Use multiple LLM backends in a single crate, simple builder-based configuration, and built-in prompt chaining & templating.
