A Python service that generates natural, kid-friendly weather descriptions using OpenWeatherMap and an LLM API (can be a small, self hosted model).

I wrote a blog post about how I use it as part of a weather eInk display:

The display is driven by a $40 single board computer and shows weather forecasts that can be easily understood.

Transforms complex weather data into simple, engaging descriptions that kids can understand. Instead of "Partly cloudy with 70% chance of precipitation," you get "It might rain later - maybe bring an umbrella! 🌧️"

The service includes:

• Command-line weather reports with HTML rendering
• A shared service layer (WeatherReportService) used by every entrypoint
• API caching for performance
• LLM interaction logging and replay capabilities
• Support for multiple LLM providers with automatic fallback

• Python 3.9+
• uv for dependency management (recommended) or pip
• OpenWeatherMap API key (free tier works fine)
• LLM API key (DeepSeek, OpenAI, OpenRouter, or any OpenAI-compatible API)

1. Install uv if you haven't already:

curl -LsSf https://astral.sh/uv/install.sh | sh

2. Clone this repository

3. Create and activate a virtual environment:

uv venv
source .venv/bin/activate  # On Windows: .venv\Scripts\activate

4. Install dependencies:

uv sync
# OR without uv: pip install .

5. Copy the example environment file and configure your API keys:

cp .env.example .env
# Edit .env with your API keys

Before running these commands, make sure your .env file is properly configured with API keys as described in the Installation section.

Generate a weather report for a specific location:

uv run python -m kidsweather --lat 38.9 --lon -77.0

Load test data from a file (files live in test_data/ by default):

uv run python -m kidsweather --load dc1

Save weather data for testing (writes to test_data/ unless you supply --save with a different path):

uv run python -m kidsweather --lat 38.9 --lon -77.0 --save dc_latest

Log LLM interactions:

uv run python -m kidsweather --lat 38.9 --lon -77.0 --log-interactions

See what the model was told:

uv run python -m kidsweather --lat 38.9 --lon -77.0 --verbose

Render weather report as HTML to a file:

uv run python -m kidsweather --render page.html

This generates a standalone HTML file with weather information, styled for display on e-ink screens or other devices.

Replay logged LLM interactions with different prompts or models:

uv run python replay.py --log-id 5
uv run python replay.py --log-id 5 --new-model deepseek-coder
uv run python replay.py --log-id 5 --prompt "You are a pirate weather forecaster."

{
  "description": "It's cool right now, maybe wear a light jacket. 🧥 The rest of the day looks cloudy but nice!",
  "daily_forecasts": [
    "Monday: Cloudy day ☁️",
    "Tuesday: Rain coming later 🌧️",
    "Wednesday: Still rainy",
    "Thursday: Cloudy again"
  ],
  "temperature": 47,
  "feels_like": 44,
  "conditions": "overcast clouds",
  "high_temp": 59,
  "low_temp": 47,
  "icon_url": "http://openweathermap.org/img/wn/[email protected]",
  "alerts": [],
  "last_updated": "Friday, April 11 at 11:30 AM"
}

1. OpenWeatherMap: Get a free API key at https://openweathermap.org/api
2. LLM Provider: Choose from DeepSeek, OpenAI, OpenRouter, or any OpenAI-compatible API

The project is organized as a Python package with a clear separation of concerns:

• Core (kidsweather/core/): Contains the main service orchestration and settings management
  • settings.py: Loads environment variables into a single dataclass tree and ensures required directories exist
  • service.py: The WeatherReportService orchestrates data fetch, formatting, LLM generation, and logging
• Clients (kidsweather/clients/): External API integration layers
  • weather_client.py: Fetches current conditions and optional historical summaries from OpenWeatherMap, applying diskcache when configured
  • llm_client.py: Wraps the primary and optional fallback LLM providers, normalising JSON responses and caching successful calls
• Formatting (kidsweather/formatting/): Data preparation and output generation
  • weather_formatter.py: Prepares both the LLM prompt context and the data needed for display
  • html.py: Contains render_to_file() function for HTML rendering using Jinja2 templates for e-ink displays
• Infrastructure (kidsweather/infrastructure/): Cross-cutting concerns
  • cache_provider.py: Cache construction and management utilities
  • llm_logging.py: Persists LLM interactions to SQLite for replay and debugging
• Templates (kidsweather/templates/): Jinja2 templates for HTML rendering
• Tests (kidsweather/tests/): Unit and integration tests

The main CLI entry point is kidsweather/__main__.py, which delegates to the Click-based command interface in main.py.

• Uses OpenWeatherMap's One Call API 3.0
• Supports any OpenAI-compatible chat completions API
• Configuration lives in kidsweather/core/settings.py and is shared across every entrypoint
• Temperatures are in Fahrenheit
• Caches API and LLM responses for 10 minutes using diskcache
• Logs LLM interactions to llm_log.sqlite3 for replay
• Includes optional automatic fallback between LLM providers
• HTML rendering uses Jinja2 templates from kidsweather/templates/
• Organized package structure with clear separation of concerns