Tools
NVIDIA Rewrote Its LLM Routing Proxy in Rust — Switchyard v0.2.0 Lands 193 Commits and Splits Into Five Crates
NVIDIA-NeMo shipped Switchyard v0.2.0 on 2026-08-10, a ground-up redesign across 193 commits that replaces the v0.1.0 Python proxy with a native Rust server and a new `libsy` orchestration library. It translates between OpenAI Chat Completions, OpenAI Responses, and Anthropic Messages so coding agents like Claude Code can run against vLLM, NVIDIA NIM, Ollama, or any OpenAI-compatible backend, and splits into five crates (protocol, translation, libsy, llm-client, server) that can be embedded independently. The repo took roughly 370 stars in a single day (633 total) and NVIDIA still labels it pre-alpha, explicitly 'not for production use.'
Source
↳ Follow the thread