zerops runs multi-tenant on incus LXC today. that's been good for density, fine for trust, and increasingly tight as we scale. the move is to put a microVM around each tenant — kernel-level isolation, smaller blast radius — and keep an LXC fleet inside that microVM for the actual workloads. we get the kernel boundary at the customer line and the density we already shipped.

honest trade-offs:

• cold start is worse than bare LXC.
• memory ballooning and overcommit need tuning.
• kernel surface inside the guest is a thing we now own.

none of these are deal-breakers, but they're real and we plan around them.

the bet: density × isolation is the unlock for going worldwide. you can't run banking-grade tenants on a shared kernel forever, and you can't pay public-cloud per-VM economics either. this is the middle.

we're shipping observability where the boring 95% (collection, retention, dashboards) is solved and the AI helps with the 5% that hurts — correlating a deploy with a latency spike, summarising on-call noise, suggesting the next probe. not a chatbot bolted on; the AI sits where it's actually cheaper than a tired human at 3 a.m.

postgres, redis-family, search, queues, object storage. opinionated defaults, sane upgrade paths, no surprise downtime. the goal is that you stop noticing them, which is the highest compliment a managed service can earn.