NVIDIA Dynamo GitHub repository (ai-dynamo/dynamo)

Repo description 'A Datacenter Scale Distributed Inference Serving Framework'; Apache-2.0 per README badge and SPDX headers (GitHub's license API reports NOASSERTION because of bundled NVIDIA-proprietary components in some paths). README states the architectural position explicitly: 'Dynamo is the orchestration layer above inference engines - it doesn't replace SGLang, TensorRT-LLM, or vLLM, it turns them into a coordinated multi-node inference system', built in Rust for performance and Python for extensibility. Disaggregated serving is listed as supported for all three backends (SGLang, TensorRT-LLM, vLLM); KVBM (block manager) is marked complete for TensorRT-LLM and vLLM and in-progress for SGLang. Use cases include 'independently scale prefill and decode' and KV-aware routing 'to avoid redundant prefill computation'.