Beyond the free credits, I see support for Bedrock and Azure, but GCP Vertex AI (now Agent Studio) support is not listed. Is this listed on the near term roadmap?
Currently I can use Service Accounts against my GCP project and pull in models via the AI Agent node, but for this new feature, the only credential that is supported is a Gemini API key, which my organization blocks.
Community members can’t commit to the roadmap here — the Gateway credits catalogue is n8n’s call, and it does grow over time (the current list lives on the service-pricing page).
Practically, you’re not blocked: the Gateway-credits choice is made per node, so you can keep using your own Google service-account credential on the AI Agent node for Vertex models, and use Gateway credits only for the providers it does cover. Anything outside the catalogue still works the normal way with your own credential. So unless you specifically need to spend the free credits on Vertex calls, your existing service-account setup is the path — and if it is the credits you need, a proxy in front of Vertex won’t draw from the Gateway balance, so a feature request is the right place to ask.
Hi @datanerdery3, the new standalone Agents builder uses a Gemini API-key credential for Google. It doesn’t expose the Vertex service-account option you use with the workflow AI Agent node, so this isn’t just a Gateway credits limitation.
I couldn’t find an announced date for Vertex support. I’d raise the service-account requirement in the Agents launch thread, where the team is collecting feedback:
Hey @datanerdery3
You can still get your Vertex models into an agent today through the OpenAI provider. Agents pass the OpenAI credential’s Base URL straight through, so run an OpenAI-compatible proxy like LiteLLM with your service account in front of Vertex, somewhere your n8n instance can reach, and point an OpenAI credential at it. For a custom endpoint the model picker lists whatever the proxy’s /models returns, so your Gemini models show up by name, and the agent falls back to /chat/completions if the proxy doesn’t speak the Responses API. No Gemini API key is involved anywhere.
I can help bridge this gap by designing an n8n workflow that works with your organization’s existing GCP authentication instead of depending on a blocked Gemini API key. The goal is to keep your Vertex AI models accessible while fitting your current security and credential requirements.
Review the current AI Agent setup
Map GCP service-account authentication
Connect Vertex AI models
Adapt the new feature workflow
Test authentication and model execution
Hi @datanerdery3 ,
One thing to watch if you go the proxy route: Vertex access tokens expire after about an hour, so pointing the OpenAI credential straight at Vertex’s OpenAI-compatible endpoint will break unless something refreshes the token. A proxy like LiteLLM handles the service-account auth itself, so that’s the safer setup.
And if you only need Vertex for a few flows, the AI Agent node with your service-account credential keeps working as-is until the Agents builder gets Google service-account auth.
@datanerdery3 If the LiteLLM route works for you, mark it as the solution so others with the same org restriction can find it.
I am not asking to modify the roadmap, I am asking with proper GCP Project-based support will be added. I can’t use a Gemini token and my question was not about the Gateway credits. I can use the old AI Agent node easily, and likely will have to fall back to that for my class that starts in two weeks.
Thank you, I already did that. I appreciate the note.
Thanks but this wont work for the my use case. I work for an eductational institution and my primary use case is for a class that starts in two weeks. I can use the “old” agent node very easily, and the setup is not a barrier for students, but introducing another tool and config/setup/app that for this (e.g. LiteLLM) is introducing hurdles that I can’t add for a 7 week course. I appreciate the note.
That makes sense, and for a 7-week course I wouldn’t add a proxy either. The workflow AI Agent node with the Google Vertex Chat Model and your service-account credential keeps students on a single tool, so I’d stay on that for this class.
If you want Vertex support in the new Agents builder later, the Agents launch thread Anshul linked is where the team collects feedback, so the service-account requirement is worth posting there.
If that covers it, feel free to mark whichever reply helped most as the solution. HomelabAgent’s reply describes exactly this setup.
