Not Diamond

Description

Not Diamond is a hosted model router for LLM applications. Customers register the set of candidate models they want to consider and call Not Diamond's API with a query; a meta-model that learns when to call each LLM predicts the best target and the request is dispatched there. The product positions itself around quality maximisation and cost reduction — simple queries are routed to cheap models, complex queries to frontier models — without the customer hand-coding a routing policy. A pre-trained router gets users started quickly; custom routers can be trained on the customer's own preference data.

Solution

Hosted decision API. The application sends each prompt to Not Diamond with the configured candidate model list; Not Diamond analyses the input and predicts which model will provide the highest quality response at the lowest cost for that specific query, then (optionally) executes the call. Routing is stateless per request; configuration lives in the Not Diamond dashboard / API call parameters, and learned preference data drives the meta-model behind the routing decision.

Primary use cases

• automatic best-model selection per query
• cost reduction over single-model deployments by routing simple queries to cheap models
• quality-maximising routing across heterogeneous models
• managed routing without authoring rules
• custom router training on customer preference data