Add vertex AI

[samuelcolvin]

Example Usage

With the default google project already configured in your environment:

from pydantic_ai import Agent
from pydantic_ai.models.vertexai import VertexAIModel

model = VertexAIModel('gemini-1.5-flash')
agent = Agent(model)
result = agent.run_sync('Tell me a joke.')
print(result.data)
#> Did you hear about the toothpaste scandal? They called it Colgate.

Or using a service account JSON file:

from pydantic_ai import Agent
from pydantic_ai.models.vertexai import VertexAIModel

model = VertexAIModel(
    'gemini-1.5-flash',
    service_account_file='path/to/service-account.json',
)
agent = Agent(model)
result = agent.run_sync('Tell me a joke.')
print(result.data)
#> Did you hear about the toothpaste scandal? They called it Colgate.

[cloudflare-workers-and-pages]

Deploying pydantic-ai with    Cloudflare Pages

Latest commit:	73ad697
Status:	✅  Deploy successful!
Preview URL:	https://25c1191d.pydantic-ai.pages.dev
Branch Preview URL:	https://vertexai.pydantic-ai.pages.dev

View logs

[adriangb]

I'm not familiar with the details of the codebase but in general looks good to me!

[adriangb]

I'm not sure this needs to be clarified in the docstring, I think it's okay if that's just how it is.

[adriangb]

I will say that it might be nice if these are forced to be keyword arguments, it'd be easy to mess up and do {}, None, True or something and with None being falsely and whatnot it might fail in surprising ways.

[samuelcolvin]

This is basically an internal API, otherwise I would agree.

[adriangb]

I tend to prefer passing this stuff in as parameters, e.g. have a parameter called _auth_provider: AuthProviderProtocol or something and you pass in a test fake, maybe a mock. You wouldn't be able to do this monkey patching business in Rust 😉

[samuelcolvin]

I'd rather keen the code clear and mock like this.