Installation and setup

Pydantic AI installs as a slim core plus one extra for each model provider you want.

Last updated: 28 Sep, 2026 · Pydantic AI 2.51

pydantic-ai-slim is the framework without any provider libraries. An extra such as [openai] adds one provider's client. The full pydantic-ai package pulls in every provider at once.

Installing the slim core and extras

[openai] adds OpenAI's client and [mcp] the MCP client used later. pydantic-evals and pytest are for the testing part.

pip install "pydantic-ai-slim[openai,mcp]==2.51.0" "pydantic-evals==2.51.0" pytest

Checking the installed version

Print the version so the rest of the course matches what you have:

from importlib.metadata import version

print(version("pydantic-ai-slim"))

2.51.0

The number you see is the version every output in this course was captured against.

What each package does

• pydantic-ai-slim is the framework: agents, tools, output types, retries.
• [openai], [groq], [google] and other extras each add one provider's client.
• Nothing here needs an API key. Lessons up to Real models: providers, keys and model names run on the built-in test model and the function model you write in FunctionModel: write a stand-in model.

Adding a provider and its keyOptional

From Real models: providers, keys and model names an agent can run on a hosted model. Add the provider's extra, set its key in the environment, then name the model when you create the agent:

from pydantic_ai import Agent

agent = Agent("groq:openai/gpt-oss-120b", instructions="You answer support tickets.")

pip install "pydantic-ai-slim[groq]==2.51.0"

export GROQ_API_KEY=gsk_...

Related

• Previous: Pydantic AI overview
• Next: Agent basics and a first run
• Reference: Install Pydantic AI