How to Use the Agno Framework with Gemini Embeddings Without an API Key?

[GRoberson]

Hello, everyone!

I'm trying to use the Agno framework to generate embeddings with Google's Gemini models. However, I noticed that the framework requires an API Key for authentication. The problem is that I'm not using Google's API Key, as I'm authenticating via gcloud auth.

Given this, I have a few questions:

1 - Is there a way to configure Agno to use gcloud auth instead of an API Key for embeddings?
2 - Does the framework support authentication via Application Default Credentials (ADC) for embeddings? If so, how can I properly configure it?
3 - If Agno cannot be used without an API Key, is there an alternative library or method to generate Gemini embeddings using only gcloud auth?
4 - Has anyone successfully integrated Agno with Gemini embeddings without using an API Key? If so, could you share an example?

I appreciate any help!

[mikelee-flex]

You can use Service Accounts if you're running outside of GCP.

1. You'll need to create the service account in the project associated with the workload
2. Enable the Vertex and Gemini APIs for the service account. FYI the Google Docs suck on clearly explaining how to do this, but searching will help you find some resources
3. Create a key file (JSON) for the service account that you can use to impersonate the service account
4. Store the JSON in a secure place (in my case I use Secrets Manager in AWS but you can use whatever your process will have access to)
5. Create a Credentials object from the JSON (code below)
6. Create a GenAI client (code below)
7. Pass the client into the agent's model object (code below)

Create a Credentials object from the JSON

gcp_credentials = json.loads(service_account_secret)

            # Create credentials object
            scopes = ['https://www.googleapis.com/auth/cloud-platform']
            self.credentials = service_account.Credentials.from_service_account_info(
                gcp_credentials,
                scopes=scopes
            )

Create a GenAI client

self.client = genai.Client(
                vertexai=True,
                project="gcp-project-id",
                location="us-central1",  
                credentials=self.credentials
            )

Pass the client into the agent's model object

agent = Agent(
                name="My Agent",
                role="Do stuff.",
                model=Gemini(
                    id="gemini-1.5-pro-002",
                    vertexai=True,
                    project_id="gcp-project-id",
                    location="us-central1",  
                    client=self.client,
                ),

[mikelee-flex]

BTW you'll need to use, and note genai is the latest SDK from Google, the other one referenced in the Agno docs will get deprecated at some point. I'll hopefully have time to create a PR to correct and update later.

from google import genai
from google.oauth2 import service_account