Authentication is often the least-tested part of an app. Live environments need network access and real credentials, while mocks miss the failures that break production. @workos/emulate runs the Work OS API locally for development and CI. Seed users, organizations, RBAC roles, and SSO connections, then test full Auth Kit login flows, signed webhooks, token refresh, and error handling. Responses and event shapes come from the Work OS Open API spec, so tests exercise the same surface your app uses in production.

Imagine a developer builds an AI-based coding assistant that runs on a desktop computer. The model is available to download, the application is straightforward, and the machine has plenty of storage. But when the program tries to load the model, it runs out of memory. This is where local AI development becomes a hardware problem. Merely downloading a large AI model doesn't mean we can run it. Even if loaded, useful response time isn't guaranteed. A large AI model can run on modest hardware only by reducing memory usage, reducing calculations, or moving work to slower hardware.

Several techniques help: Quantization gives each weight a smaller representation; Layer-wise offloading moves weights as needed; Mixture of experts uses selected parts per token; Distillation lets a larger model teach a smaller one; Pruning removes less-contributing work; Speculative decoding proposes several tokens before checking. An AI model contains parameters or weights that influence input-to-output mapping. An 8B model has ~8 billion weights organized into layers. Input text is divided into tokens, processed into probabilities for the next token. Inference uses trained weights unchanged, unlike training which requires expensive infrastructure.