Provider and model
Each event identifies the model that handled the call, so cost changes do not disappear inside one blended total.
AI cost attribution
Find the feature, model, and customer behind every reported AI cost.
Add the server SDK around the model calls your product already makes. It records provider-reported usage and the dimensions you choose, then prices each event against a versioned server catalog.
What is recorded
Enough to calculate cost. Not enough to read the conversation.
Input, output, and cached tokens
Cached usage stays separate because providers may price it differently from uncached input.
Status, latency, and retries
Failed or retried application calls remain visible when the provider returns usage, so reliability cost is not hidden.
Your own dimensions
Send opaque account, user, feature, and version identifiers that match the way your product is operated.
How it works
One event becomes a traceable cost.
- Observe the model response. The backend wrapper reads the usage returned by the provider.
- Attach product context. Your server supplies opaque account, user, feature, and version IDs.
- Price it centrally. The Grow or Die server applies the price version effective for that event.
- Keep uncertainty visible. Unknown models or missing usage stay unpriced rather than receiving a guessed cost.
Private by design
Prompts, responses, and model API keys are not sent.
Grow or Die needs usage counts and your opaque business dimensions, not conversation content. Cost is calculated on the server so price changes do not require every customer to upgrade an SDK.
Set up my product