I love Kysely! I use it for a couple of production applications and it works quite well. Type support is better than Knex, and Kysely is really lightweight so I can use it in my projects without being concerned about performance issues.
Really, when you look at options like this, you start to break them down into 3 distinct categories:
1. Raw adapter usage - writing SQL. Performant, but can get tedious at scale, and weird to add types to.
2. Knex/Kysely, lightweight query builders. Readable, composable, and support types well, but a step removed from the performance of (1). Some would argue (1) is more universally understandable, too, although query builders make things easy for those more-familiar with programming languages than query languages.
3. Full-fledged ORMs like TypeORM, Sequelize, Prisma. Often require much more ecosystem buy-in and come with their own performance issues and problems.
I usually choose 1/2 depending on the project to keep things simple.
I have pretty much had no issue with it so far. The only thing that I would call out is that you _must_ run a migration initially to set things up, or your queries will hang. This has stumped me a few times (despite being obvious after-the-fact). It also interfaces really well with postgres, and has nice support for certain features (like jsonb).
I made a tool that generates Typescript types out of a live Postgres database. I've had a request for Kysely support (https://github.com/kristiandupont/kanel/issues/273), but I more or less forgot about it. I would love to hear if that would be helpful.