The Google Gen AI SDK is the Python client for Google's generative models. One client targets either the Gemini Developer API (with an API key) or the Gemini Enterprise Agent Platform, formerly Vertex AI (with a Google Cloud project and location), with sync and async variants, typed request and response objects, and a choice of beta or stable API versions.
The SDK covers text and multimodal generation, streaming, chats, embeddings, token counting, file uploads, and context caching. It also wraps Imagen for image generation, upscaling, and editing, and Veo for text-to-video, image-to-video, and video-to-video generation.
For agent work, it supports function calling (including automatic calling of plain Python functions), structured output through JSON Schema or Pydantic models, and experimental Model Context Protocol (MCP) support that passes an MCP server to the model as a tool. The Interactions API adds server-side conversation state, built-in Google Search grounding and code execution, and specialized agents such as Deep Research. Supervised fine-tuning and batch prediction are available on the Agent Platform only.
Google publishes Gemini API agent skills for coding assistants such as Claude Code and Cursor so that they generate code against current SDK syntax.
Features
- Two backends: one client for the Gemini Developer API and the Gemini Enterprise Agent Platform
- Content generation: text, image, and multimodal generation with sync, async, and streaming calls
- Function calling: automatic invocation of Python functions or manually declared function schemas
- Structured output: responses constrained by JSON Schema or Pydantic models
- MCP support: pass an MCP server as a tool to the model (experimental)
- Interactions API: stateful conversations, Google Search grounding, code execution, and agents such as Deep Research
- Imagen and Veo: generate, upscale, and edit images, and generate videos from text, images, or video
- Embeddings and tokens: embed content, and count or compute tokens remotely or locally
- Files and caches: upload files and reuse cached context across requests
- Tuning and batch jobs: supervised fine-tuning and batch prediction on the Agent Platform
