github.devotd.
NewAI agents

langchain-ai/langgraphjs — Graph Agent Framework and CLI 1.5.4

Covers langchain-ai/langgraphjs, a TypeScript framework for building resilient language agents in graph form, and its latest CLI release @langchain/langgraph-cli@1.5.4.

Repositorylangchain-ai/langgraphjs ↗

Covers langchain-ai/langgraphjs, a TypeScript framework for building resilient language agents in graph form, and its latest CLI release @langchain/langgraph-cli@1.5.4.

GitHub description: Framework to build resilient language agents as graphs.

stars
★ 3.4K
Forks
610
Language
TypeScript
License
MIT
Latest release
@langchain/langgraph-cli@1.5.4
Read original source ↗View on GitHub ↗

langchain-ai/langgraphjs is a TypeScript framework for building resilient language agents in graph form. The repository is written in TypeScript and released under the MIT license. The latest release is the CLI package @langchain/langgraph-cli@1.5.4. The code is published in the github.com/langchain-ai/langgraphjs repository.

The README describes LangGraph as a low-level orchestration framework for controllable state-based agents. It carries long-running workflows forward on the basis of stateful graphs and persistence (keeping state around so it can be resumed later), and supports persistent execution, human-in-the-loop, and short- and long-term memory. Used together with LangSmith, it can cover debugging, observability (watching execution), and deployment, and it can be used alongside the LangChain products or independently. The repository was created on January 9, 2024 and has 3,367 stars.

That said, 1.5.4 is a CLI package release, and the provided records contain neither its changelog nor any performance benchmarks. Among the collected evaluation scores, momentum comes out low at 2.1 out of 10. The README calling it "low-level" only means a trade-off that requires you to design the flow yourself, and there are no figures comparing it against other orchestration tools. So this repository can be worth looking at when reviewing graph-based agent designs. However, judging runtime performance or the effect of adoption requires separate measurement.