LangGraph SDK

The LangGraph SDK is the client library for interacting with LangGraph Platform deployments — providing programmatic access to create, run, monitor, and manage LangGraph workflow executions from application code, with support for streaming outputs, async execution, and multi-tenant workflow management.

What the LangGraph SDK provides

The LangGraph SDK is the interface between application code and a deployed LangGraph Platform instance. It provides methods for creating new workflow threads (individual execution contexts), running workflows with specified inputs, streaming intermediate outputs as they are generated, checking workflow status, resuming interrupted workflows, and accessing historical run data. The SDK handles the HTTP communication with the LangGraph Platform server, authentication, and error handling, so application code works with Python objects rather than raw API calls. Both synchronous and asynchronous interfaces are available to fit different application architectures.

Thread and run management

LangGraph SDK organizes workflow execution around two concepts. A thread is a persistent execution context — it holds the state for a workflow instance and accumulates history across multiple interactions. Runs are individual executions within a thread — each message or input creates a new run that continues from where the thread left off. This model supports conversational agents (each conversation is a thread, each turn is a run) as well as long-running autonomous workflows (a task is a thread, each step completion creates a new run). Thread management through the SDK includes creating threads, listing active threads, retrieving thread state, and deleting threads when they are no longer needed.

Streaming and real-time output

Streaming support in the LangGraph SDK allows applications to receive workflow outputs as they are generated rather than waiting for the entire workflow to complete. The SDK provides streaming iterators that yield events as the workflow executes — model token outputs, tool call initiations, state updates, and completion events. This is the mechanism behind real-time agent UIs that show users what the agent is doing while it works. Consuming streaming output requires handling an event stream asynchronously, which suits web applications and chat interfaces but requires more care in synchronous application contexts.

LangGraph SDK — FAQ

Do I need the LangGraph SDK if I'm self-hosting LangGraph?

The LangGraph SDK is designed for use with the LangGraph Platform's API server. For self-hosted deployments that use the same server implementation, the SDK works against the self-hosted endpoint by configuring the base URL. For fully custom LangGraph deployments that wrap the framework in their own API server, the SDK may not apply directly — you interact with your own API instead. The SDK is most relevant when using the LangGraph Platform's managed service or when self-hosting the official LangGraph server.

Can the LangGraph SDK handle authentication for multi-user applications?

The SDK supports API key authentication for server access and can be configured with per-request headers for applications that need to pass user-specific credentials or metadata. For multi-user applications where different users need isolated workflow contexts, the standard pattern is to use separate threads per user and to manage access control at the application layer — ensuring that API calls made on behalf of a user only access that user's threads. The LangGraph Platform provides basic multi-tenancy through thread isolation; more granular authorization is typically implemented in the application layer.