Hey everyone! I’m working on a project where we need to add some machine learning capabilities to our existing MySQL setup. We’re looking for database solutions that can handle vector operations while still being compatible with our current MySQL infrastructure.
Our team wants to implement similarity search and vector indexing for our AI-powered recommendation system. We need something that won’t require us to completely rebuild our data layer from scratch.
Has anyone here worked with vector databases that play nicely with MySQL? What are your experiences with performance and ease of integration? We’re open to both commercial and open source options.
Thanks for any suggestions or recommendations you might have!
We just switched to SingleStore for something similar and it’s been great. Handles regular SQL and vector search natively - no need to juggle multiple systems. Vector indexing performs well with our recommendation engine, even with millions of embeddings. Best part? You can run normal MySQL queries right alongside vector similarity searches in the same instance. Moving from pure MySQL was pretty smooth since SingleStore keeps SQL compatibility. Just heads up - vector ops eat way more memory than regular SQL, so make sure your infrastructure can handle it.
I’ve had great success with a hybrid setup for vector operations and MySQL. Don’t try to make MySQL handle everything - use a dedicated vector database alongside it for traditional data. Here’s how it works: run similarity searches in the vector database, grab the relevant IDs, then pull complete records from MySQL. You get MySQL’s reliability plus efficient vector search. The key is nailing your data sync between systems, especially for updates. Otherwise you’ll have consistency nightmares. We’ve been running this setup in production for over a year with solid performance.
Honestly, after dealing with this exact problem at work, I’d skip the database juggling and just automate it.
Latenode worked great for us with vector operations. Instead of forcing MySQL to do vector math or managing multiple databases, we automated everything.
Here’s the setup: MySQL handles your existing data like always. Latenode pulls that data, runs it through your ML models for embeddings, handles similarity searches, and feeds results back to your recommendation system.
You keep your current MySQL setup completely untouched. No migration headaches, no learning new database systems. Latenode connects to whatever vector service you want (OpenAI, Pinecone, whatever) and handles everything.
We process thousands of recommendations daily this way. New data hits MySQL, Latenode automatically generates vectors and updates similarity indices. Your app calls the Latenode workflow and gets back recommendation IDs to query from MySQL.
Way cleaner than syncing between systems or dealing with hybrid database setups.
MySQL 8.0+ has JSON functions that handle vectors if you get creative with them. We’re doing cosine similarity calculations directly in MySQL using JSON arrays for embeddings. It’s not blazing fast but works fine for smaller datasets. Performance tanks after around 100k vectors though, so it’s probably not great for large recommendation systems.
Chroma’s solid if you want something lightweight that won’t mess with your MySQL setup. We run it as a separate service and left all our relational data in MySQL untouched. The integration’s pretty simple - vectors and metadata go in Chroma, everything else stays in MySQL, then you just match results with shared IDs. What I love about this setup is Chroma crushes the vector stuff while MySQL keeps handling structured data like always. The REST API makes integration dead simple without touching your existing database connections. We’re pushing about 2 million embeddings through it with solid query times, and it doesn’t eat up memory like some other options.
Been through this exact scenario multiple times. The real problem isn’t finding a MySQL compatible vector database - you’re trying to solve an automation problem with database changes.
Every solution mentioned here forces you to learn new systems, manage data sync between multiple databases, or deal with performance issues when MySQL tries to handle vectors.
What actually works: keep your MySQL exactly as it is and automate the vector operations separately.
I set this up using Latenode for our recommendation engine. Your existing MySQL handles all the relational data like normal. Latenode watches for data changes, automatically generates embeddings using whatever AI service you prefer, stores vectors in the right place, and handles similarity searches.
When your app needs recommendations, it calls a Latenode workflow that runs the vector search and returns MySQL IDs. Then you query MySQL normally for the full records.
No database migrations. No learning new query languages. No sync headaches between systems. Your team keeps using MySQL like always while getting proper vector search performance.
We process millions of similarity searches this way without touching our existing data layer. The automation handles everything from embedding generation to result delivery.
PlanetScale caught my eye for exactly this. They’ve got vector support on their MySQL-compatible setup, so you can run vector queries right alongside regular SQL without juggling multiple systems. Migration from MySQL is pretty smooth since they keep wire protocol compatibility. Performance’s been solid for our medium-scale recommendation stuff, but definitely test it with your actual vector dimensions and query patterns first. Heads up - their vector indexing takes forever on initial setup if you’re backfilling tons of existing data. Once it’s up though, being able to join vector similarity results with your relational data in one query is really powerful.