An OpenAI-compatible API for work the mainstream hosts block
Security work produces prompts that look like attacks because they are, in a lab, on a contract, against a system you are allowed to touch. You need an API that knows the difference between that and a random caller asking for a worm.
Why the mainstream APIs fail here
Catalogues that list uncensored models often still filter them, put no cap on a key, or log the prompt "for safety". The combination you actually need is: the model answers, the cap holds, the prompt is not kept.
How MagmaRouter is used for this
MagmaRouter is that combination. Abliterated and uncensored models in the catalogue, frontier models on the same key, a hold-then-settle cap that cannot be outrun by a stream, and a signed receipt of a hash instead of a prompt log. The Playground is the same path as the API.
A working loop
- Get a key.
- Point the SDK at /v1.
- Put a cap on it before you give it to a script.
Prompts are not stored. A signed receipt of the hash comes back on every call, so you can prove later what MagmaRouter logged. Details are on the privacy page. Per-model prices are on /models and the pricing model is on /pricing.