Skip to content

pydantic_ai.models.gemini

Custom interface to the generativelanguage.googleapis.com API using HTTPX and Pydantic.

The Google SDK for interacting with the generativelanguage.googleapis.com API google-generativeai reads like it was written by a Java developer who thought they knew everything about OOP, spent 30 minutes trying to learn Python, gave up and decided to build the library to prove how horrible Python is. It also doesn't use httpx for HTTP requests, and tries to implement tool calling itself, but doesn't use Pydantic or equivalent for validation.

We could also use the Google Vertex SDK, google-cloud-aiplatform which uses the *-aiplatform.googleapis.com API, but that requires a service account for authentication which is a faff to set up and manage.

Both APIs claim compatibility with OpenAI's API, but that breaks down with even the simplest of requests, hence this custom interface.

Despite these limitations, the Gemini model is actually quite powerful and very fast.

GeminiModelName module-attribute 

GeminiModelName = Literal[
    "gemini-1.5-flash",
    "gemini-1.5-flash-8b",
    "gemini-1.5-pro",
    "gemini-1.0-pro",
]

Named Gemini models.

See the Gemini API docs for a full list.

GeminiModel dataclass

Bases: Model

A model that uses Gemini via generativelanguage.googleapis.com API.

This is implemented from scratch rather than using a dedicated SDK, good API documentation is available here.

Apart from __init__, all methods are private or match those of the base class.

__init__ 

__init__(
    model_name: GeminiModelName,
    *,
    api_key: str | None = None,
    http_client: AsyncClient | None = None,
    url_template: str = "https://generativelanguage.googleapis.com/v1beta/models/{model}:{function}"
)

Initialize a Gemini model.

Parameters:

Name Type Description Default
model_name GeminiModelName

The name of the model to use.

 required
api_key str | None

The API key to use for authentication, if not provided, the GEMINI_API_KEY environment variable will be used if available.

 None
http_client AsyncClient | None

An existing httpx.AsyncClient to use for making HTTP requests.

 None
url_template str

The URL template to use for making requests, you shouldn't need to change this, docs here.

 'https://generativelanguage.googleapis.com/v1beta/models/{model}:{function}'

GeminiAgentModel dataclass

Bases: AgentModel

Implementation of AgentModel for Gemini models.

GeminiStreamTextResponse dataclass

Bases: StreamTextResponse

Implementation of StreamTextResponse for the Gemini model.

GeminiStreamStructuredResponse dataclass

Bases: StreamStructuredResponse

Implementation of StreamStructuredResponse for the Gemini model.

get 

get(*, final: bool = False) -> ModelStructuredResponse

Get the ModelStructuredResponse at this point.

NOTE: It's not clear how the stream of responses should be combined because Gemini seems to always reply with a single response, when returning a structured data.

I'm therefore assuming that each part contains a complete tool call, and not trying to combine data from separate parts.