dagucloud/dagu
DevOps & Self-hostingA self-hostable YAML workflow engine that schedules scripts, containers, and SSH jobs without an external database.
What it does
Dagu is a local-first workflow engine for internal operations and automation. Instead of writing jobs inside a framework, you describe a DAG in declarative YAML next to your existing scripts, and Dagu handles order, dependencies, schedules, retries, logs, and run history. Steps can be local shell commands, Docker containers, Kubernetes Jobs, SSH commands on remote hosts, HTTP requests, or reusable actions.
The project targets teams whose main job is not orchestration: sysadmins, data engineers, and small infrastructure groups that already have working scripts and want scheduling plus visibility. It is presented as an alternative to cron (no dependencies or history) and to heavier platforms such as Airflow or Temporal, where business logic must be rewritten into their programming model. A single Go binary carries the server, scheduler, and executor, with no external database or message broker, running on Linux, macOS, and Windows.
Quick start
Install with one of the supported methods, for example Homebrew:
brew install dagu
Other options include a shell installer, npm, Scoop, Docker, Helm, and a Windows installer (installation options).
Create hello.yaml in an empty folder:
steps:
- id: hello
run: echo "hello from Dagu"
Run it once, then start the Web UI from the same folder:
dagu start hello.yaml
dagu start-all --dags .
Open http://localhost:8080 and, on first launch, create the admin account. If an installer registered Dagu as a background service, it may already be running on port 8080; add workflows through the UI or place YAML files in the DAGs directory the installer reported. You can also try the hosted live demo.
Workflow capabilities
Beyond running commands in order, Dagu supports parameter forms rendered in the Web UI, parallel fan-out over Sub-DAGs with max_concurrent, cron schedules with overlap policies (skip, all, latest) and catch-up windows, per-step retry policies and continue_on handling, lifecycle handlers, and human tasks that pause a run for operator input. Build workflows can reuse a step's output when its command and input files are unchanged (incremental workflows). External CLIs can be pinned and cached through tools:, using aqua as the default provider.
Runs write logs, state, and artifacts to local files. Prometheus metrics, structured JSON logging, email/webhook notifications, and an artifact browser support day-to-day operation.
Execution models
All modes are self-hosted and run the same YAML. dagu start-all puts the HTTP server, scheduler, and coordinator in one process. Set DAGU_HEADLESS=true to drop the Web UI for CI or CLI-only use. For scale-out, run dagu coordinator and dagu worker separately; workers poll the coordinator outbound and can be routed by labels. Dagu also offers an experimental embedded Go API for starting runs from a host application.
Constraints and licensing
- The core is GPL-3.0; embedded API and commercial embedding terms are covered separately in
LICENSING.md. - SSO, RBAC, audit logging, incident integrations, and multi-server monitoring are described as self-host license features; community self-hosting requires no license key but expects you to operate storage, upgrades, networking, and workers (pricing details).
- The default bind address is
127.0.0.1. For shared or network-exposed instances, the docs recommendbuiltinauth, keeping metrics private, and enabling peer TLS in distributed deployments. Mounting the Docker socket or using host executors grants privileged access to the host machine. - Container steps need extra Docker daemon configuration when Dagu itself runs in Docker.
- Build workflows with result reuse currently run locally only.
- Scoped notification routing is broken in v2.11.0–v2.11.2; use v2.11.3 or later.
HotRepo's take
Dagu's 4,617 stars and 403 forks put it in the same conversation as cron replacements and lightweight orchestration, and its Go-language daily snapshot on 2026-11-16 recorded a 14th-place position with 8 stars added that day. The design choice that stands out is keeping workflow structure as configuration rather than framework code: one binary, file-based state, and no database to operate. That makes it easy to try, but it also means reliability and observability are tied to the local filesystem unless you add workers and a shared volume. The paid self-host features suggest the project is also building a commercial layer around the open core.
Written by AI from the project's README, updated . The project's own docs are the source of truth.
You might also like
- windmill-labs/windmillSelf-hostable code-first platform that turns scripts in 20+ languages into workflows, UIs, APIs and AI-assisted internal apps.
- argoproj/argo-workflowsRuns multi-step batch, ML and CI jobs as container DAGs inside Kubernetes, for teams already on a cluster.
- apache/airflowAirflow defines workflows as Python DAGs and schedules, monitors, and backfills their tasks across workers.
- conductor-oss/conductorSelf-hosted workflow engine that persists every step of microservice and AI-agent orchestration so runs survive crashes.
- apache/dolphinschedulerA low-code platform for scheduling and orchestrating interdependent data-pipeline tasks across teams and clouds.