Problem or use case
Text search on SQLite on omnigent reads every message on every search and uses a LIKE query so it gets slower as history grows.
On Postgres this is already fixed by using a trigram index (#4502), SQLite can have support for this too.
Proposed solution
Add an FTS5 trigram index on SQLite 3.34+.
Alternatives considered
The existing conversation_items_fts table uses a word tokenizer, so it can't match substrings (andbox in sandbox).
Problem or use case
Text search on SQLite on omnigent reads every message on every search and uses a LIKE query so it gets slower as history grows.
On Postgres this is already fixed by using a trigram index (#4502), SQLite can have support for this too.
Proposed solution
Add an FTS5 trigram index on SQLite 3.34+.
Alternatives considered
The existing
conversation_items_ftstable uses a word tokenizer, so it can't match substrings (andboxinsandbox).