About Inferock Bench
Inferock Bench is a local proxy that sits between your application and OpenAI, Anthropic, Gemini, or OpenRouter shaped API calls. It captures per-call token usage, failures, and retries, then generates an independent receipt showing what you were billed and how much you may be overpaying. The tool launched this week and is open source under the FSL-1.1-Apache-2.0 license, converting to Apache-2.0 in two years.
Review
Inferock Bench addresses a specific gap in LLM operations: providers give you totals, not per-call evidence. The tool runs locally, so your provider API key never touches their servers. You point your existing SDK at it by changing two settings, apiKey and baseURL, and it starts recording every call.
Key Features
- Per-call receipts that capture spend observed, bill-bounded money loss, time loss, and a separate "invoice-check exposure" line that is never summed into money loss
- Detection of answers cut off mid-stream that still got billed, empty replies with billed tokens attached, and token counts that don't match visible output
- Retry tracking that surfaces silent duplicate charges
- Cache discount visibility to identify credits you may be missing on your invoice
- Local proxy architecture supporting OpenAI, Anthropic, Gemini, and OpenRouter shaped calls
Pricing and Value
The tool is listed as free on Product Hunt. It is open source under FSL-1.1-Apache-2.0, converting to Apache-2.0 in two years. Setup is via a single command, npx inferock-bench, with the makers stating it takes about a minute to run.
Pros
- Runs entirely locally, so provider keys stay on your machine and aren't sent to third-party servers
- Works across multiple providers without provider-specific integration work
- Separates the "invoice-check exposure" line from money loss calculations, which keeps the headline number conservative
- Public changelog and issue tracking on GitHub, which the makers chose specifically because the project's premise is not trusting vendor-reported data
- Quick setup requiring only a baseURL and API key change
Cons
- Forward-looking only; the makers confirmed it cannot retroactively audit calls that happened before installation, since it only records traffic it actually sees
- Out of scope for detecting well-formed 200 responses with silently corrupted values inside; a maker confirmed a call that returns 200 with a wrong value passes as clean because the proxy can't tell the value was wrong
- Not well suited for web-based logins like claude.ai or other subscription interfaces, since those don't expose per-call API detail and the tool requires an API key and baseURL to function
Inferock Bench is most useful for developers running multi-model agent setups who need to explain billing jumps with per-call evidence rather than estimates. Teams disputing AI provider charges will find the receipts useful for walking into those conversations with data. The tool is still new, and the makers have noted that cost-per-accepted-outcome roll-ups are a suggested future revision, not something available today.
Open 'Inferock Bench' Website
Your membership also unlocks:
