Changelog

All notable changes to Generous.nz are recorded here, grouped by version. Every deployed build has a version number, a git tag, and an entry below.

Versioning scheme

Versions follow MAJOR.MINOR.PATCH:

0.x — pre-launch. Minor bumps mark a completed build phase or major feature set; patches are bug fixes on the current build.
1.0.0 — first public launch.
1.x+ — post-launch: minor = new features, patch = fixes, major = breaking changes or significant re-architecture.

Each version is tagged in git (v0.2.0) and the current version is kept in package.json. See docs/version-control.md for how to cut a release.

Admin-managed DeepSeek API key.

The admin AI page now has a "Provider key" panel where the DeepSeek API key (and optional model name) can be entered and saved — no server .env edit or restart needed.
The key is encrypted at rest (AES-256-GCM, keyed from AUTH_SECRET) in the settings table. The full key is never returned to the browser; the page only shows a masked sk-…abcd status.
Saved keys are used immediately by AI moderation, fraud screening, order fraud checks, content generation, and the "Run test review" button. A "Remove saved key" button clears it and falls back to the server .env key.

The AI provider now reads its DeepSeek key from the database first and the server .env as a fallback.

A valid sk-… key from platform.deepseek.com is still required for the AI provider to authenticate; the previous server key returned 401.

AI moderation model fix.

The DeepSeek model name was set to a non-existent model; it now defaults to the real deepseek-chat model, and the admin AI page hint matches.

The AI provider is currently returning "401 Authentication Fails" — the DeepSeek key in the server .env needs to be a valid sk-… key from platform.deepseek.com. Once a valid key is in place, product moderation runs fully automatically: approve when the AI is confident, flag with a reason otherwise.

Fully automatic AI moderation for products.

Wholesale items now auto-approve when the AI review is confident they're fine (previously they always needed a human). Anything the AI isn't sure about is flagged with the AI's reason, and banned items are still auto-rejected.
Editing a product re-runs the AI review — a fixed or resubmitted item is re-scored automatically and approved again if it's fine, or flagged with a reason if not (previously edited items just sat in the queue with no AI check).
The AI product review now considers the campaign/business context (what the fundraiser is for, the business name, category, brand) and is prompted to catch attempts to disguise banned or illegal items with misspellings, euphemisms or vague wording.

Voucher codes moved to the product page + sold codes never reused.

Voucher code management moved out of the catalogue cards and into the product's edit page (where there's room): generate or paste codes there. Creating a new voucher takes you straight to its edit page so you can add codes immediately.
Sold vouchers stay sold: a code that has been purchased is never returned to the pool, even if the order is later refunded — it stays marked as assigned to that order, so it can't be resold.
Voucher sales page now shows the voucher code, the purchaser, and a "🎁 Gifted to [name] ([email])" line when the buyer selected the gift option at checkout.

Vouchers can be bought as gifts.

Voucher purchases now have a "This is a gift" option on the buy button. When selected, the purchaser enters the recipient's name, email and an optional message, and the voucher is emailed to the recipient (with the message) and a copy goes to the purchaser.

Brands page explainer + one-click first brand.

The Brands page now explains what brands are (sell multiple product lines under one store and one payout account) with real use cases, and how to set one up (create → tag products → it appears on the catalogue).
One-click first brand: if the store has no brands, a highlighted card creates a brand named after the business with a single click.

Removed references to other sites.

Removed all mentions of other platforms (e.g. "XXXXX-style") from user-facing copy and code comments. Feature descriptions now stand on their own.

Zero-downtime deploys.

The server deploy process now builds the site into a separate directory and atomically swaps it in, so the live site keeps serving the current build throughout the update — no more "Internal Server Error" / "client-side exception" windows during releases.
Dependencies are only reinstalled when the lockfile actually changes, so node_modules is no longer wiped on every push.

The intermittent TypeError: a is not a function / module-not-found errors during deploys were the old process serving while npm ci and next build replaced files under it. Deploys no longer disturb the running site.