Interactions API Google

Questions
1

Support for Google Gemini “Deep Research Max” model (Interactions API) in AI Agent Node

Describe the problem/error/question

Hi everyone,
An attempt to use the new Google Gemini model deep-research-max-preview-04-2026 within the AI Agent / Google Gemini Chat Model node has encountered an issue.
The Error:
[GoogleGenerativeAI Error]: Error fetching from … [400 Bad Request] This model only supports Interactions API.

Share the output returned by the last node

[GoogleGenerativeAI Error]: Error fetching from https://generativelanguage.googleapis.com/v1beta/models/deep-research-max-preview-04-2026:streamGenerateContent?alt=sse: [400 Bad Request] This model only supports Interactions API.

Information on your n8n setup

  • n8n version: 2.17.7
  • Running n8n via (Docker, npm, n8n cloud, desktop app): Portainer
3

Hi, @Sven_B :waving_hand:

This is likely not your configuration issue.

The model deep-research-max-preview-04-2026 appears to be an Interactions API-only model from Google, while the current n8n Google Gemini Chat Model node is calling the standard streamGenerateContent endpoint.

So the node likely does not yet support Gemini Deep Research - Interactions API models.

You may need:

  1. Standard Gemini models for now
  2. Use HTTP Request node directly against Google Interactions API
4

Hi @Haian_Abou-Karam

thank you very much for your reply.

Does n8n still don’t support the Google Interactions API? That’s really too bad. Because I specifically want to use the new Gemini model in my workflow.

Best regards,
Sven

5

@Sven_B

Just to be clear, my previous reply is based on my current knowledge and isn’t an absolute, final answer! It is entirely possible that I missed a recent update or a hidden feature.

It might be a good idea to wait a little bit and see if another community member or someone from the n8n team chimes in with different information or a timeline for an official update.

If it turns out that native support definitely isn’t there yet, we can just apply the workaround:

You can bypass the AI Agent node entirely and use n8n’s standard HTTP Request node to communicate with the Interactions API directly.

Here is how you can set up that HTTP Request node to get your research tasks running in the meantime:

{
  "input": "Your overarching research objective here",
  "agent": "deep-research-max-preview-04-2026",
  "background": true
}
6

Are you using any tool use? If so, change to other AI models as Gemini do not support tool use reliably

7

You may need to adjust fields depending on Google’s latest API requirements and authentication method.

8

Hi @kjooleng, :waving_hand:

Thanks for the input!

Just a quick clarification: Gemini Deep Research Max is an autonomous agent built specifically for complex tool-use tasks. It manages its own research tools internally via the Interactions API, making it much more reliable for this specific workflow than standard models.

9

Hi @Haian_Abou-Karam

Thank you for the heads up.

This is the typical tool use error I normally see from Gemini

10

I totally hear you, @kjooleng : Gemini’s tool-calling can definitely be hit-or-miss in standard chat nodes!

However, I believe this specific 400 Bad Request is actually a structural protocol error rather than a performance issue.

It’s essentially a ‘wrong address’ error:

Google requires this specific model to use the Interactions API because it’s a long-running autonomous agent that needs to run in the background.

11

Hi @kjooleng, @Haian_Abou-Karam

Thanks for your replies. It works for me with the HTTP Request Node. However, this is only a workaround for me. It would be much more sophisticated if n8n supported the interaction API for this purpose, among others.

Best regards,

Sven

12

:grinning_face: I’m really glad to hear the HTTP Request node got you up and running, @Sven_B

I completely agree with you: while the workaround functions, a native integration for the Interactions API would be a much more elegant and n8n-style way to handle these agentic models.

Since you’ve already figured out the API logic, this could actually be a fantastic opportunity to build and share a Community Node! It’s a great way to contribute to the ecosystem while we wait for official support.

To help others who might run into this same error, would you mind marking the workaround as the Solution? It keeps the community tidy and helps the next person find the answer faster.

Happy automating!