NeuroCore vs LangGraph

LangGraph is excellent for explicit graph/state-machine orchestration. NeuroCore is a higher-level application chassis. They solve different problems and compose well.

Category difference

	LangGraph	NeuroCore
Primary unit	A graph of nodes/edges in Python	A YAML blueprint of skills
You write	Graph wiring + node functions	Declarative YAML + reusable skills
Distribution	Library code	pip install neurocore-skill-* plugins
Config	In code	neurocore.yaml + .env + env overrides
Providers	Bring your own	Injected (anthropic/openai/ollama/…)
Operations	Build it yourself	Built-in run history, replay, resume, approval
Scaffolding	—	neurocore new <template> + CLI

LangGraph optimizes designing control flow. NeuroCore optimizes shipping and operating an application around skills — packaging, configuring, running, streaming, recording, and resuming.

Use them together

NeuroCore doesn’t replace your agent stack — it’s the outer chassis. When you need LangGraph-specific control flow, wrap a compiled LangGraph graph as a NeuroCore skill:

from neurocore import AsyncSkill, SkillMeta

class LangGraphSkill(AsyncSkill):
    skill_meta = SkillMeta(name="my-graph", version="0.1.0",
                           consumes=["input"], provides=["output"])

    def init(self, config):
        super().init(config)
        from my_graphs import build_graph
        self._graph = build_graph()       # a compiled LangGraph app

    async def process(self, context):
        result = await self._graph.ainvoke({"input": context.get("input")})
        context.set("output", result["output"])
        return context

Now that graph is a discoverable, configurable, recordable NeuroCore component — usable from any blueprint, with run history and approval gates for free. The same adapter pattern works for LlamaIndex query engines, CrewAI crews, and the OpenAI Agents SDK.

The one-liner

LangGraph helps you design agent graphs. NeuroCore helps you ship agent applications.