RunAnywhere

San Francisco startup serving open-weight models, with fast versions for agent loops, from the United States.

RunAnywhere is a San Francisco company from Y Combinator's Winter 2026 batch. It began with open-source SDKs for on-device inference and now hosts frontier open-weight models through Wally, an OpenAI-compatible API, on cloud capacity in the United States and a serving stack it builds itself. The catalogue covers GLM, DeepSeek, Kimi and MiniMax models, with separate fast versions of some of them for agent loops. A fit for long-context coding and agent work on current open models with a US host.