Levelrail

Levelrail is an open source, self-hosted deployment platform that turns one or more Linux servers into a private cloud: push to a git repo and get a running app with HTTPS, logs, metrics, and one-click rollback. It ships as a single Go binary for the control plane, uses Docker as the runtime, and is licensed under Apache 2.0.

It is built for people running 3 to 50 services on 1 to 10 machines who do not want to learn Kubernetes. Each server runs a small agent that talks to Docker's Engine API directly, so there is no SSH and no shelling out to the docker CLI. Metrics and logs are part of the core, not a Grafana you install afterwards.

Why Levelrail

Related MCP server: Coolify MCP Server

Try it in two commands

You need a Linux server (amd64 or arm64) with systemd, root access, and ports 80 and 443 open. Docker is installed for you if it is missing.

curl -fsSL https://levelrail.com/install.sh | sudo sh

The installer checks the host, starts Levelrail as a systemd service, and prints a dashboard URL with a one-time setup token. Open it, choose a password, and the setup wizard walks you through a domain, a git provider, and your first app. Then deploy a repository from the CLI the installer put on the box:

levelrail-cli import https://github.com/your-org/your-app --deploy

Without --deploy it prints the deployment plan and stops. Prefer containers, a pinned version, or a source build? See Installing and Docker. Full walkthrough: Getting started.

What you get

• Sleep when idle. An app with no requests for a time you choose stops and uses no memory or CPU, then starts on the next request. See Sleep when idle.

• Many servers, no open ports. Add nodes with a one-time token. Agents dial out over mTLS, with WireGuard mesh networking, internal DNS, cordon, and drain. See Multi-node.

• Git-native. GitHub, GitLab, Bitbucket, and Gitea webhooks and preview environments per pull request.

• A large catalog of one-click templates for self-hosted services, each a compose file the platform deploys and manages. See Template catalog.

• Alerting. Threshold, crashloop, and certificate expiry rules delivered over many notification channels, including Slack, Discord, email, Telegram, PagerDuty, ntfy, and a generic webhook.

• Pipelines. CI/CD runs on your own nodes, defined in a YAML file. See Pipelines.

A tour of the dashboard

Every image is captured from a real, running control plane by scripts/screenshots/capture.sh. Click one to open the guide for that screen.

How it works

flowchart LR
  Git["Git push<br/>or import"] --> API
  subgraph CP["Control plane (one Go binary)"]
    API["HTTP API<br/>+ dashboard"] --> DB[("SQLite")]
    API --> REC["Level-triggered<br/>reconciler"]
    API --> ING["Caddy ingress<br/>(automatic TLS)"]
    API --> BLD["BuildKit"]
  end
  REC <-->|"gRPC, mTLS<br/>(agent dials out)"| AG1["Node agent"]
  REC <-->|"in-process on<br/>a single node"| AG0["Local agent"]
  AG1 --> D1["Docker Engine API"]
  AG0 --> D0["Docker Engine API"]
  CLI["levelrail-cli<br/>and MCP server"] --> API

• Control plane (cmd/levelrail): one Go binary with the dashboard embedded. It reconciles declarative desired state against observed Docker state, the Kubernetes controller pattern without the rest of Kubernetes.

• Node agent (cmd/levelrail-agent): dials out to the control plane and streams Docker events up. In single-node mode it runs in-process, so one node and ten nodes share one code path.

• Ingress: Caddy embedded as a library. Builds: BuildKit as a library, with Railpack detection for apps without a Dockerfile. State: SQLite in WAL mode, pure Go.

• CLI and MCP: levelrail-cli and levelrail-mcp are thin clients over the same HTTP API under /api/v1.

Details: Architecture.

How it compares

Most self-hosted platforms in this category manage servers by SSHing in and shelling out to the docker CLI, then parsing text, which tends to force polling loops. Levelrail uses a reverse-dialed agent and a level-triggered reconciler instead.

This is positioning, not a ranking, and all of them are worth using. The one area where Levelrail trails is breadth of templates: a smaller curated catalog than Coolify's, an intentional bet on curation (see ADR 015). The full, sourced comparison is in Comparison.

Footprint. Measured on an Apple M4 Max with a dev build and a loopback client, the idle control plane used 67.7 MB RSS and 0.03% CPU with 0 apps, and 90.8 MB RSS and 1.60% CPU with 500 suspended apps. No containers were running in that benchmark, so real apps add load it does not measure. Method and caveats: Performance.

Status

Releases are listed on the releases page, and every feature is labelled stable or beta, with the evidence behind it, on Feature status. Single node is the best-tested path. Anything marked beta works but has less real-world mileage, and a few areas stay off by default behind APP_EXPERIMENTAL. The roadmap lists what is done and what is next.

Guides

Step by step walkthroughs, each tested against a real install:

For AI agents and crawlers

• Agent skill: skills/levelrail/SKILL.md teaches a coding agent to deploy, wait, diagnose and roll back safely.

• MCP server: levelrail-mcp exposes the same API and permission model. Start in read-only mode. See MCP and AI control.

• Repo guide: AGENTS.md for working in this codebase, and llms.txt as the index.

• Docs for LLMs: llms.txt and llms-full.txt.

Documentation

• docs.levelrail.com: the hosted docs, with getting started, architecture, app spec reference, and the full index.

• docs.levelrail.com/llms.txt: the docs index in a plain format for LLMs and AI assistants.

• docs/: the same content as Markdown, for browsing on GitHub.

• Start here: Getting started, Installing, App spec reference, CLI reference.

Community and support

• GitHub Discussions: questions, ideas, show and tell.

• GitHub Issues: bug reports and feature requests.

• GLINR Discord: live chat with maintainers and other users, with a dedicated #levelrail forum channel.

• support@levelrail.com: direct email support.

• Security policy: report a vulnerability privately.

• Contributing: dev loop, tests, and commit conventions.

Build from source

Needs Go 1.26 or newer and Docker.

go build ./cmd/levelrail
go build ./cmd/levelrail-agent
go build ./cmd/levelrail-cli

The frontend is a Vite project in web/ that is embedded into the control plane at build time. See Installing. Generic, stdlib-only Go packages (SSRF-safe HTTP client, cron parsing, health probes, and more) live in a separate module, kit/, usable without the platform.

Contributors

Star the repo

If Levelrail saves you a weekend, a star on GitHub helps other self-hosters find it. Bug reports and honest feedback help more.

License

Apache 2.0, see LICENSE.

Founding partners

Founding partner sites and libraries, from the same studio:

Levelrail is built by Gagan Deep Singh, founder of GLINR STUDIOS and theSVG. Say hello at thegdsks.com.

Powered by GLINR STUDIOS · icon from theSVG