Frameworks, Runtimes, and Harnesses

Status: ACTIVE (pulled from docs.langchain.com) Source: https://docs.langchain.com/oss/python/concepts/products Timestamp: 2026-05-11

LangChain maintains several open source packages. Understanding the differences helps choose the right tool.

Framework Runtime Harness
Value add Abstractions, integrations Durable execution, streaming, HITL, persistence Predefined tools, prompts, subagents
When to use Getting started quickly, standardizing team approach Low-level control, long-running stateful workflows Autonomous agents, complex non-deterministic tasks
Examples LangChain, Vercel AI SDK, CrewAI, OpenAI Agents SDK, Google ADK LangGraph, Temporal, Inngest Deep Agents SDK, Claude Agent SDK, Manus

LangChain (Framework)

Provides abstractions that make it easier to build with LLMs: structured content blocks, agent loop, middleware. Built on top of LangGraph but you don't need to know LangGraph to use LangChain.

When to use: Quick agent building, standard abstractions, straightforward applications.

LangGraph (Runtime)

Low-level orchestration framework for building, managing, and deploying long-running, stateful agents. Provides durable execution, streaming, human-in-the-loop, and persistence.

When to use: Fine-grained control, durable execution needs, complex workflows combining deterministic and agentic steps, production infrastructure.

Deep Agents SDK (Harness)

Opinionated, batteries-included framework with built-in tools on top of LangGraph: planning, task delegation, file system, token management, subagents.

When to use: Long-running agents, complex multi-step tasks, need predefined tools and prompts.

Feature Comparison

Feature LangChain LangGraph Deep Agents
Short-term memory Yes Yes StateBackend
Long-term memory Yes Yes Yes
Multi-agent Subagents Subgraphs Subagents
Human-in-the-loop Middleware interrupt() interrupt_on parameter
Streaming Yes Yes Yes
Skills Multi-agent skills - Skills