Fixed
- Search finds an article under its other names again: "العائلة اللغوية" leads to "أسرة لغات", as in Kiwix (#86). Title indexes rebuild in the background after the update; search keeps working while they do.
- Quick search answers in about a tenth of a second on a large library, including for words not searched recently. It read hundreds of rows per ZIM from disk; it now reads the title index alone, and finds matches it used to miss.
- Making a subreddit into a ZIM works again: SQLAlchemy 2.1, released 2026-09-24, broke the Reddit maker's build. It is pinned below 2.1, and a maker installed with 2.1 is reinstalled the next time it runs online.
- A search made in the first seconds after Zimi starts, before its title indexes are ready, no longer keeps its empty answer for 15 minutes.
- Manage → Server shows the background work still running under the cache sizes: title indexes, the Q-ID scan, did-you-mean and ZimiTube descriptions, each with what it is on and how far along, refreshing until it is done; then it says nothing is building.
- Create a ZIM: a site capture's Max pages takes any number and its size budget any size, and 0 means no limit for either. They used to be cut to 50,000 pages and 64 GB without a word.
Upgrade
docker pull epheterson/zimi:latest
docker compose down && docker compose up -dOr pip install --upgrade zimi for Python installs.