How is breadboard different from...

[sorvell]

• https://www.langchain.com/
• https://cloud.google.com/vertex-ai
• https://openai.com/blog/introducing-gpts
• https://huggingface.co/

[dglazkov]

Excellent question! We really need to have a nice crisp answer here.

Some rough notes that need to be made more coherent:

• breadboard is not a service or an online community (so it's not like huggingface)
• breadboard is model-agnostic (so it's not like gpts)
• breadboard does not aspire to be a full development platform that encompasses the entire workflow (so it's not like Vertex AI)
• breadboard could be used with Vertex AI or huggingface
• breadboard is first and foremost a method for composing AI recipes, rather than a cohesive collection of AI recipes (so it's not langchain) A good Webby metaphor is Lit (breadboard) vs. Material Design (langchain)
• you can build recipes with breadboard and create curated collections of them that may look like langchain, or you could use langchain recipes in breadboard, or breadboard recipes in langchain. They sort of operate at different layers, even though they overlap (langchain has LCEL, which if you squint a little, looks similar to breadboard)
• breadboard focuses primarily on the composition aspect of building AI recipes, and relies on the idea of a common interchange format for describing recipes, aspiring to grow as the hourglass architecture.

[sorvell]

Thanks for the answers, that's really helpful! So much that I'll risk pressing my luck =)...

• If breadboard is most like LCEL, what distinguishes it? Why isn't LCEL "good enough"? Is there something specifically missing or is breadboard a fresh take on a similar solution because it isn't obvious that LCEL has all the answers?
• Does Vertex AI have a layer in their stack that's similar to breadboard and might it make technical sense at some pointto integrate breadboard with Vertex AI?

[dglazkov]

Both breadboard and LCEL are WIPs -- hard to answer definitively. Directionally, LCEL is LangChain-first and breadboard is composition-first. Applying the Webby analogy from above, Material Design implementation might grow a component framework that if you squint, looks like Lit, but ultimately is mostly about composing MD bits more easily. Currently (and probably for the long term), LCEL is a subset of capabilities of breadboard.

Ideally (🤞🏻) the breadboard runtime one day powers LangChain's composition. I'd love to help LangChain get there.

Re: Vertex AI. I certainly hope that Cloud folks realize how cool Breadboard is and adopt it for AI recipe orchestration. They don't really have anything like that currently.

[sorvell]

Thanks, that's really helpful and some exciting ambitions!