Model inspection lab

Inspect model quality, endpoint provenance, and real model identity

This page runs a fast black-box inspection against public AI endpoints. It scores multiple dimensions, captures response fingerprints, estimates the likely underlying model family, and tells you whether the endpoint looks first-party, relayed, or locally hosted.

Model inspection lab

Start a model inspection

Paste the API base URL, API key, and model alias. The evaluator will probe protocol behavior, formatting discipline, reasoning control, code execution, and calibration signals.

Independent judge model

Optional. Add a separate trusted model to review traces, headers, route hypotheses, and spoofing signals for a second-pass forensic opinion.

Use a judge endpoint that is independent from the target API. This improves reliability when you suspect wrappers, relays, or fake model aliases.

Public endpoints only

For safety, this public site does not probe localhost, RFC1918/private networks, or internal hostnames. Those targets are classified as probable local runtimes without sending live traffic.

Sensitive input handling

The key is used only for the live inspection request and is not persisted to the page state beyond the current browser session. Use a low-scope test key whenever possible.

Quick presets

Inspect model quality, endpoint provenance, and real model identity

This page runs a fast black-box inspection against public AI endpoints. It scores multiple dimensions, captures response fingerprints, estimates the likely underlying model family, and tells you whether the endpoint looks first-party, relayed, or locally hosted.

High-dimension black-box test

The lab scores not just a single completion, but protocol fit, structured output control, reasoning behavior, code execution fidelity, and self-calibration.

Model-family probability mining

Alias patterns, observed response metadata, and behavioral signals are combined to estimate which real model family is most likely behind the endpoint.

Endpoint authenticity verdict

The report separates model identity from API provenance so you can see whether the endpoint behaves like a first-party vendor API, a relay, or a local stack.