LangGraph

Founded by Harrison Chase in 2024

Build and orchestrate AI agents with fine-grained control and reliability

Pricing model

Free trial, Paid

Rating

★ People love it

Time to value

Moderate Setup (1-3 hours)

You can use LangGraph to design and run AI agents that handle complex, multi-step tasks. It gives you low-level control over agent workflows, including single-agent, multi-agent, and hierarchical setups. You can add human-in-the-loop checkpoints to approve or redirect agent actions, persist memory across sessions for context, and stream agent reasoning token-by-token in real time. It works with any model provider and is free to use under an MIT license.

What LangGraph does

Define agent workflows with custom control flow logicAdd human approval steps at specific points in an agent's executionStore and retrieve conversation history across multiple user sessionsStream agent output token-by-token to a user interfaceConnect multiple agents in a hierarchical or peer-to-peer structureDebug agent decisions step-by-step using LangSmith integrationRoute tasks between specialized subagents based on contextDeploy agents to production with state persistence and error handlingAdd human-in-the-loop checkpoints to approve or redirect agent actions mid-runBuilt-in memory that persists conversation history and context across sessionsNative token-by-token streaming so users can see agent reasoning in real timeSupports single-agent, multi-agent, and hierarchical agent control flows in one frameworkLow-level primitives that let you customize agent logic without being locked into a preset architectureMIT-licensed and free with no overhead added to application performanceWorks with any LLM provider without requiring a specific modelIntegrates with LangSmith for debugging, evaluation, and one-click deployment