Prior Art

Prior art: what we borrow

The point of this section is to speed up the plan with real projects: take code or approach, integrate a component as is, or use as a UI reference.

9.1 Panel-level PaaS (UX and domain model reference)

Project	What is good	What Maurice takes
Coolify (PHP/Laravel, Docker)	Most complete domain model: projects, environments, services, domains, S3 backups, notifications, service catalogue, "magic" variables linking services	Entity structure and UX of key screens; env references; catalogue templates (compose templates converted to Quadlet)
Dokploy (TypeScript, Swarm, Traefik)	Clean TS stack (Next.js, tRPC, Drizzle), readable code, DB backups to S3, templates, monitoring	If the control plane is TS — API/UI structure reference; backup and schedule logic
Komodo (Rust + TS)	Multi-server "core + periphery agent", single API/TS client, builds and deploys	Control plane ↔ agent model; message format between them
ZaneOps (Python, Swarm, Caddy)	Caddy as a managed proxy via admin API, simple service model	Caddy config generation and full rebuild on restart
Dokku (shell, Docker)	git push → deploy, buildpacks, minimalism, plugins	Plugin system idea for integrations; buildpacks as a build option
Rivetr (Rust)	Single binary, low footprint, webhook deploys, runtime abstraction Docker/Podman	Lightweight control plane example; GitHub/GitLab/Gitea webhook handling

9.2 Quadlet-native panels (systemd/Podman reference)

Project	What is good	What Maurice takes
Rookery (Go)	Control plane + agent per rootless scope; units on disk as source of truth; import of podman run / compose / running containers into Quadlet with warnings	Agent architecture; compose → Quadlet converter (key for migrating off Coolify)
quadletman	Web UI for rootless Quadlet with OS-user isolation	Working with several system users from one process
quadlet_manager (Go + React)	Unit management via systemd D-Bus, xterm.js web terminal, logs	D-Bus layer instead of shelling out to systemctl; web terminal
quadman (TUI)	Acting as other users via sudoers with an explicit command list; events timeline; templated web@ units	Minimal sudoers for the agent; merged podman + systemd event stream
podlet	podman run / compose / k8s YAML → Quadlet converter	Integrate as a library or CLI in the validator and importer
cockpit-podman	Reference use of the libpod API from the browser	API call reference for containers, logs, stats

9.3 Components integrated as is

Task	Component	Notes
Proxy and TLS	Caddy (admin API, on-demand TLS, DNS-01 plugins)	Not rewritten; the panel generates JSON config
L4 and protection	HAProxy	Added when needed; config generated
Registry	zot / distribution	Ready OCI registries, digest-addressed
Backups	restic + rclone; optionally Backrest (web UI over restic)	Schedule and retention; Backrest as UI/API reference
DB dumps	pg_dump / pgBackRest, clickhouse-backup, dragonfly SAVE	Wrappers in catalogue templates
Logs	journald → Vector/Alloy → VictoriaLogs	Single-binary store, label search; see Telemetry
Metrics	node_exporter, cAdvisor (Podman) or systemd-exporter → VictoriaMetrics	PromQL-compatible, charts in the Maurice UI; Grafana optional
Notifications	Apprise or direct Telegram Bot API	Apprise gives dozens of channels at once
Tunnels	cloudflared + Cloudflare API	Routes via API, token held by the panel only
LLM adapter	Vercel AI SDK (TS) / LiteLLM (proxy)	Any provider via the user's key
MCP	Official MCP SDKs (TS/Go)	Server over the panel API
STT for Telegram	Whisper (faster-whisper locally) or a provider STT	Voice to text
Dev environments	toolbx / distrobox, code-server, Coder (reference)	See Remote Development
Agent memory	Entire CLI	See Agent Memory

Conclusion. Only the control plane (data model, Quadlet generation, plan/apply, UI) and the agent are written from scratch. Everything else is integration or borrowing. This roughly halves the Alpha and MVP effort compared to a fully custom core.