Riften builds earned intelligence: models that come out of a company’s own work instead of out of a lab. One environment variable points an existing application at Riften instead of OpenAI or Anthropic. Riften routes each request to the least expensive model likely to handle it, using open-weight models we host whenever they are sufficient. Customers lower inference costs before any training begins. Because Riften sits in the request path, real tasks and their outcomes create training evidence the company controls. Riften uses that evidence to improve routing and train private models on open weights around recurring work. Those models earn production traffic by meeting the same bar as the APIs they replace. Over time, frontier APIs handle only the tasks that still require them. The rest runs on models the company owns, on infrastructure it controls, inside the applications where the work happens.
