Hey n8n community,
I am excited to release my first community node: n8n-nodes-vercel-ai-sdk - npm
It comes with some really cool features:
Simple Prompt + Messages Prompt
Multimodal inputs (Audio, video for Gemini)
Support for Search Grounding (Gemini)
Messages injectable as JSON
Easy integration of Deepseek r1 and also Groq Deepseek r1
Would love your feedback 
so cool and useful, shame N8N does not natively support search grounding feature. I guess $55 Millions could help closing the gap…
So what if its the other way around, like if I want to use ollama provider or custom provider to connect to n8n instead of ollama directly? Sort of like a proxy, Can we create a emulator node that mimics ollama so that the workflow is fully utilized from n8n side of chat? Basically the n8n custom node will handle orchestration and calling any tools etc from the workflow of n8n and on Vercel AI side we deal with it as if its an Ollama provider. do you think its possible? That way I can keep the quick n8n workflow for quick AI workflows and use vercel ai sdk for the great UI frontend…
Or could you suggest a better way I couldn’t think of?
This is fantastic - many thanks! I’d been using HTTP nodes for the multimodal stuff, this will make it all so much more straight forward. The search feature pretty much eradicates most of my MCP tool usage as well 
Hi again @felixvemmer - this node really is excellent - thanks again for this. However I’m struggling to get a conversational interface working together with the multimodal file processing with Gemini; I’m wanting the user to be able to keep referring back to the file in the same that way we would using an AI Agent with memory. Do you have some example JSON you could share that demonstrates how to do this?
Many thanks again!
This is a great resources. Thanks for sharing.
@Paul_Vincent this is a tough one 
not sure if this can be easily done with the current setup. Sorry!
No worries Felix; it’s not essential for us, just a quality of life thing really. Is it possible for multiple files to be submitted using the JSON Messages route though? When I submit several together only the first is being analysed, which I’m assuming can be accounted for by using JSON, but the structure I use in HTTP nodes isn’t working here. I’m guessing Vercel uses a slightly different structure to the core API?
Thanks again for the node - I hope n8n integrate similar functionality into their agent / chain Gemini nodes!