Takes effect on the next question — no restart. Effort changes how much the model thinks rather than the price per token, so its cost effect varies; the model changes the rate directly. Switching model resets the prompt cache, so the first answer after a change costs a little more.

Every figure the assistant gives is computed by the same functions that draw these pages, so the two can never disagree. It will recommend actions your own data supports, but not specific investments — and anything consequential on tax is worth confirming with your accountant.

Kept as a way in if Google is unreachable or misconfigured. Setting one here stores a scrypt hash and stops APP_PASSWORD being read, which takes the plaintext out of .env.