The best LLM for LangGraph
LangGraph's strength is the loop — nodes call models, route on the result, and cycle until done. But every cycle is another model call, and a stateful graph that stalls on a provider error mid-run is a frustrating thing to debug. CodeBurst gives every node an OpenAI-compatible model endpoint that fails over across providers and repairs empty tool turns, so your graph runs to completion.
Loops multiply your exposure to provider failures
A linear chain makes a handful of calls. A LangGraph cycle — plan, act, observe, repeat — can make many, and any one of them hitting a provider's rate limit or returning an empty tool turn can halt the run. The more capable your graph, the more it loops, and the more it needs a backend that doesn't depend on a single provider being healthy.
What CodeBurst adds to every node
| Failure | CodeBurst |
|---|---|
| A node's call rate-limits | Reroutes to a healthy provider in the same request; the node returns normally. |
| Empty tool-synthesis turn | codeburst-agent retries with a corrected format. |
| One node needs more reliability | Give just that node codeburst-swarm for a multi-model vote. |
Configure the model once
LangGraph uses LangChain chat models, so point ChatOpenAI at CodeBurst and use it in your nodes:
from langchain_openai import ChatOpenAI
llm = ChatOpenAI(
model="codeburst-agent",
base_url="https://codeburst.ai/api/v1",
api_key="YOUR_CODEBURST_KEY",
)
# use `llm` inside your LangGraph nodes exactly as before —
# bind tools, invoke, stream; failover happens behind the endpoint.Your graph structure, tools and state are untouched. Swap models by changing one string, and provider failover comes for free.
Get started
Get an API key How agent failover works →FAQ
How do I use CodeBurst with LangGraph?
Configure ChatOpenAI(base_url="https://codeburst.ai/api/v1", model="codeburst-agent", api_key=...) and use it in your nodes.
Why route LangGraph through a router?
Loops mean many calls; CodeBurst fails over and repairs tool turns so the graph doesn't stall.
Does my code change?
No — only the model config; nodes, edges and tools stay the same.