Backend-neutral graph IR, automatic kernel fusion, and async capture/replay for LLM inference.
Build the workspace, run graphs and kernels as a compute engine, or use the LLM Driver API and embed crates in your own project.
Driver
Run poot-serve for an OpenAI-compatible HTTP API that bounds requests. It serves generation, embeddings, reranking and LoRA adapter administration.
poot-serve
Read how tracing, fusion, capture/replay, and backend lowering fit together on a fine-grained graph IR.
