model-substitution

Local security audit for AI API relays and LLM proxies: detects prompt injection, model substitution, tool-call rewriting, SSE anomalies, error leakage, and Web3 wallet risks. Website: https://apirelayaudit.com

Verify which LLM an OpenAI-compatible API really serves — single-token behavioral fingerprinting (Jensen-Shannon) from the paper "One Token Is Enough" (arXiv:2607.10252). Zero-dependency TypeScript library + CLI. Catches model substitution by resellers & gateways.

Which model is really behind your API relay or agent IDE? Behavioral fingerprinting + anytime-valid sequential tests (FPR<=1%). LLMs can't be random - measured on 9 frontier models.

Publicly verifiable proof that an LLM endpoint actually spent the compute you paid for — zero provider cooperation. Closes the effort gap (Hollow-LLM) and the transferability gap (IRIS).

Is the model behind that API endpoint the one it claims to be? Request-by-request checks against the genuine model.

Black-box fingerprinting of LLM endpoints: serving-stack DNA, tokenizer vectors, capability matrix, behaviour distributions. Zero dependencies.

CLI framework for auditing LLM providers and detecting model substitution.

Check an AI service before you pay: claims, the model underneath, your results, the markup and the terms. An agent skill

Detect what model is really behind any OpenAI-compatible proxy, reseller or gateway. 6-way behavioral forensics - tokenizer fingerprint, trap-question bank, speed side channel, tool-call integrity - with no self-declaration needed. Zero-dependency single binary (Win/Linux/macOS), exit codes built for CI.

Is your LLM provider serving the model you're paying for? A detector with a MEASURED FPR