langgraph 1.2.19 — langgraph==1.2.19
- Source
- github.com
- Author
- github-actions[bot]
- Date

Context
LangGraph is LangChain's open-source library for building agents as graphs, where each node is a step such as a model call or a tool call. Version 1.2.19 is a maintenance release, and nearly every change concerns , the live feed of events a running graph emits so an application can show progress as it happens.
The clearest fix is in pull request 9406. A tool can pause itself to wait for an answer, such as an approval, by calling LangGraph's interrupt function. Because the underlying LangChain library treated that pause like any other exception, the tools stream reported the waiting tool as a tool error, and a tool containing a that was waiting for approval showed as failed too. Now a pause ends the tool's events without an error. Until the run resumes, the paused tool has a start event and no end event; on resume it runs again and reports both.
The other fixes in the release notes also make the stream match what the graph actually did. Message and tool events now carry the namespace of the graph that produced them, a graph streamed inside a node now emits its node-output messages, RemoteGraph.stream respects subgraphs=False under a streaming parent, the trace_policy setting no longer rewrites streamed node payloads, and awaiting a task while streaming returns its value. The one new feature lists a ToolNode's tool names in graph node metadata. The changelog also includes version 3.1.3 of the checkpoint-sqlite package.
- streaming — Sending a model's response token by token as it is generated, so the reader sees text immediately instead of waiting for the whole answer.
- multi-agent — Using several AI agents on one problem — splitting work in parallel, checking each other, or filling different roles like planner and reviewer.
Checking sign-in…
Loading comments…




