AgentPlayground

A quick playground for AI agents development.

This project is currently a minimal Google ADK-based "weather agent" example.

Layout

• agent_playground/ — package code (agent factory, tools, runtime, conversation loop)
• agent_playground/guardrails/ — model/tool guardrails (callbacks for validation)
• agent_playground/warnings_suppression.py — optional suppression of noisy Pydantic serializer warnings
• __main__.py — root shim entrypoint (kept for convenience)
• tools.py — compatibility shim (re-exports agent_playground.tools)

Requirements

• Python 3.11+
• A configured environment with Google ADK + dependencies installed

Setup

Pick one of the following:

Option A: Conda (recommended)

conda env create -f environment.yml
conda activate agent-playground
python -m agent_playground

Option B: venv + pip

python -m venv .venv
.\.venv\Scripts\Activate.ps1
python -m pip install -r requirements.txt
python -m agent_playground

Environment variables

• OPENAI_MODEL (required): model name used by LiteLlm (example: gpt-4o-mini)
• OPENAI_API_KEY (may be required depending on your LiteLLM/provider setup)
• AGENT_PLAYGROUND_SUPPRESS_PYDANTIC_SERIALIZER_WARNINGS (default: true): suppresses noisy Pydantic serializer warnings

Tip: you can set these in PowerShell for the current session:

$env:OPENAI_MODEL = "gpt-4o-mini"
$env:OPENAI_API_KEY = "..."

Alternatively, copy .env.example to .env and fill in your values.

Guardrails

This repo includes simple guardrails implemented using ADK callbacks:

• agent_playground/guardrails/model_callback.py: before_model_callback validations
• agent_playground/guardrails/tool_callback.py: before_tool_callback validations

The weather_agent has these guardrails enabled (see before_model_callback and before_tool_callback in the agent factory).

Pydantic serializer warning suppression

agent_playground/warnings_suppression.py implements check_pydantic_warnings_suppression(), which reads AGENT_PLAYGROUND_SUPPRESS_PYDANTIC_SERIALIZER_WARNINGS from .env/environment and suppresses only the noisy Pydantic serializer warnings.

This suppression is called during startup in agent_playground/__main__.py.

Why it exists: in some runs we observed the warning spam could make agent execution appear “stuck” (logs flood / noisy stdout), even though it’s not a fatal error.

Run

Recommended:

python -m agent_playground

Alternative (kept for backwards compatibility):

python __main__.py

Once running, type messages into the terminal. Use exit or quit to stop.

Run with ADK Web UI

The ADK CLI discovers agents by scanning subfolders for an agent.py that exports a top-level root_agent. This repo provides that entrypoint at agent_playground/agent.py.

From the repo root (the parent folder that contains agent_playground/), run:

adk web --port 8000

Then open http://localhost:8000 and select agent_playground.

Tutorial sources

This project was created using these tutorials as references:

• https://codelabs.developers.google.com/onramp/instructions
• https://google.github.io/adk-docs/get-started/python/
• https://google.github.io/adk-docs/tutorials/agent-team/
• https://colab.research.google.com/drive/10aC9vrBD8y_UlR9CcmuXuvBBPnkZ8i7M