About 黑曜

One endpoint, many models.

The shape of it

One endpoint, one key, and a catalogue the mainstream APIs will not serve. Point an OpenAI or Anthropic client at it, send a model id from the catalogue, and it answers, meters the call, and bills your key.

Three things make it different from the mainstream APIs:

  • Uncensored models under one key.
  • Per-key spend control enforced before the call: holds and daily quotas bound the spend, not a reconciliation afterwards.
  • No prompt retention. The log records model, tokens, cost and status — never the text; every response even carries a signed receipt you can verify.

Who it's for

Red teams, detection engineers, trust-and-safety teams building abuse classifiers, and researchers generating adversarial data. The common thread is work that a frontier API refuses on sight even when the caller is authorised to do it. Capabilities, context and unit prices are listed in the model catalogue.

Start here

Follow the quickstart to send your first request, copy a client config directly, or go straight to running Claude Code or Codex CLI against it. You can also try it in Chat first.