LangChain is launching LangSmith LLM Gateway, a runtime governance layer that sits between AI agents and LLM providers. It enforces spend limits at org/workspace/user/API key levels, redacts PII and secrets before requests reach the model, and surfaces policy violations as traceable events directly inside LangSmith. Setup requires only a base_url swap. The gateway eliminates the need to stitch together separate guardrails platforms, audit pipelines, and observability stacks. Future plans include tool call governance, MCP server controls, model fallbacks, and deeper security controls.
Table of contents
The gap between observability and enforcementWhat’s in beta todaySetup is intentionally boringGovernance on the same surface as build, observe, and evaluateChoosing where governance should liveWhere this is goingTry it today9 Impressions