It’s been a rough time for vector databases.
A wonderful, hot technology that’s suddenly run into some headwinds. Or maybe a small hurricane would be more accurate.
That’s never happened to a hot technology before, right?
Well, other than BigQuery, Hadoop, Google Glass, Second Life, Quibi. And mainframes. And a few others, enough to Segway to my next point (yes, the Segway failed too).
If you’ve been following our blog here at SWIRL, you know we’ve been arguing for some time that vector databases add unnecessary complexity, cost, and delays to AI projects.
That they lack too many database features: reliable ACID semantics. Analytics. Effective RBAC.
That you still need to maintain your old, specialized databases to do all the non-vector work.
Meanwhile, SWIRL creates on-the-fly vector embeddings in memory. All the benefits of vector databases without the hardware. Plus no issues with confidential data being saved in the embeddings since the embeddings disappear after use.
And now Amazon has announced native vector embeddings in S3. Once again, all the vector benefits, no need for additional infrastructure.
So why would anyone go through all the headache, hassle, and expense of using a vector database?
There are some possible reasons:
- Because nothing says “cutting-edge” like recreating your entire tech stack to solve a problem you didn’t actually have.
- Because they miss the thrill of manually rebuilding ETL pipelines every time someone sneezes near the schema.
- Because their cloud budget wasn’t hemorrhaging fast enough.
- Because they thought "vector database" would look cool on the pitch deck next to the words “synergy” and “blockchain.”
- Because they haven’t met SWIRL yet.
Zero ETL. Zero-trust security. Real answers.
And no data remodeling, duplication, or delay.
SWIRL.