LangChain VectorStore integration for ShannonBase (langchain-shannonbase)

[apoorva-01]

Hey folks, your README lists LangChain under the integrations you'd welcome, so I built one: langchain-shannonbase.

It's a LangChain VectorStore that talks to the native VECTOR type directly (STRING_TO_VECTOR / DISTANCE), so anyone already working in LangChain can use ShannonBase as their store and drop it into whatever chains, retrievers, or agents they've already got.

store = ShannonBaseVectorStore(embedding=OpenAIEmbeddings(...), table="docs", host="127.0.0.1", database="rag")
store.add_texts(["Refunds within 30 days."])
store.similarity_search("return policy?", k=2)

I know sys.ML_RAG already does the whole retrieve-and-generate loop in a single SQL call, and honestly that's the nicer path when you're all-in on ShannonBase. This is for the other crowd: people with an existing LangChain app, or who want to bring their own embeddings (OpenAI, whatever) and keep the wiring in Python. Different front door, same engine underneath.

Works on MySQL 9 and HeatWave too since they share the surface, but ShannonBase is what I point people to for a local spin (no HeatWave subscription needed to try it).

Couple of asks - I'm happy to send a PR linking it wherever you keep the ecosystem stuff, if that's useful. And if there are corners of the VECTOR surface I should watch out for, or changes on the way, point me at them and I'll keep it current.

Thanks for building this. Genuinely fun to write against :)

[ShannonBase]

Yes, PRs are welcome! And yes, you gave this example:

store = ShannonBaseVectorStore(embedding=OpenAIEmbeddings(...), table="docs", host="127.0.0.1", database="rag")
store.add_texts(["Refunds within 30 days."])
store.similarity_search("return policy?", k=2)

This is a great drop-in solution for teams that are already working with LangChain's VectorStore interface and don't want to rewire their existing chains, retrievers, or agents. It lets them keep everything in Python while still leveraging ShannonBase as the backing store.

It's not the only way to do it—sys.ML_RAG is definitely cleaner if you're building fresh on ShannonBase—but for teams with existing LangChain investments, this gives them a smooth on-ramp without having to change their whole stack.