LangChain has released LangSmith LLM Gateway in public beta, giving teams one place to control how production agents call models across providers. The gateway sits between an application and model endpoints, so policies do not have to be rebuilt inside every agent.
Teams can set time-bound spending caps and rate limits at organization, workspace, API-key or user level. A request that exceeds its spending allowance receives a 402 response. Multi-tenant applications can also identify customers through a request header and enforce separate limits while routing traffic through one shared key.
Fallback rules can move requests to another model or host when a provider is unavailable or rate-limited. The gateway supports OpenAI, Anthropic and Fireworks, plus custom OpenAI- or Anthropic-compatible endpoints. It is built around bring-your-own provider keys, while optional hosted open-model inference is available through Fireworks.
Sensitive-data controls can detect and replace personal information and secrets before requests reach providers and redact them from LangSmith traces, but that feature is limited to Enterprise customers. The gateway itself is available now to Plus and Enterprise plans; broader guardrails and deployment controls are planned rather than available today.