Arcweld is a proposed server intended for the primary goal of experimenting with novel LLM inference technologies, the secondary goals of a) reducing the cost of using hosted APIs when it would not interfere with the primary goal and b) providing space to train models.