pydantic-ai

Build typed AI agents by defining an Agent() with a model string, deps_type, and output_type, then using @agent.tool decorators to convert Python functions into structured, LLM-callable tools with typed dependencies.

RunContext dependency injection provides typed dependencies to your AI agent tools at runtime. You define a typed deps container dataclass, and RunContext[Deps] passes it securely to tool functions, enabling stateful, structured workflows without boilerplate.

Yes, you can use Google Gemini models with Python AI agents. This approach supports Google Gemini provider integration, allowing you to specify the model string in your Agent() definition for cross-provider AI workflows.

Expose an AI agent over HTTP using AG-UI integration. This feature provides HTTP access to your typed AI agents, allowing web applications to interact with your structured Python tools and dependency-injected workflows.

Create dynamic instructions for an LLM agent using the @agent.instructions decorator. This allows you to generate instructions dynamically via RunContext, adapting the agent's behavior based on runtime dependencies and state.

You need Pydantic AI to define structured tools for LLMs. It requires an Agent() definition with typed dependencies and output types, ensuring safe, structured workflows by leveraging typed dependencies and tool decorators rather than untyped function calls.