← Back to stack

Builder infrastructure

LangChain

The platform for agent engineering

Visit official site →

LangChain is best known as the open-source framework for building LLM applications, but the entity as it exists today is a full commercial platform spanning the agent development lifecycle: build (LangChain, LangGraph, Deep Agents), test (datasets, evaluations, prompt engineering), deploy (agent server, cloud or self-hosted) and monitor (LangSmith tracing). The commercial layer is LangSmith, available as Cloud or self-hosted.

From official docs

What LangChain is

From official docs

The documentation site organises everything around a four-stage agent lifecycle and a platform layer beneath it. The frameworks stay open source and free to use; the money is in LangSmith and the LangGraph Platform deployment surface. The Control Plane API is substantial - the llms-full.txt index alone enumerates whole API families for deployments and revisions, agent connections, and an auth service that manages OAuth providers and tokens on behalf of agents, including creating an OAuth provider by MCP auto-discovery. That last detail is the giveaway about where the product is aimed: agents that hold their own credentials and call third-party services on a user's behalf.

What it can do

Vendor's own claim
  • Building agents in code via the LangChain and LangGraph open-source frameworks, plus Deep Agents
  • LangSmith: tracing, debugging and observability for agents in production
  • Evaluation with datasets and prompt engineering tooling
  • Agent Server for deploying and serving agents at scale, on Cloud or self-hosted
  • LLM Gateway to route, control and observe traffic across model providers
  • LangSmith Fleet - building and running agents without code
  • LangSmith Engine - automated detection and remediation of recurring agent issues
  • A Prompt & Context Hub, managed sandboxes, and an open-source terminal coding agent (dcode CLI)
  • A Control Plane API covering deployments, agent auth/OAuth connections and GitHub integrations

Who it is for

Our read

Engineering teams building agentic applications who want a single vendor across the framework, the deployment runtime and the observability layer. The self-hosted LangSmith option and the published HIPAA / SOC 2 Type 2 / GDPR posture point squarely at regulated enterprises as the paying end of the market.

When to choose something else

Our read

Nothing here is usable without engineers - the entry point is "build agents with code", and even LangSmith Fleet's no-code promise sits on a platform whose primary API surface is a deployment control plane. It is also a broad commitment rather than a component: adopting the framework, the agent server, the gateway and the observability layer together means one vendor owns your build, runtime and telemetry at once, which is exactly the lock-in that OTel-native competitors sell against.

Implementation considerations

Our read

Setup assessment: custom

"Sign up for LangSmith" is free and quick, and adding tracing to an existing LangChain app is small. But the platform-shaped path - deploying an agent server, configuring cloud or self-hosted LangSmith, wiring the LLM Gateway and governing users and compliance - is an infrastructure project whose length depends entirely on which pieces you adopt.

Prerequisites

  • Engineering capability in the framework's supported languages
  • A LangSmith account (Cloud) or infrastructure for self-hosted LangSmith
  • For deployments: a GitHub integration - the Control Plane API's create-deployment flow takes an integration_id obtained from the GitHub integrations endpoint
  • For agent auth: configured OAuth providers per third-party service the agent must act against

Developer and agent access

From official docs
Authentication
Not documented in the artefacts 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, returning the integration_id used when creating a deployment
  • POST /v2/deployments and create-deployment-revision, get/delete deployment, interrupt deployment revision
  • POST /v2/auth/authenticate - get an OAuth token or start the authentication flow
  • POST /v2/auth/providers/mcp-discover - create an OAuth provider via MCP auto-discovery
  • POST/GET/DELETE /v2/auth/agents/{agent_id}/connections - create, list and remove an agent's connections
  • POST /v2/auth/agents/{agent_id}/providers/{provider_id}/tokens - bind an imported OAuth token to an agent
  • GET /v2/auth/tokens/exists and /exists/batch - token-presence checks per provider

Verify before you adopt it

  • The machine-readable OpenAPI spec the API reference is generated from
  • API authentication scheme (the sample requests in the API reference use a documentation placeholder host, not the real base URL)
  • The Trust Center page backing the HIPAA / SOC 2 Type 2 / GDPR claims
  • The LangSmith status page URL
  • The marketing site (this entity's canonical URL is the documentation site)
  • A machine-readable OpenAPI specification was not included in the reviewed source set.