Riften is an OpenAI- and Anthropic-compatible gateway that routes each LLM request to the lowest-cost model capable of completing it, using open-weight models Riften hosts whenever they suffice. Because it sits in the request path, real task results create training evidence the company controls, which Riften uses to improve routing and train private models on open weights around recurring work. Over time more traffic moves off rented APIs onto company-owned weights and controlled infrastructure, so the company's work becomes intelligence competitors cannot buy.