LangGraph

The LangGraph overview page contains the clearest map of how the vendor's stack fits together, which is worth quoting because the naming is otherwise confusing: LangChain is the agent framework (abstractions and integrations for models, tools and agent loops); LangGraph is the orchestration runtime (durable execution, streaming, human-in-the-loop, persistence); Deep Agents is an agent harness layered on top of LangGraph (planning, subagents, filesystem tools, context management); LangSmith is the platform for tracing, evaluation, prompts and deployment across frameworks; LangSmith Engine detects issues in LangGraph traces and proposes fixes, and can open a pull request with the fix directly from its Engine tab; LangSmith Fleet is the no-code agent builder. Installation is one line - `pip install -U langgraph` - and the hello-world example is a three-node StateGraph, so the floor is low even though the ceiling is not.

• Graph-based agent orchestration mixing deterministic nodes with LLM-driven nodes
• Durable execution for long-running, stateful workflows
• Persistence of agent state across steps and runs
• Streaming
• Human-in-the-loop interruption points
• Runs standalone - LangChain components are used throughout the docs but are not required
• Optional LangSmith tracing via a LANGSMITH_TRACING environment variable and API key
• Deployment via the Agent Server to Cloud or self-hosted, managed through a Control Plane API

Python (and JS) engineering teams building bespoke agents where control matters more than speed of assembly - workflows that run for a long time, hold state, need to pause for a human, and need certain steps to be provably deterministic rather than model-decided. Regulated or high-stakes domains where "auditable where it matters" is a requirement, not a nice-to-have.

The docs themselves tell you who should not use it: "LangGraph is very low-level", it recommends you first understand models and tools, and it explicitly redirects anyone "just getting started with agents or [who wants] a higher-level abstraction" to LangChain's prebuilt agents instead. If your workflow is a straightforward tool-calling loop, the graph is overhead you will pay for in code you have to write and maintain yourself. It also deliberately does not abstract prompts or architecture, so none of the scaffolding that makes a first agent quick to stand up is provided here.

Setup assessment: 30m

Getting a graph running locally is a pip install and a few lines of Python. We put it at 30 minutes rather than 5 because a useful graph requires designing state and control flow, and because the first thing the docs recommend after install - wiring LANGSMITH_TRACING and an API key, then setting up LangSmith Engine - is a second setup step. Production deployment via the Agent Server is a separate, much larger exercise.

Prerequisites

• Python (pip/uv install langgraph) and working familiarity with models and tools as concepts
• A model provider - LangGraph orchestrates, it does not supply the model
• For tracing: a LangSmith account and API key
• For deployment: the Agent Server on Cloud or self-hosted, plus a GitHub integration for the deployment control plane

Authentication

Not documented in what we retrieved.

Machine-readable API

No OpenAPI link in the reviewed source set

• GET /v1/integrations/github/install - list GitHub integrations available for LangGraph Platform Cloud SaaS
• POST /v2/deployments - create a deployment (takes integration_id from the GitHub integrations endpoint)
• Create/get deployment revision, interrupt deployment revision, delete deployment(s)
• GET free deployment count
• Agent OAuth connections and token management under /v2/auth/agents/{agent_id}/

• The machine-readable OpenAPI spec that the API reference cites as its source
• Persistence/checkpointer backend documentation
• The JS/TS LangGraph documentation tree
• Social profiles
• A machine-readable OpenAPI specification was not included in the reviewed source set.
• Published pricing was not available in the reviewed source set; confirm current commercial terms with the vendor.