RunAnywhere

RunAnywhere is an inference company that spans hosted models, on device acceleration and developer SDKs. Its hosted product, Wally, serves frontier scale open models behind an OpenAI compatible API. Sign in happens in the browser, and the vendor states that hosted execution is always explicit: a request leaves the machine only when the developer says so, and hosted is a flag rather than a fallback. For local use, the company builds NPU inference engines called MetalRT and QHexRT, aimed at the silicon devices already ship with, including the Apple Neural Engine and Qualcomm Hexagon NPUs. It states that every published benchmark names its hardware and method. Its open source SDKs offer six bindings on one C++ core so that behavior does not drift between platforms, and a console shows usage and spend. The company describes itself as a research first inference lab. The audience is developers who build local or hybrid AI applications on phones, laptops and other consumer hardware, and who want to move heavier jobs to a hosted open model without changing their stack. Agent setup instructions are published for use with coding agents. Users sign in to the console, buy credits and run requests, with usage and spend tracked live. Credit rates and per model prices are not published on the site, and no pricing page is available.