LangSmith’s documentation now describes routing OpenAI Decisions API requests through the LLM Gateway, rather than presenting the API only as a direct provider call.
Observed facts: The documented endpoint is POST /openai/v1/decisions, and the example uses a workspace-scoped LangSmith key plus an OPENAI_API_KEY provider secret. Mechanism: The gateway routes native Decisions API requests while applying existing access, rate-limit, and budget controls and tracing inputs, answers, and token usage to LangSmith. This allows a structured-answer API to sit inside an existing governance and observability path instead of requiring a separate integration surface.
Why now: Structured classification and scoring outputs are useful to agent workflows, and routing them through shared controls reduces the need to treat each API family as an operational exception. First implication: Teams already using LangSmith Gateway can evaluate Decisions requests within their existing access and spend-policy setup. Evidence boundary: The documentation establishes the route, configuration, and stated controls, but does not demonstrate latency, accuracy, rollout scope, or comparative performance. First watch condition: Watch for documented coverage of additional Decisions request types and concrete operational reports from deployments. This is an integration update, not evidence that a particular decision workload is more reliable or cheaper.