Generate tiny playable browser games from one prompt using Pydantic AI agents and GLM-5.2 via Nebius Token Factory. The app is a local FastAPI project with a Spotify-style single-page UI. It streams agent progress while a game is being generated, previews the finished game in a sandboxed iframe, and saves successful games as standalone HTML files you can reopen later.

Project path: advance_ai_agents/pydantic_game_agent

Features

One-prompt game generation with named Pydantic AI agents.
Live run stream in the preview area while agents work.
Sandboxed iframe preview for the generated game.
Local generated-game library with one-click open links.
Safety checks for blocked browser APIs and incomplete game output.
Saved games exposed via GET /play/latest, GET /games/{game_id}, and GET /api/games (generated-games/ is gitignored local output).

Tech stack

Python 3.11+
FastAPI: web UI and SSE streaming routes
Pydantic AI: typed multi-agent workflow
GLM-5.2 / Nebius Token Factory: OpenAI-compatible game generation

How it works

The workflow uses Pydantic AI Agent objects with static instructions and typed output_type models where structured data is needed.

Game Designer Agent creates a typed GameSpec from the prompt.
Game Builder Agent generates a standalone HTML/CSS/JS game.
Game Reviewer Agent reviews the generated HTML text plus the spec.
Repair Agent runs one targeted repair pass only if review or static safety checks fail.

The reviewer reads the generated HTML/CSS/JS and the typed spec, then returns a structured review with approval, issues, and fix instructions.

Getting started

Prerequisites

Python 3.11+
Nebius Token Factory API key

Install

git clone https://github.com/Arindam200/awesome-ai-apps.git
cd awesome-ai-apps/advance_ai_agents/pydantic_game_agent
python3 -m venv .venv
source .venv/bin/activate
pip install -e ".[dev]"
cp .env.example .env

Set your Nebius key in .env:

NEBIUS_API_KEY=your-nebius-token-factory-key
NEBIUS_MODEL=zai-org/GLM-5.2
NEBIUS_BASE_URL=https://api.tokenfactory.nebius.com/v1/

NEBIUS_MODEL defaults to zai-org/GLM-5.2. The demo does not require token or temperature tuning.

Run

uvicorn app.main:app --host 127.0.0.1 --port 8000

Open http://127.0.0.1:8000.

Test

pytest

Safety

The app rejects or asks the model to repair generated HTML that contains external script tags, network calls, page navigation, browser storage APIs, or embedded frames. The preview iframe uses sandbox="allow-scripts" so generated games run JavaScript without full page privileges.

Troubleshooting

Missing API key: set NEBIUS_API_KEY in .env and restart the server.

Wrong model: use NEBIUS_MODEL=zai-org/GLM-5.2 without unquoted spaces.

Empty model response: GLM-5.2 can return reasoning before final content; try a shorter prompt or run generation again.

Project structure

pydantic_game_agent/
  app/
    agents.py       # Pydantic AI agents and Nebius model setup
    main.py         # FastAPI routes and SSE stream
    models.py       # Pydantic data models
    safety.py       # Generated HTML safety checks
    storage.py      # Local generated-game persistence
    static/         # CSS and logos
    templates/      # FastAPI/Jinja HTML UI
  tests/            # Route, safety, and storage tests
  .env.example
  pyproject.toml

README.md

Pydantic Game Agent

One-prompt browser game studio