The idea is:
Create a generic embeddings node that can be set up with any http api. You can create embeddings manually with an http request node, but an http request cannot be linked to a vector store node because the node type is not classified as “embeddings” type. This limits the number of APIs you can use for vector retrieval.
My use case:
There are lots of embeddings API providers out there (i.e. Jina) that that do not publish n8n embedding node wrappers. I want to use some of these APIs for vector retrieval tasks, but cannot create a custom embedding node type that is capable of interfacing with the vector store nodes. This means I have to recreate the vector store nodes as a series of code blocks and https requests, which is fiddly and time consuming. If there was simply a generic embeddings node I could point to an arbitrary API endpoint, that would unlock a lot of providers that are currently unsupported.
I think it would be beneficial to add this because:
The tool is most useful when it can access the widest range of API providers for AI and embedding tasks.
Any resources to support this?
Are you willing to work on this?
Sure!