Inferock Bench is a local diagnostic proxy designed to audit AI API usage, providing independent verification of billed costs against actual delivery performance.
What the product does and how it is positioned
Inferock Bench serves as a measurement instrument for developers to cross-check AI provider bills against real-time delivery evidence. By routing traffic through a local proxy, it captures per-call data to identify discrepancies such as billed failures, token mismatches, and retries.
The tool emphasizes transparency by separating observed provider spend from bill-bounded money loss and invoice-check exposure. It is designed to help users inspect the reliability and cost-efficiency of their AI integrations without relying solely on provider-supplied totals.
Source-supported ways to use the product
Users can verify if billed API calls correspond to successful model outputs, helping to identify and document potential overcharges or billing errors.
Developers can measure token usage and latency for specific AI models locally, allowing for more accurate cost estimation and performance monitoring.
The tool functions by intercepting API requests at the local level, recording metadata such as status codes, timing, and provider-reported token counts. It then applies the Inferock Standard to interpret these observations into financial and operational metrics.
This approach allows for the identification of specific failure patterns, such as empty replies or truncated outputs, which are often billed as standard completions. By maintaining a local ledger of these events, users can build a defensible record for billing inquiries.
Checks to run with your own material and workflow
What was checked and when
Answers based on the source-checked product record
No, the tool is designed as a local process. Provider keys are stored locally on your machine with owner-only permissions and are only attached to requests sent directly to the AI provider.
No, it does not perform automated billing disputes. It provides the evidence and documentation necessary for you to inspect your bills and conduct your own billing-integrity inquiries.
The tool is extensible-by-design, but only explicitly supported providers have the necessary adapters and pricing taxonomies to generate measured receipts. Other providers will not produce measured results.
No, the receipt is an interpretation of observed events based on the Inferock Standard. It is intended to highlight potential discrepancies for further review rather than serving as a definitive proof of provider error.
Cache discounts are labeled as invoice-check exposure. They are kept separate from bill-bounded money loss and are accompanied by guidance to verify the details against your actual provider invoice.
