FAQ: is Levelrail production-ready, is it a Kubernetes replacement, what license, etc.

[thegdsks]

FAQ, answered honestly rather than optimistically. Will keep updating this as the real answers change.

Is Levelrail production-ready?

No. It's early, active development. Single-node and multi-node both run today, but there's no stable release, and APIs, the app spec format, and the on-disk data layout can all still change without notice. Track actual status on the roadmap.

Is this a Kubernetes replacement?

No, and it won't try to be. Levelrail is for a team running somewhere between 3 and 50 services across 1 to 10 machines who wants a private cloud without taking on Kubernetes's operational surface: no CRDs, no custom orchestration standard, no bin-packing scheduler. If you actually need Kubernetes's scale or ecosystem, use Kubernetes.

How is this different from Coolify or Dokploy, concretely?

Coolify and Dokploy drive remote servers by SSHing in and shelling docker/docker compose commands, then parsing text output. Levelrail's agent talks to the local Docker Engine API directly, no CLI shelling anywhere, no SSH session per deploy. See the comparison doc for the specific, sourced differences against Coolify, Dokploy, CapRover, Dokku, and Kamal.

Can I run this on a single VPS?

Yes, that's the default mode. Multi-node is opt-in: the agent works in single-node mode running in the same process as the control plane, communicating over an in-memory transport that implements the same interface real network agents use. Same code path either way.

Does it support Docker Compose?

Yes. internal/compose parses a supported subset of compose.yaml, including per-service build: directives, and expands it into per-service builds and deploys the same way a multi-service app.yaml does, with a frontend Compose creation flow. HTTP-based healthcheck: stanzas translate into real readiness/liveness probes; restart: and networks: are parsed and surfaced as informational notices rather than enforced, since the reconciler always keeps containers running and every service in an app already shares one network.

What's the license?

Apache 2.0.

Does it use AI anywhere in the deploy path?

No, and it's an explicit non-goal. AI is a read-and-suggest layer on top of the platform API (an MCP server, shipped, wrapping the same versioned REST API and bearer-token model the CLI uses), never part of the reconciliation path itself. The HTTP API is designed so that layer bolts on cleanly, that's what "AI-ready" means here, not AI making deploy decisions.

Where do I report a bug or ask something not covered here?

Issues for bugs and feature requests, the Q&A category here for anything else.