InferMux
Inference routing for AI systems.
Your application calls one model today and three tomorrow. InferMux sits between your code and your providers — you configure routing rules, and it handles the dispatch, failover, and cost tracking without changing your call sites.
go get github.com/greynewell/infermuxOne interface, many providers
Switching model providers usually means rewriting integration code. InferMux abstracts the provider behind a common Go interface so you can swap OpenAI for Anthropic, or route some traffic to a self-hosted model, without touching application logic.
InferMux is part of the MIST Stack. It shares a message protocol with MatchSpec (evaluation), SchemaFlux (data), and TokenTrace (observability). Routing decisions, cost data, and latency metrics all flow through the same wire format. The methodology behind the stack is Eval-Driven Development.