Hermes Weather Plugin

Weather plugin for Hermes Agent. 12 tools covering current conditions, forecasts, alerts, model imagery, radar, and meteorological calculations.

Data tools call NWS/SPC/METAR APIs directly in Python. All image rendering happens in Rust -- no matplotlib in the rendering path.

Tools

Data (Python)

Tool	What it returns
wx_conditions	Current obs: temperature, wind, sky, dewpoint
wx_forecast	NWS 7-day or hourly forecast
wx_alerts	Active warnings, watches, advisories
wx_metar	Raw/decoded METAR for any ICAO station
wx_brief	Conditions + forecast + alert count in one call
wx_global	Non-US locations via Open-Meteo
wx_severe	SPC Day 1 categorical outlook + active watches

Images (Rust)

Tool	What it returns
wx_model_image	NWP field rendered as PNG -- 22+ products, 3 models, batch support
wx_radar_image	NEXRAD Level 2 reflectivity PPI (1024px, 200km range, 10 dBZ noise floor, dark background)
wx_storm_image	Radar with storm cell analysis overlaid

Calculations (Rust via PyO3)

Tool	What it returns
wx_calc	Any of 205 meteorological functions (dewpoint, CAPE, LCL, wind chill, etc.)
wx_sounding	Model sounding at a point -- 40 pressure levels + all derived parameters

Model Images

22+ fields rendered with color tables from Solarpower07. Lambert Conformal projection, state/country borders, colorbars.

Instability: CAPE (surface, mixed-layer, most-unstable, 0-3km), CIN Shear/Helicity: SRH 0-1km, SRH 0-3km, updraft helicity, 0-6km bulk shear, 0-1km bulk shear Surface: Temperature, dewpoint, RH, wind gust, cloud cover, precipitation, visibility Composites: STP, SCP, EHI Reflectivity: Composite reflectivity (27-color discrete palette)

Comma-separated batch: "cape,srh,uh,stp" renders 4 images in one tool call.

Average render time: 177ms per image.

Supported Models

Model	Resolution	Frequency	Forecast Range
HRRR	3 km	Hourly	0-48h
NAM	12 km	6-hourly	0-84h
RAP	13 km	Hourly	0-51h

Sounding Parameters

wx_sounding downloads pressure-level data and computes:

• CAPE/CIN: Surface-based, mixed-layer, most-unstable
• Levels: LCL (pressure, temperature, height AGL), LFC, EL
• Indices: Lifted Index, K-Index, Total Totals
• Shear: 0-1km and 0-6km bulk shear, SRH 0-1km and 0-3km
• Storm motion: Bunkers right-mover
• Composites: STP (fixed-layer)
• Moisture: Precipitable water, freezing level
• Lapse rates: 0-3km, 700-500mb
• Profile: Standard levels (1000, 925, 850, 700, 500, 300, 250 mb) with T, Td, wind

Stack

Plugin (Python)
  ├── Data: requests → NWS / SPC / METAR / Open-Meteo APIs
  ├── Model images: rusbie → cfrust → wrf-render
  │                  (download)  (decode)  (rasterize)
  ├── Radar: radar-render binary (rustdar)
  └── Calculations: metrust-py (205 functions, PyO3 → Rust)

No eccodes, no Fortran, no C libraries in the Rust components. The only system dependency is a working Python environment.

Python Packages

metrust      — 205 meteorological calculations (PyO3 → Rust)
cfrust       — GRIB2 decoder (pure Rust, replaces cfgrib/eccodes)
rusbie       — NWP downloader with byte-range .idx filtering
rustweather  — Plotting wrapper
rustplots    — MetPy-compatible plotting
wrf-rust     — Solarpower07 color tables + rasterizer

Rust Binary

radar-render — NEXRAD Level 2 download + parse + render (from rustdar)

Setup

# Install Python packages
pip install metrust cfrust rusbie rustweather

# Install from source (not yet on PyPI)
pip install -e /path/to/rustplots
pip install -e /path/to/wrf-rust

# Build radar binary
cd /path/to/rustdar
cargo build --release --bin radar-render

# Copy plugin to Hermes
cp -r weather ~/.hermes/plugins/

# (Optional) Set radar binary path if not at ~/rustdar/
export RADAR_RENDER_PATH=/path/to/radar-render

Timings

Operation	Time
Model image render (Rust)	~177ms
22 maps including download	~18s
15 maps from cache	~3.1s
Radar image (NEXRAD L2)	~3s (download + render)
Sounding (40 levels + params)	~15s (download-heavy)
METAR lookup	~300ms

File Structure

~/.hermes/plugins/weather/
├── plugin.yaml          # Hermes plugin manifest
├── __init__.py          # register(ctx) — wires 12 tools
├── schemas.py           # Tool schemas (what the LLM sees)
├── nws.py               # NWS/METAR/SPC/Open-Meteo API client
├── skill.md             # Usage guide for the LLM
├── tools/
│   ├── __init__.py
│   ├── data.py          # NWS API handlers
│   ├── images.py        # Rust renderer + radar handlers
│   └── calc.py          # metrust calculations + sounding
└── README.md

Credits

• Color tables: Solarpower07 -- discrete color palettes and product style definitions used for all model imagery
• Meteorological calculations: metrust -- 205 functions verified against MetPy test suites
• Plugin platform: Hermes Agent by Nous Research