langchain-ai/langgraph-swarm-py is a swarm-style multi-agent library that uses LangGraph so that multiple agents work by handing control over to each other. It is written in Python and released under the MIT license. The distribution package name is langgraph-swarm, and the latest release is 0.1.0. That said, the 0.1.0 change log is not listed in the release information, so it cannot be confirmed. The code is published in the github.com/langchain-ai/langgraph-swarm-py repository.
The problem this library addresses is work handoff between multiple agents. A swarm is a pattern in which an agent, upon receiving a request, keeps the work going by handing over control according to its own expertise. The README explains that it remembers the last activated agent and continues the conversation in the next interaction. It provides inter-agent collaboration and customizable handoff tools by default, and also supports streaming responses (receiving results little by little before they are complete), short-term and long-term memory, and human-in-the-loop (a human stepping in midway). The repository was created on February 19, 2025 and has now collected 1580 stars.
Some things remain unverified as well. There is no public information backing performance benchmarks or real-world use cases. Since 0.1.0 is an early-stage version, it is hard to predict how the interface will change going forward. So this library can be viewed as a reference case for learning the swarm pattern, but there is not enough basis for judgment to apply it straight to a production environment.