AI usage and estimated cost
AI usage records associate model tokens with features and conversations and estimate token cost where pricing is available.
What it is
INSIDE records token usage for supported AI operations and associates it with the feature that performed the work. Where model pricing is available, it also calculates an estimated cost in US dollars. These records help explain AI activity and resource use.
Who uses it and why
Administrators and supervisors use usage evidence to investigate which AI activities consume resources. Developers and Powerfront use the associated feature and conversation details to investigate individual requests.
Where to find it
Powerfront manages the related model and account configuration in Settings › AI › Models and Settings › AI › Accounts. Ask Powerfront for a usage investigation when you need token or estimated cost details for an operation.
How it works
For a supported generation, INSIDE records input tokens, output tokens, total tokens and generation duration. It associates the record with the AI account, model, feature and website, plus available conversation, visitor, operator, channel and subsite details.
An AI task can make several model calls. Each logged generation contributes its own usage, so a single reply can involve more than one usage record. Where available, specialist work also identifies the originating and executing features.
Estimated cost applies the model's input and output rates. Cached input and reasoning tokens use separate rates when those rates exist; otherwise they use the ordinary input or output rate. Those detailed tokens are removed from the ordinary quantities before calculation to avoid counting them twice.
If INSIDE cannot find the model or suitable pricing, it cannot produce that estimate. A missing estimate does not mean the operation was free.
Configure it
There is no separate switch for recording usage. Ask Powerfront to check that the account uses the intended model and that the available model pricing supports the estimate you need.
When requesting an investigation, give Powerfront the approximate time, AI feature and conversation, where applicable. Compare equivalent activities and model choices when reviewing changes in usage.
Limits and good to know
- Estimated cost is in US dollars and is not a reconciled provider invoice.
- Model token estimates do not establish a complete bill for voice, embeddings, guardrails or external agent services.
- Logging failures can leave gaps without failing the conversation. Recorded totals should not be treated as guaranteed complete billing evidence.
- Suggestion use, editing, ignoring and translation adoption are separate activity measures. They do not equal token counts.
- Recording usage does not enforce a monetary spending cap.
Depends on and works with
View all connections as a list
This feature uses
- Specialist sub-agents for AI tasksEnhances
Specialist invocations can retain originating and executing feature context with their usage.
- AI assistance activityRelated
Compare token usage with AI assistance activity when reviewing resource consumption and operator adoption.
- AI accounts and shared defaultsConfigured by
The AI account supplies the model and account identity attributed to usage.
- AI model catalogueConfigured by
Model catalogue pricing determines whether token cost can be estimated and which rates apply.
Used by
- AI assistance activityEnhances
Generation usage records add token counts and cost estimates to AI activity analysis.
FAQ and troubleshooting
Why is there no estimated cost for an operation?
The model or its pricing may be unavailable, or the execution path may not record an estimate. Ask Powerfront to check the account, model and usage record.
Why are there several records for one answer?
The AI may call the model repeatedly while using tools or completing its work. A visible reply is not necessarily one generation.
Why does the estimate differ from my provider invoice?
The estimate uses available model token rates. It does not establish complete coverage of all services, provider charges or billing adjustments.
Can I set a spending limit through usage metering?
This metering does not provide a monetary quota. Execution limits can bound particular AI work, but they are not a billing cap.