Weather

Current conditions and forecast, answered as a plain fact rather than performed in character — so the numbers stay exact.

How to use it

"what's the weather"

"is it going to rain tomorrow"

"how cold is it tonight"

"what's the forecast"

How it works

• Power response (brain/api.py, _power_response → _WEATHER_RE): weather questions are caught *before* any LLM call. when is derived from simple substring checks ("tomorrow" / "today"/"tonight" / default "now") and handed to weather_v2.get_weather().
• Tool fallback (brain/tools.py, get_weather): if a weather question reaches the tool-use round-trip some other way (e.g. embedded in a longer request), the LLM can still call get_weather(when, location) directly. location defaults to home; common fillers like "home"/"here"/"outside" are normalized to "no location given" since the geocoder can't resolve them.
• Data source: weather_v2.py — Open-Meteo, 16-day forecast, Bay Area locations.

Status & known limits

• If weather_v2 returns an error string (starts with "sorry"), the power response falls through to the model rather than voice an error — so an occasional weather answer *does* go through character narration.
• Location support beyond "home" is geocoded per-request; no saved list of named locations yet.