COGS monitoring for AI products
Know what your AI app actually costs to run.
One dashboard for all your AI spend — by app, not by invoice. Paste one admin key per provider; Costbase discovers every API key and pulls your usage straight from each provider's official reporting APIs. No proxy. No SDK. No code changes.
5 providers · 1 dashboard · ~2 minutes to connect
Pulls straight from the official reporting APIs of
OpenAI
Anthropic
xAI
OpenRouter
ElevenLabs
Gemini — soon
Groq — soon01
Connect
Click a provider, paste one admin key. Verified live in seconds.
02
Discover
Every API key in the account appears automatically, with a full year of history backfilled.
03
Name
Rename each key to the app it powers — “Transcription App,” “Support Bot.” One click, inline.
04
See
One chart, all providers. Cost per day, grouped by app, provider, model, or key.
Pull, don't proxy.
Most cost tools are gateways: you reroute production traffic through their servers to get a chart. That means a new dependency in your critical path, added latency, and your prompts transiting someone else's infrastructure.
Costbase reads your usage from each provider's official admin reporting APIs — the same data behind their own billing pages. It can be down, slow, or deleted and your apps never notice.
Cost visibility shouldn't require re-architecting your stack.
Proxy tools
Every production request flows through a third party.
Costbase
Your request path is untouched. Usage is read out-of-band.
Five providers, one view
OpenAI, Anthropic, xAI, OpenRouter, and ElevenLabs, merged into a single series. Several accounts on one provider? Add each as its own connection.
The usage others bury
Whisper transcription, text-to-speech, embeddings, and images — not just chat completions. Audio spend finally shows up next to everything else.
Fresh on its own
Hourly background sync against each provider's reporting APIs, plus a Sync-now button when you can't wait.
Honest numbers
Provider-reported dollars wherever the provider gives them; computed from tokens × a maintained pricing table where they don't — and every figure is marked as one or the other.
Read-only by design
Admin keys are used for reporting endpoints only — never to create, rename, or delete anything in your accounts. They're encrypted at rest, shown only as their last 4 characters after saving, and revocable in your provider console at any time.
FAQ
How fresh is the data?
Providers publish usage with a lag of minutes to a few hours. Costbase re-pulls the last 3 days every hour, so figures converge on exactly what your invoice will say. This is a cost-visibility tool, not a request debugger — by design.
Do you see my prompts or production keys?
No. Costbase reads aggregate usage from each provider's official admin reporting APIs — the same data behind their billing pages. Your prompts, responses, and production API keys never touch it.
How are per-app dollar figures calculated?
Where a provider reports dollars at key level (OpenAI's Costs API), we use theirs. Where it only reports tokens per key (Anthropic), we compute cost as usage × current published prices and mark it ≈. xAI reports spend per team, so it appears account-level.
What do I paste, exactly?
One admin/management key per provider account — created in each provider's console in about a minute. It's used only for read-only reporting endpoints, encrypted at rest, and never displayed again after saving.
You can't cut what you can't see.
Connect a provider and watch a year of spend resolve into apps you recognize.
Get started — free