Agent frameworks · ranked 05
LangGraph
Graph-shaped agents with state, from the LangChain team.
Best for Agents that are workflows, not a single loop
LangGraph is an open-source library from the LangChain team for agents that are graphs of steps with explicit state. Use it when the agent has branches, loops, or a human in the loop and a single prompt chain is not the design. It is available in Python and JavaScript.
Best use cases
- Multi-step agents with durable state
- Human approval in the middle of a run
- Production agents that grew out of a LangChain prototype
Pros
- + Open source
- + State and control flow are explicit
- + Sits in the LangChain ecosystem if you already do
Cons
- − More machinery than a weekend chat endpoint needs
- − You still bring a model and an opinion about the graph
Pricing
The library is open source. Hosted deployment options: check site.
Alternatives
Agent frameworks
6.7
LangChain
The original LLM application framework, plus a hosted LangSmith side.
Best for A general LLM app framework with a huge ecosystem
Agent frameworks
8.0
Mastra
TypeScript framework for agents, workflows, and evals.
Best for TypeScript agents you keep in your repo
Agent frameworks
7.6
Pydantic AI
Python agent framework from the Pydantic team, typed end to end.
Best for Typed Python agents