langchain-ai/langgraphjs — reverse-engineered prompt

Build me a TypeScript library for creating AI agents and long running workflows as graphs. I want a simple developer friendly API where I can define steps, connect them, keep shared state between steps, and run everything in a controlled way instead of a messy chain of prompts.

It should support the big things people expect from serious agents, like saving progress so a run can survive crashes and resume where it stopped, short term and long term memory, and human approval pauses where someone can inspect or edit the state before continuing. Please make it usable for general stateful workflows too, not just chatbots.

Include solid examples and docs that show how to build a few real agent flows, how to persist checkpoints, and how to pause and resume execution. I also want the project to feel production ready, with tests, clear error handling, and a nice package setup for Node. If you need details, look up the current docs online and mirror the overall behavior closely.