RUG::B

RubyLLM 1.x gave Rubyists a beautiful way to work with AI: one delightful DSL across major AI providers, easy Rails integration, a model registry for capabilities and costs, instrumentation, agents, workflows, and RAG.

But once the work moves into production, a new set of problems appears. Calls get expensive fast. Providers have bad days. Answers need provenance. Workflows need to be traced. And sometimes the provider you need speaks more than one protocol, or is not yet supported.

RubyLLM 2.0 is about crossing that line. It adds caching and batch execution to cut costs, model fallbacks so your app can keep answering, citations so users can see where answers came from, and workflow-aware tracing so you can finally understand complex flows. It also separates providers from protocols so multi-protocol providers work natively, and smaller providers can be added with a simple generator.

1.x was a beautiful way to work with AI. 2.0 is how Rubyists build real AI products without giving up any of that beauty.