Less waiting on big libraries
Notes, focused Graph, search rebuilds, and exports got faster this week. Here are the numbers, and where nothing changed.
Arivu was doing the same database lookups over and over. This week we stopped, and a few things got noticeably quicker.
Nothing about the results changed: same ordering, same limits, same data. It shipped in Arivu 1.7.0.
What got faster
We tested on a copy of a library with 5,000 notes and bookmarks, plus links and feedback, and timed each request 25 times:
| Request | Before | After | Faster by |
|---|---|---|---|
| Notes list | 169 ms | 75 ms | 55% |
| Focused Graph | 13.2 ms | 3.1 ms | 77% |
Rebuilding search and exporting are where big libraries felt it most. On a test library of 10,000 bookmarks with linked notes:
| Task | Before | After |
|---|---|---|
| Rebuild search | 54.7 s | 0.5 s |
| Full export | 28.7 s | 2.6 s |
We re-ran both on the final build and got about the same speed, so the gains held up.
Where nothing changed
Search and the recent Graph did not get faster in our tests. Don’t read these numbers as “Arivu is faster everywhere.”
The test libraries contain short text. Long documents, a slower disk, or several people using the app at once will give you different numbers. Nor do these numbers cover how long an AI provider takes to answer. If capacity matters to you, measure your own setup.
If you want to check our work, the testing notes in the open-source repo have the setup and the commands.