Comparison
A systemd alternative for Go apps
systemd reliably runs a Go binary as a service, but it has no build, deploy, versioning, rollback, or health-gated cutover. Phelix adds that deploy layer, and can run alongside systemd.
Released Phelix hosts: Linux and macOS. Builds Go and Rust with your host toolchain.
Where each fits
One keeps a service running; the other manages releases.
Reach for systemd when you want a battle-tested init to keep a binary running as a service on a Linux host, with unit files, restarts, and resource limits you manage directly. It is already on the machine and it is excellent at that job.
Reach for Phelix when hand-written units stop scaling: you want versioned builds, real deploy strategies, health-gated zero-downtime cutover, and one-command rollback across one or many hosts, instead of editing a unit and restarting for every release.
Side by side
A neutral capability comparison for Go services.
| Capability | Phelix | systemd |
|---|---|---|
| Role | Build, deploy, supervise, and roll back Go and Rust apps | Init and service manager for Linux |
| Keep a process alive | Yes, per-host supervision | Yes, via unit files and Restart= |
| Build step | Builds Go and Rust binaries with the host toolchain | None; runs a binary you built elsewhere |
| Deploy strategies | Classic, blue-green, rolling, and canary | Swap the binary and restart the unit |
| Zero-downtime cutover | Health-gated proxy cutover between versions | Not provided for app traffic; a restart drops the listener |
| Versioning and rollback | Retained versions, one-command rollback | You track versions and revert by hand |
| Health | HTTP, TCP, and PID tiers gate traffic | Process liveness and an optional watchdog; no traffic gating |
| Multiple servers | Per-agent metrics and logs across hosts | Per host; no cross-host view |
| Secrets | One encrypted store per app: name-only imports, launch-time resolution, rotation with version history, and no value ever printed | EnvironmentFile units read plaintext variables, and encrypted credentials sit outside the build, deploy, and rollback path |
| Security scanning | Built in: local, read-only scans against OSV.dev with evidence-classed findings and an exit 25 CI gate | Out of scope; unit files run services, so vulnerability scanning stays separate tooling |
| Server dependency scanner | Installed package inventory, distribution vulnerability matching, evidence, source freshness, and explicit coverage | Use separate host scanning or deployment tooling |
| Verified server remediation | Selected package preview, privilege and package-state checks, fresh-scan verification, and reboot status | Use separate host scanning or deployment tooling |
| Deployment lifecycle hooks | Defined build, deploy, traffic, stop, and rollback points with timeouts, retries, output, history, and remote execution | ExecStartPre, ExecStartPost, and stop commands cover service lifecycle points; deployment traffic steps need orchestration |
| Deployment rollout impact analysis | Baseline and observation windows, per-metric evidence, impact scores, offline reports, and remote synchronization | Use separate host scanning or deployment tooling |
| Scheduled job workloads | Timezone-aware cron, bounded attempts, concurrency policy, history, redacted logs, and remote execution | Timers and oneshot units support scheduled tasks; execution history is inspected through the journal |
| Automatic replica scaling | Worker and rolling-service scaling with replica limits, evaluation windows, cooldowns, stabilization, and remote controls | Manage service instances and write a controller for replica decisions |
| Per-replica resource visibility | CPU and memory per replica with explicit unavailable and stale states, even when autoscaling is disabled | Service and cgroup resource accounting; collect and present replicas separately |
| Configuration lint and schema | Validate phelix.yaml, reject unknown keys during linting, diagnose ports, and inspect the configuration schema | Unit files and systemd validation; a remote application configuration editor is separate |
| Dashboard configuration management | Synchronized visual and YAML editing, CLI validation, revision checks, atomic saves, and category permissions | Unit files and systemd validation; a remote application configuration editor is separate |
| Proxy traffic reporting | Request counts, errors, and latency in status and dashboard monitoring | Use separate host scanning or deployment tooling |
| Unified local and dashboard logs | One application log workflow with redacted output and connected dashboard reporting | The system journal; add remote collection when needed |
| Remote workload and monitor control | Connected gRPC worker, job, scaling, hook, security, and monitor controls under host authorization and local security consent | Use separate host scanning or deployment tooling |
What Phelix adds on top
The release lifecycle a unit file does not cover.
With a plain unit, a new release means replacing the binary and restarting, which drops the listener for a moment and leaves version history and rollback up to you. Phelix builds and numbers each version, deploys it through a chosen strategy with health-gated cutover, and keeps every version one command from rollback. Health uses HTTP, TCP, and PID tiers rather than liveness alone.
v1.1.0 also makes that release flow safe to automate: a coding agent can propose a deploy through the MCP adapter, but only an immutable, plan-bound change passes the execution-authorization boundary, something a unit file leaves entirely to you.
systemd can still run underneath
They are not mutually exclusive.
Phelix operates applications on infrastructure you already control, and its own monitor runs as a Linux service. You can keep systemd for host-level services and let Phelix own the application release lifecycle on top. If all you need is to keep one binary alive on one host, a unit file is still the simplest answer.
One binary between git push and production.
Install the CLI, verify the host, and follow the quick start to build your first application.