v0.4.0: local reranking, 83 models - add Infinity reranking service (michaelf34/infinity:0.0.77-cpu) - mxbai-rerank-xsmall-v1 cross-encoder for RAG pipelines (~70MB, ONNX) - local-infinity-rerank model routed via LiteLLM - resource limits in recommend-limits.sh - 83 models across 12 providers