LangGraph v1.0.10: Type-Safe Streaming API With Per-Mode TypedDicts, Retry Middleware, and Content Moderation Middleware Ship on March 10

LangChain Docs·medium signal

LangGraph released v1.0.10 on March 10 with three notable developer-experience upgrades: type-safe streaming via version='v2' parameter (each stream mode now returns a typed TypedDict — ValuesStreamPart, UpdatesStreamPart, MessagesStreamPart, etc. — enabling full editor autocomplete and type narrowing), model retry middleware with configurable exponential backoff for flaky model calls, and content moderation middleware using OpenAI's moderation API to filter unsafe tool inputs. The type-safe invoke API also coerces output to declared Pydantic models or dataclasses automatically. These additions address the three most-cited LangGraph production pain points: type ambiguity at streaming boundaries, model call reliability, and input safety.