All notable changes to Switchyard are documented here. The format is based on Keep a Changelog, and this project adheres to Semantic Versioning.
First public release of Switchyard — a typed, composable control plane for LLM traffic that sits between client applications and LLM backends.
- Four-role chain —
RequestProcessor → LLMBackend → ResponseProcessor → TranslationEngine, executed by the Rust-backed core. See Architecture. - Protocol translation — convert between OpenAI Chat Completions, Anthropic Messages, and OpenAI Responses wire formats, so each client keeps speaking its native API regardless of the upstream backend.
- YAML route bundles (
switchyard serve --routing-profiles) — one bundle, many named routes, each its own chain. Supported routetypes:model,passthrough,random_routing,cascade,deterministic(LLM-as-classifier),latency_service, andnoop. - Routing strategies — weighted random split, signal-driven cascade escalation (see Cascade Routing), LLM-as-classifier strong/weak routing, and latency-aware multi-endpoint failover.
- One-command launchers —
switchyard launch claude,launch codex, andlaunch openclawspin up a local proxy and drop you into the target CLI. All three default to LLM-as-classifier routing (validated coding-agent trio) with--model/--routing-profilesto opt out. - CLI —
serve,launch,configure(saved defaults,--show,--list-models), andverify/launch --smokeround-trip checks. - Observability — Prometheus
/metrics, a JSON/v1/stats(/v1/routing/statsalias), and per-request cost/token/latency stats. See Metrics Reference. - Python library —
SwitchyardRecipes(passthrough_recipe,random_routing_recipe,cascade_recipe,deterministic_routing_recipe, …) and typedChatRequest/ChatResponsecontainers for in-process use. - Rust core (PyO3) — chain execution, the latency-aware router, and the tool-result signal collector are implemented in Rust and re-exported to Python.
- Packaging —
pip install nemo-switchyardwith optional extras[server],[cli],[gpu],[all]. See Installation.
--plan-executelauncher flag — slated for removal; plan-execute will be configured through a--routing-profilesYAML bundle instead.
- The
--deterministiclauncher flag was removed during pre-release development — LLM-as-classifier routing is now the implicit default for theclaude/codex/openclawlaunchers. - Inference Hub integration docs are out of scope for this release.