Apps · Infra & proxies

LangGraph

Stateful graph runtime for agent workflows.

Tokens / 7d

58B

Routing directive

auto:agent

On our metal

23%

Licence

Open source

LangGraph nodes each make their own model call, which makes per-node model selection the main cost lever. Routing directives let a graph stay portable instead of hardcoding vendor model ids.

→ langchain.com/langgraph → All apps → Integration docs

How it connects

ChatOpenAI pointed at the gateway base URL per node.
Checkpointer replay reuses cached prefixes on retries.
Structured output enforced with a shared schema registry.

Traffic pattern

Graph replays repeat identical prefixes, so caching pays for itself on retry-heavy pipelines.

gpt-5.5claude-sonnet-4.5claude-haiku-4.5

Point it at the gateway

from langchain_openai import ChatOpenAI
llm = ChatOpenAI(model="auto:agent",
  base_url="https://api.llmcloud.ai/v1")