PydanticAI

Type-safe agents with OpenAIChatModel

PydanticAI

PydanticAI supports custom OpenAI-compatible providers. Construct an OpenAIChatModel explicitly so requests use Radium’s Chat Completions endpoint.

Before you start

You need Python 3.10 or newer, a Radium API key, and a terminal or PowerShell window.

1. Create a project and install PydanticAI

macOS or Linux

mkdir radium-pydanticai && cd radium-pydanticai
python3 -m venv .venv
source .venv/bin/activate
python -m pip install "pydantic-ai-slim[openai]"

2. Set your Radium credentials

macOS or Linux

export RADIUM_API_KEY="your-radium-api-key"
export RADIUM_MODEL="hal-1.0"

3. Create main.py

import os
from pydantic_ai import Agent
from pydantic_ai.models.openai import OpenAIChatModel
from pydantic_ai.providers.openai import OpenAIProvider

api_key = os.getenv("RADIUM_API_KEY")
if not api_key:
    raise SystemExit("Set RADIUM_API_KEY before running this program.")

model = OpenAIChatModel(
    os.getenv("RADIUM_MODEL", "hal-1.0"),
    provider=OpenAIProvider(
        api_key=api_key,
        base_url="https://api.radium.cloud/v1",
    ),
)

agent = Agent(
    model,
    system_prompt="Follow the user's instruction exactly.",
    model_settings={"max_tokens": 512, "temperature": 0},
)

result = agent.run_sync("Reply with exactly: Radium connected.")
if not result.output or not str(result.output).strip():
    raise RuntimeError("Radium returned no visible text.")
print(f"RADIUM_RESPONSE: {str(result.output).strip()}")

Do not use the shorthand openai:hal-1.0. Constructing OpenAIChatModel explicitly ensures PydanticAI uses the custom Radium provider and Chat Completions endpoint.

4. Run it

python main.py

Expected output:

RADIUM_RESPONSE: Radium connected.

Choose a model

export RADIUM_MODEL="tycho-1.0"  # or hal-1.0 or clarke-1.0
python main.py

Migrate an existing PydanticAI agent

Replace its model/provider configuration:

model = OpenAIChatModel(
    "hal-1.0",
    provider=OpenAIProvider(
        api_key=os.environ["RADIUM_API_KEY"],
        base_url="https://api.radium.cloud/v1",
    ),
)
agent = Agent(model)

Your system prompts, dependency types, tools, and output types can remain unchanged.

Troubleshooting

Problem	Fix
Authentication errors	Set the key in the terminal running Python
Calls going to OpenAI instead of Radium	Use the explicit OpenAIProvider construction shown above
404	Verify /v1 and the exact model name
Empty output	Keep the output allowance at 512 or higher

Validation

These instructions were verified with Python 3.12.7 and pydantic-ai-slim[openai]==2.31.1 on August 20, 2026. You do not need that exact package version.

Text generation and tool_plain execution passed with hal-1.0, clarke-1.0, and tycho-1.0.

Next steps

API quickstart

Call Radium directly

Tool calling

The tool_use and tool_result contract