Search Got a Lot Faster — v1.0.2 & v1.0.3 Update
Two quick patches shipped this week. Both focus on the same thing: making search actually feel instant.
What was wrong
After the v1.0.1 launch, I ran into a frustrating issue. Search worked fine the first time, but repeat searches would slow down or return nothing. Worse, if indexing was running in the background, search could freeze for 5+ seconds. Not acceptable for a tool you reach for dozens of times a day.
v1.0.2 — Search Engine Overhaul
The entire search pipeline got rewritten.
The root cause was a SQLite query structure issue — compound SELECT statements with UNION ALL were not respecting ORDER BY and LIMIT correctly. Fixing that alone brought folder search results back from zero.
Beyond the SQL fix, I separated the cancellation logic so that quick search (filename matching) and full search (content matching) don't interfere with each other. Search results now survive tab switches and language changes instead of disappearing. And a dedicated read-only connection pool means search queries no longer compete with indexing writes for database access.
The Data Setup page also loads noticeably faster — all the status checks that used to run one after another now fire in parallel.
v1.0.3 — Speed, Then More Speed
v1.0.3 followed a day later with six targeted fixes.
The biggest win was the push batch system. During large file scans, the app was flooding the UI with thousands of individual status updates per second. Now those updates get buffered and flushed every 200ms. Search response during active indexing went from 3-4 minutes to under 1 second.
Other changes: filename search now uses SQLite FTS5 instead of LIKE queries (much faster on 60k+ files), search phases run in parallel instead of sequentially, and repeated queries within 30 seconds return cached results instantly.
The numbers
| Scenario | Before | After |
|---|---|---|
| Search during indexing | 5+ seconds | < 1 second |
| Repeated same query | Full re-query | Instant (cached) |
| Filename lookup | Sequential LIKE scan | FTS5 index match |
What's next
Search quality is the foundation everything else depends on — MCP integration, email search, all of it. These two patches close out the search reliability work. Next up is building the test infrastructure to make sure it stays reliable.
If you have been holding off on trying LocalSynapse, now is a good time. Grab the latest from GitHub Releases.