The interactions interface is incompatible with the Vertex AI Express mode (API Key authentication).

,

The interactions (NextGen) interface is incompatible with the Vertex AI Express mode (API Key authentication).

When using Express mode, the client initializer enforces a mutual exclusion between the API key and the project/location parameters. Because the project and location are missing, the SDK fails to construct the required Vertex AI resource path (/v1beta1/projects/{project}/locations/{location}/interactions). Instead, it falls back to a default path (/v1beta1/interactions) on the Gemini Developer API server, which does not exist and results in a 404 Not Found error.

To use the interactions interface, you must use the standard GCP Credentials authentication mode by providing a project, location, and valid Google Cloud credentials.

good catch, this looks like a real gap between express mode and the interactions endpoint. worth filing it as a bug on the vertex ai sdk repo so its tracked: Issues · googleapis/python-aiplatform · GitHub. for now the workaround you found, standard gcp credentials with project and location, is the right move until express mode support is added.

Hi H9per,

Apologies for the slow response, I was double checking this internally. The answer right now is that Vertex doesn’t fully support the Interactions API, but that should change soon.

Any document about supported interactions via API for now ?