Stacks and Flows

LangChain

Leading OSS framework for LLM-powered agents and stateful multi-step workflows — Python and JS, with LangGraph for orchestration.

framework open-source Updated 2026-05-09
Visit LangChain →

Pros

  • Largest ecosystem in agent frameworks — connectors, memory backends, vector stores, evals
  • LangGraph (the orchestration sibling) is genuinely good at stateful, multi-step agents
  • Model-agnostic — swap OpenAI, Anthropic, open models, or local LLMs at the config layer
  • Massive community; almost every tutorial and template online uses it
  • LangSmith (paid) provides production-grade tracing and evals

Cons

  • Reputation for over-abstraction is partly earned — you can fight the framework before you fight the problem
  • API churn has historically been rough; v0.1 → v0.2 → v0.3 broke a lot of code
  • LangChain (the framework) and LangGraph (the orchestrator) confuse newcomers about what to use when
  • No affiliate program; LangSmith has paid tiers but no revenue share for content sites
  • Production monitoring is paid (LangSmith) — the OSS path is observable, but not turnkey

Best for

  • Teams who want model portability and the largest connector ecosystem
  • Engineers building stateful, multi-step agent workflows (LangGraph)
  • Anyone who wants to read 100 community examples before writing their own

What it is

LangChain is the most popular open-source framework for building LLM-powered applications and agents. It started in 2022 as a “chain primitives” library and has grown into a sprawling ecosystem: the langchain core, LangGraph for stateful orchestration, LangSmith for observability and evals (paid SaaS), and a templates/hub for community work.

For agent builders, LangGraph is the part that matters most. It’s a graph-based orchestration framework where nodes are functions or agents and edges are typed transitions — closer to what an experienced engineer would draw than what the original LangChain “chains” abstracted away. As of 2026 LangGraph is the recommended path for serious agent work; LangChain core is more useful as the connector layer underneath.

Who it’s for

LangChain/LangGraph is the right pick for engineering teams who want maximum optionality. Model-agnostic, vector-store-agnostic, deployment-agnostic. If you genuinely don’t know yet whether you’ll run on OpenAI or Claude, on Pinecone or Weaviate, on AWS or Cloudflare — LangChain lets you defer that decision longer than any other framework.

It’s a poor fit for teams who want minimal abstraction (use the official OpenAI or Claude SDK), and for teams who want a polished no-code UI (use Dify or Lindy).

Strengths

  • Ecosystem. Every major model, vector store, document loader, memory backend, and tool integration has a LangChain wrapper. You will not be the first person solving your problem.
  • LangGraph is solid. Once you’re past the LangChain-core learning curve, LangGraph is a genuinely good orchestration layer. State machines, checkpointing, human-in-the-loop, time travel — all first-class.
  • Model portability. Swap GPT-4o for Claude Sonnet for a local Llama by changing a config line. This is real and it works.
  • Community. Thousands of templates, tutorials, courses, and community-maintained integrations. Stack Overflow answers exist.
  • LangSmith for production. The paid observability layer is genuinely good — traces, evals, prompt management. Worth its price if you’re shipping.

Weaknesses / Watch out

  • Abstraction tax. LangChain has a reputation for being “abstractions on top of abstractions.” Some of that is fair, especially for simple use cases where the official model SDK would be 5 lines and LangChain is 50.
  • API churn history. v0.1 → v0.2 → v0.3 transitions broke a lot of community code. The current API is more stable, but there’s still scar tissue.
  • Two frameworks, one brand. “Should I use LangChain or LangGraph?” is the most-asked beginner question. The answer is “use LangGraph for agent orchestration, LangChain for the connectors LangGraph uses underneath” — but it shouldn’t take a paragraph to explain.
  • Production observability is paid. LangSmith is good but it’s not free. Without it, you’re stitching together OpenTelemetry yourself.
  • No affiliate. Pure SEO traffic value for publishers — no revenue share on LangSmith referrals at the time of writing.

Best paired with

  • LangGraph specifically for agent work — don’t reach for LangChain core’s older agent abstractions in 2026; LangGraph is the path.
  • Anthropic Claude or OpenAI as the model layer — both are well-supported, and the framework’s portability story actually pays off here.
  • Postgres or Redis for LangGraph checkpointing in production deployments.

Verdict

Recommended for portability-conscious teams. LangChain/LangGraph is the largest, most flexible framework in the category, and LangGraph specifically is the right answer for stateful multi-step agents in 2026. The abstraction tax is real but lower than it used to be, and the model portability is a genuine asset for teams that want to hedge across vendors. For minimal-abstraction OpenAI or Claude work, the official SDKs win on simplicity. For visual flow building, Dify or Flowise. No affiliate — pure technical recommendation.

Sources

FAQ

Is LangChain free?
LangChain has a free tier or open-source edition. See pricing details on the official site for paid features and usage limits.
What is LangChain best for?
Teams who want model portability and the largest connector ecosystem Engineers building stateful, multi-step agent workflows (LangGraph) Anyone who wants to read 100 community examples before writing their own
What are the main downsides of LangChain?
Reputation for over-abstraction is partly earned — you can fight the framework before you fight the problem API churn has historically been rough; v0.1 → v0.2 → v0.3 broke a lot of code LangChain (the framework) and LangGraph (the orchestrator) confuse newcomers about what to use when
Who should use LangChain?
Leading OSS framework for LLM-powered agents and stateful multi-step workflows — Python and JS, with LangGraph for orchestration. See our review for the full pros and cons.