One key for every model: why we built Nemu Gate
SDK sprawl, key rotation, and per-provider quirks were eating our days. So we built a drop-in gateway that speaks OpenAI and Anthropic natively, and never touches your billing.
The problem with SDK sprawl
Every provider ships its own SDK, its own auth scheme, its own retry semantics and its own idea of what a streaming chunk looks like. A team using three providers ends up maintaining three integration layers that all do the same job slightly differently, and every model launch means another round of glue code.
The keys are worse than the SDKs. They live in three dashboards, rotate on three schedules, and leak in three different ways. Nobody owns the map of which key is used where.
One endpoint, two dialects
Gate answers in both of the wire formats that matter, OpenAI's and Anthropic's, from a single address. Your existing code keeps its SDK and swaps one base URL. The gateway translates between dialects on the fly, so an Anthropic SDK can call a model served over the OpenAI format and never know the difference.
Model names become your names. You map an alias like "smart" to a provider model once, and every app, script and teammate picks it up instantly. Swapping the model behind an alias is a console edit, not a deploy.
Your keys, never our billing
The part we refuse to compromise on: your provider keys stay yours, encrypted in a dedicated vault, and every token is billed by your provider at your rates. Nemu adds no markup and never resells inference. Credits pay only for the services nemu itself runs: search, reading, research and indexing.