This repository was archived by the owner on Mar 29, 2026. It is now read-only.
Repository navigation
Replies: 0 comments
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Uh oh!
There was an error while loading. Please reload this page.
Hi,
The migration guide
https://vanna.ai/docs/migration
does not clearly explain how (or if) it is possible to keep using the legacy Postgres / pgvector connector with Vanna 2.0+.
In Vanna 0.x we used patterns like:
`from vanna.pgvector import PG_VectorStore
from vanna.openai import OpenAI_Chat
class MyVanna(PG_VectorStore, OpenAI_Chat):
...`
However, in a fresh Vanna 2.x installation, the module vanna.pgvector is no longer available (even when installing extras like postgres or pgvector). This makes it impossible to instantiate a legacy vn object in the same environment, which in turn makes the LegacyVannaAdapter unclear in practice.
So the questions are:
Is vanna.pgvector.PG_VectorStore officially supported in any Vanna 2.x release?
If not, how is the LegacyVannaAdapter expected to be used if the legacy connectors are not importable anymore?
Is the intended migration path:
a) run Vanna 0.x in a separate environment/process and share the same Postgres RAG DB, or
b) reimplement pgvector-based RAG using external tools (LangChain / LlamaIndex) when moving to Vanna 2?
If Postgres + pgvector is still a supported RAG backend in Vanna 2, is there an official example showing how to do this without vanna.pgvector?
Right now the migration guide suggests that legacy code can be wrapped, but in practice the required legacy modules are no longer present in the Vanna 2 package, which makes this confusing.
Any clarification on the officially supported approach would be greatly appreciated.
Thanks!
All reactions