Agent frameworks
langchain-ai/langgraph
LangGraph is an orchestration library for agents whose work cannot be reduced to one prompt. It represents a process as a graph of explicit steps and transitions, while retaining state between them. That makes retries, pauses, human review and recovery visible in the design instead of hiding them inside a long conversation.
THE PRACTICAL EXPLANATION
LangGraph is an orchestration library for agents whose work cannot be reduced to one prompt. It represents a process as a graph of explicit steps and transitions, while retaining state between them. That makes retries, pauses, human review and recovery visible in the design instead of hiding them inside a long conversation.
WHERE TO USE IT
Use it for case management, research, document review or operational workflows that may pause for approval, revisit an earlier step or continue over a long period. It is particularly useful when the path depends on evidence gathered during the run.
WHO MAY USE IT
Python engineering teams building controlled agent services, especially where product owners and risk specialists need to understand how work moves from one state to another.
HOW TO USE IT
Begin with the business state machine, not an autonomous-agent demonstration. Name the states, permitted transitions, evidence required at each transition and the person authorised to approve consequential actions. Implement one narrow graph, add durable checkpoints and test interrupted runs before adding more agents.
GETTING THE BEST RESULTS
WHY IT MAY BE USEFUL
Build resilient, stateful agents as explicit graphs with durable execution and human checkpoints.
Best considered for: Long-running agents, controlled workflow state and multi-agent orchestration.
READ BEFORE YOU ADOPT IT
Powerful, but teams must design state, recovery and approval rules carefully.
Confirm the current licence, maintenance status, dependency risk, data path, model access, tool permissions and human approval points at the source. A public repository is inspectable raw material—not proof that a system is secure, supported or suitable for your production environment.