We're building the
framework we always wanted.

Six months of LangChain before we broke.

In late 2024, our team was building a multi-agent research system for a fintech client. We chose LangChain because it was the obvious choice. Six months later, we had 4,000 lines of orchestration code and a system that failed 1 in 10 times under production load.

We tried CrewAI. Better ergonomics, but the state management still wasn't there. We tried LangGraph — closer to what we wanted, but the learning curve was steep and the deploy story was non-existent.

So we did what engineers do when nothing fits: we built it ourselves. First as an internal library, then as an open-source project, then as the thing you're reading about right now.

n00dles is our answer to the question "what would a multi-agent framework look like if it was designed for production from day one, by people who've had their pipelines wake them up at 2am?"

Okay, but why "noodles"?

We get asked this constantly, and honestly, the metaphor holds up better than we expected when we picked the name over beers:

• 🍜Noodles are long, connected, tangled threads — that's an agent chain. One agent's output twists straight into the next agent's input.
• ⛓Multiple noodles in one bowl — that's parallel agents. A dozen pipelines can run at once, sharing the same plate, never tangling with each other.
• 🥣A bowl holds it all together — that's the orchestration runtime. Without it, you just have a mess of cold noodles on the counter. With it, dinner's ready.