Blog

4 min read

Spotify Wrapped popularized a version of the “year in review” format. Now everything is “wrapped”. Why can’t Dolt be wrapped too?

Turns out, Dolt can be wrapped! We’re ripping off Spotify’s gimmick for the second time! It’s “Dolt Wrapped 2025: A Year in Review”. We’ll cover some categories and give you some blog article links just in case you missed them the first time.

Best Feature#

In 2025, we saw a lot of Dolt feature launches. To name a few in no particular order:

But my favorite is automatic garbage collection. Dolt makes a lot of disk garbage, especially during import. This was surprising to many users who would try to import a dump file and notice Dolt used a ridiculously large amount of disk, sometimes 10-20X the size of the dump file. In February, we released automatic garbage collection and in November we turned it on by default in Dolt versions > 1.75. Dolt no longer delivers full hard drive surprises.

Best Use Case#

We’re all in on agents. Once we saw Claude Code fix a skipped bats test, we knew you would never unleash these little robots on a system that wasn’t version controlled. Version control is one of the three pillars of agentic AI. Dolt is the only version controlled SQL database. It’s almost like Dolt was built for AI. But actually, Dolt is older than the transformer.

We’re not alone in thinking AI needs version control. The founding engineers at Cursor think so. The UC Berkeley computer science department thinks so. Andrej Karpathy, one of the leading thinkers in AI, thinks so.

Best Tech Blog Article#

Nick wrote a couple great technical articles this year. I personally enjoyed how he adapted vector indexes to be version controlled. That article really shows off the deep computer science that makes Dolt possible but it’s a tad inaccessible.

So, I went with his more popular article “People keep inventing Prolly Trees”. The Prolly Tree is the novel data structure that powers Dolt. The article surveys all the places similar data structures have been invented and used.

Funniest Blog Article#

The funniest blog article of the year is Coding Agents Suck Too. It’s funny as a standalone artifact but it’s even more funny because it was published a mere six weeks before our company went all-in on agents.

Biggest Surprise#

Last year, we were surprised for Dolt to be 10% slower than MySQL on sysbench. So, this year we’re obviously again surprised to be as fast as MySQL on sysbench. We started 15X slower on this metric so getting to parity seemed a bridge too far, even last year. But, we did it.

Most Excited For in 2026#

We’re fast approaching Dolt 2.0. All the milestones we hoped to achieve before release are all in the current release: automatic garbage collection, archive format, performance parity with MySQL, and Beta vector support. We’re just ripping out some dead code that supports pre-1.0 versions and adding adaptive encoding, a feature we built for Doltgres. Who doesn’t get excited for a major release?!

Conclusion#

That’s a wrap for 2025. See you in 2026. Stop by our Discord and wish us a happy new year.