Scoring is public
Detector logic, formulas, and ranking are public. No black box.
View detector source ->AI API relay authenticity check
Clodex checks whether a relay truly passes through Claude / GPT / Gemini or quietly swaps the request to a cheaper backend. Claude uses server-side thinking signatures; OpenAI and Gemini use protocol and usage-field checks.
100% open source · scoring is auditable line by line
Clodex temporarily uses your API key only for the check, so scoring, key handling, and security must be verifiable in code.
API keys never hit disk
Keys live in job memory only and are not written to report JSON, logs, or a database. Reports show only a masked key.
Security policy ->Fully self-hostable
Do not trust the public service? Run the code yourself so keys never leave your network.
Self-hosting guide ->
Claude
Cryptographic signal
Claude API relay check
Checks thinking signature, protocol shape, tool use, structured output, streaming consistency, usage, and long context.
Check Claude relay ->
OpenAI
Behavior/protocol level
OpenAI / GPT relay check
Looks for Chat Completions issues and fingerprints that GPT requests are being rewrapped from Claude or Gemini backends.
Check OpenAI relay ->
Gemini
Protocol level
Gemini relay check
Checks Gemini through the OpenAI-compatible path: function calling, JSON schema, streaming consistency, and usage.
Check Gemini relay ->Why choose Clodex
01
Claude thinking signature
The strongest Claude authenticity signal: an Anthropic server-side signature that a relay should not be able to forge.
02
Three protocols in one place
Claude, OpenAI, and Gemini are covered separately. If the key belongs to another protocol, the UI suggests the right page.
03
Preflight before submit
Before a full run, Clodex quickly checks available models so you do not wait for a zero-score report.
04
Shareable reports
Every report gets a permanent URL and a JPG card for chats or documentation.
05
Open-source trust
Keys are not retained, and the server code can be audited or self-hosted.
06
Field-level evidence
Reports show not just a score, but concrete fields, stream events, usage, and detector details.
How it works: 3 steps
- 1
Enter relay URL and API key
Choose Claude / OpenAI / Gemini, paste base_url and key. The model list is loaded automatically when the relay exposes /v1/models.
- 2
Start the check
Clodex runs detectors in parallel. Token cost is paid by your API key to the provider or relay.
- 3
Read the report
Check the score, critical issues, and each detector detail. For Claude, the key signal is thinking_signature.
100 points mean different things
Claude can use a cryptographic signal through thinking signature. OpenAI and Gemini do not expose the same server-side proof, so those checks focus on behavior, protocol, and usage.