. Set it high and it stays quiet all month.

Most projects don’t run on a single provider anymore. A real one leans on Claude for one job and OpenAI for another, generates voices on ElevenLabs, rents a GPU on RunPod, and serves itself through Cloudflare. Each has its own login and its own unit of measure. One counts tokens, another characters, a third bills GPU-seconds, a fourth requests and gigabytes stored. There’s no single number anywhere for “what is AI costing me this month.”

Left as separate bills, that scattering costs you three concrete things, even when no single one looks alarming.

No total. Answering the most basic question, what am I spending on AI this month, means logging into every console and summing it by hand.
No forecast. Every dashboard shows the past. None project ahead, so you learn you overspent once the invoice lands.
No comparison. Without building a spreadsheet you can’t see which provider is your biggest line item, or that one tripled this month while another held flat.

What does it take to get AI spend under control?

Four things. Each is a subject in its own right, with its own set of guides.

Know what you’re actually paying for. On an AI coding bill the output is the small half. The money goes on the reading, and in a long session the same material gets read again, and charged again, on every turn. AI coding costs breaks down where it really goes.
Get it into one number. One view reads every provider’s usage, prices each provider’s own unit into money, and adds the lot into a single running total. Cost tracking tools walks the routes to that total, and how much work each route leaves you holding.
Look forward, not back. A total says where the month stands. A forecast says where it lands — that’s managing AI spend.
Pull the biggest lever. Caching a prompt that barely changes between calls is the cheapest saving on offer and the easiest to get subtly wrong. Prompt caching covers the difference.

Most people start in the middle. You can’t decide what to cut before you can see what you’re spending.

How does CostCompass track costs across every provider?

CostCompass is a spend-intelligence dashboard built around that job. It reads each provider’s own usage and billing data instead of routing your traffic or living in your code, which is one of three ways cost tools collect billing.

The CostCompass dashboard showing a single month-to-date total of $6,789.40 across all connected providers, with a forecast and a daily burn rate. — One running month-to-date total across every connected provider — with a forecast of what next month will cost and the daily burn rate.

You connect each provider once, and from then on:

Everything becomes a cost. Each provider’s usage is read from its own API and turned into money — priced at published per-unit rates, or taken from the provider’s own reported figure where it gives one — so all those different meters land as one comparable number.
One running month-to-date total. Across every provider you’ve connected, it’s the single figure none of the individual consoles can give you.
A forecast. It projects next month from your burn rate so far. The mechanics of that projection are covered in forecasting your AI spend.
A per-provider breakdown, and a per-model one wherever a provider exposes it, so when the total moves you can see which provider, and often which model, moved it.
Your keys are encrypted with your vault password before they reach us, so what we store is ciphertext only you hold the key to. The security model documents how.

A by-provider breakdown of month-to-date spend — OpenAI, Claude, RunPod, ElevenLabs, Cloudflare, and Google — each metered in its own unit (tokens, GPU-seconds, characters, requests) but combined into one running total. — Every provider in one breakdown — tokens, GPU-seconds, characters, and requests, all normalized to one comparable total, so your biggest line item is obvious at a glance.

By topic

Frequently asked questions

Which providers can CostCompass track?: A growing list of the AI and compute services developers build on — model providers like Anthropic's Claude, OpenAI, and Google, plus voice, GPU, hosting, and analytics services. The full current list, each with a short connect walkthrough, lives on the providers page. Connect the ones you use and they fold into the same total.
How can one number cover providers that bill so differently?: CostCompass reads each provider's own usage data and turns it into a cost — pricing metered usage at the provider's published rates, or using the provider's own reported figure where it gives one. Tokens for Claude and OpenAI, characters for ElevenLabs, GPU-seconds for RunPod, requests for Cloudflare Workers and gigabytes stored for R2. Each provider meters differently, but everything arrives in your total already expressed in money, so the amounts line up and add together.
Does it forecast, or just show what I've spent?: Both. You get a running month-to-date total of what you've spent so far, and a forecast that projects what next month will cost at your burn rate to date. Provider dashboards give you the history; the forecast is the part they leave out.
Will I know if my spend suddenly jumps?: Yes, when you pull your running total. CostCompass doesn't send alerts, so the jump won't reach out to you. Each time you click Refresh it re-reads every connected provider, and a sudden jump shows up in your month-to-date figure and pushes the forecast up. You catch a spike forming while there are still days left in the month to act on it.
Where do my provider API keys live?: Stored on our servers as ciphertext, and only you hold the password that decrypts it. Your keys are encrypted in your browser with your vault password before they're saved, so what we hold is ciphertext. The password stays in your browser. When CostCompass fetches your usage, the key is decrypted in your browser and forwarded to the provider through a broker built not to log it, keeping the plaintext out of our database and logs. The security page documents exactly how.
Why might CostCompass's total differ from a provider's invoice?: Where CostCompass prices usage itself, it uses each provider's published rates at the moment usage is recorded, so that part of the figure is reproducible and doesn't shift when a provider later changes prices. Discounts that don't appear in the raw usage (batch-API rates, automatic prompt caching, promotional credits, committed-use deals) can make the actual invoice a little lower, and a few providers report their own cost or are read from a draining balance, which carries its own small approximation. Treat the CostCompass total as a close, slightly conservative running estimate.
Why use CostCompass instead of checking each provider's own dashboard?: Each provider's dashboard shows only its own slice, and only what you've already spent. So answering "what is AI costing me this month" means logging into a dozen consoles and adding it up by hand, after the fact. CostCompass turns that into one running, forward-looking number — a month-to-date total, a forecast, and a per-provider breakdown, down to the model wherever a provider exposes it, across everything you connect.

About the author

Joubert Berger builds CostCompass, a spend-intelligence dashboard that pulls usage from AI and compute providers into one month-to-date total, a forecast, and a per-provider breakdown. This guide reflects how CostCompass reads each provider's own usage API — see thesecurity model for how your keys are handled.