Infino AI
The single-Parquet-copy thing is the detail that got me. Most teams I talk to are running a search index, a vector database, and a warehouse, and they're all holding the same data with slightly different shapes. That's three systems to keep in sync, three bills, three failure modes. Infino is basically saying, what if you just had one file that everything can query, and that's it. For a small team shipping an agent this week, that's a real unlock.
I'd want to know how far the SQL and vector paths scale before I sign on. The pitch works beautifully at the prototype stage, and agents are mostly prototypes right now. But the moment you have real users hammering it, the question becomes whether one copy can hold up under concurrent load. Still, the direction feels right. The agent stack is getting too fragmented too fast, and anything that collapses the moving parts deserves a hard look.
What I like most is that it's boring in the best way. Parquet, SQL, and vectors are all proven pieces; Infino just puts them in the same box and points it at agents. That's a pragmatic move I can get behind.