LangChain.
Orchestration primitives that scale from prototype to production agents.
What it is.
The original agent framework. Open-source Python and JavaScript libraries plus LangGraph for stateful workflows plus LangGraph Cloud for managed runtimes. The default starting point for builders who refuse to lock into a single model vendor.
Where it fits.
Multi-model stacks, third-party orchestration, and teams that want explicit graph control over agent flow. LangGraph is the production tool — the original LangChain library is increasingly used as the integration layer beneath it.
- Largest ecosystem of integrations in the category
- Model-agnostic — no vendor lock-in
- LangGraph offers explicit state machines for agent flow
- API surface accumulated cruft over multiple major versions
- Defaults can hide performance footguns from new builders
Frequently asked.
Is LangChain still the default agent framework?
Yes for multi-vendor and third-party orchestration. Single-vendor stacks increasingly use Anthropic, OpenAI, or Google's first-party agent SDKs.
What is LangGraph?
LangChain's stateful workflow primitive. Explicit graph nodes and edges replace the implicit chains the original library shipped.
Is LangSmith required to use LangChain?
No, but it is the dominant observability tool for LangChain stacks. Used by most production deployments above 100K daily invocations.
