Hello!
I know that Supasbase can be used easily as a vector store tool with the Ai Agent node in n8n. However, I am considering whether I could use Xano embeddings instead of relying on Supabase. The added advantage in my case is that my backend is already on Xano, so using it would make more since as it would be easy to create relational fields and avoid duplicating data to store on Supabase as an external vector store.
I am interested to know if anyone explored this option for use with n8n Ai Agents?
Hey @DrHariri , Xano embeddings are not currently supported and you could raise a feature request for that.
There still could be a workaround if Xano implements OpenAI API specs since OpenAI (sub-)node allows for a custom base URL for the services that host the embeddings with OpenAI specs format. An example of such a service is OpenRouter.ai, though Xano is not listed there as the hosted model.
Hey! Thank you for your comment.
Yes I was hoping what others might be doing (even if using something other than Xano that is not directly supported in n8n). I think there are two parts:
Understand best ways to handle splitting of text before generating embeddings (Since I can’t use recursive splitters etc). The actual generation of embeddings is extremely easy to handle within Xano itself, or, I could use OpenAi’s Embeddings API.
Understanding best ways to pass the retrieved data (in this case would be retrieved through API) to an Ai Agent, since we can’t directly use the Vector store in Ai Agents or QA chains.
I am really hoping to learn from others but will begin to rely on APIs to do this. Thanks!
Hey, Greetings from Germany.
I deep dived into embeddings in Xano and use it for knowledge base semantic search.
the knoledge base is a standard table with fields and a vector field (indexed).
I created a trigger in Xano on update, checking if the new field values are is different to old field values. If different, I create a embedding using openAI embeddings via api.
So the vectores are only producing costs on the API of openAI if the content has changed.
For retrieving as a RAG i created an API Endpoint for a question. The Endpoint created an embedding using the same model than the trigger for the question and starts a search on the cevtor field.
Works fine…
I also build everything in Xano before. But I really miss a Xano node offering automatical table and field selection features from the OpenAPI Xano is offering…
Regards Kai
Thank you for sharing your experience. I was thinking of doing this in the same way. So within N8N, you send the question through the endpoint and return the results as just an array of records and provide that (if needed) to the Ai Node/Agent?
yes, this is the way I did it and it works really fine. I also had contact with the support of Cano. They told me that there are no limits in case of rows of the vector data sets. so I really trust them because it’s based on pgvector like supabase…