So @cazadordeestrellas basically nailed the basics, but I’d throw in a couple notes from actually dealing with Hailou at work (hype is different than reality, right?). First off, the “edge device” thing is more marketing than meat if you’re not already all-in on their ecosystem—getting models optimized for their “edge runtime” can be a pain, and unless you like futzing with docker images and custom configs, you might bail early. Oh, and “version control” is solid but a bit opaque; rollback can get wonky if you’re dealing with weird dependency chains (I once broke prod, the logs were NOT super helpful vs. something like MLflow). API generation works but don’t expect full auto-swagger docs, and permissioning is basic RBAC—it’s for keeping out the SDE2s, not SOC2 auditors, so don’t get too comfy security-wise.
The dashboard? Yeah, basic, totally agree. Misses a lot of the stuff you get with true observability platforms (Datadog, anyone?). And on pricing: the free tier is skin-tight, so if you’re playing for real, plan for sticker shock as requests scale. I will say though, the pipelining is clutch if you hate building APIs from scratch. TL;DR: Grabs you if you want fast deployment and hate DevOps, but don’t kid yourself—it’s not some miracle worker, more like Heroku for ML inference that gets you 70% of the way and leaves you holding the rest. YMMV.