Keel - automated Kubernetes deployments for the rest of us

• Website https://keel.sh
• Slack - kubernetes.slack.com look for channel #keel

Keel is a tool for automating Kubernetes deployment updates. Keel is stateless, robust and lightweight.

Keel provides several key features:

• Kubernetes and Helm providers - Keel has direct integrations with Kubernetes and Helm.

• No CLI/API - tired of f***ctl for everything? Keel doesn't have one. Gets job done through labels, annotations, charts.

• Semver policies - specify update policy for each deployment/Helm release individually.

• Automatic Google Container Registry configuration - Keel automatically sets up topic and subscriptions for your deployment images by periodically scanning your environment.

• Native, DockerHub, Harbor, Quay and Azure container registry webhooks support - once webhook is received impacted deployments will be identified and updated.

• Polling - when webhooks and pubsub aren't available - Keel can still be useful by checking Docker Registry for new tags (if current tag is semver) or same tag SHA digest change (ie: latest).

• Notifications - out of the box Keel has Slack, Hipchat, Mattermost and standard webhook notifications, plus Shoutrrr for services such as ntfy, Gotify, Telegram, Matrix, Pushover, OpsGenie and PagerDuty, more info here

Support

Support Keel's development by:

• Star this repository
• Follow on Twitter

Helm quick start

To put the Admin UI and API behind oauth2-proxy without Keel Basic Auth, use the explicit external-proxy authentication mode.

Prerequisites:

• Helm
• Kubernetes

You need to add this Chart repo to Helm:

helm repo add keel https://keel-hq.github.io/keel/ 
helm repo update

Install through Helm (with Helm provider enabled by default):

helm upgrade --install keel --namespace=kube-system keel/keel

If you work mostly with regular Kubernetes manifests, you can install Keel without Helm provider support:

helm upgrade --install keel --namespace=keel keel/keel --set helmProvider.enabled="false" 

The Helm provider targets Helm v3 and is enabled by default. Helm v2 (Tiller) is no longer supported.

To install using terraform:

resource "helm_release" "keel" {
  provider   = helm.helm
  name       = "keel"
  namespace  = "keel"
  repository = "https://keel-hq.github.io/keel"
  chart      = "keel"
  version    = "v1.0.4"

  set {
    name  = "basicauth.enabled"
    value = "true"
  }

  set {
    name  = "basicauth.user"
    value = "admin"
  }

  set {
    name  = "basicauth.password"
    value = "admin"
  }

  set {
    name  = "image.repository"
    value = "keelhq/keel"
  }

  set {
    name  = "image.tag"
    value = "0.19.1"
  }
}

That's it, see Configuration section now.

Nightly pre-release images

To try out changes on master before they are part of a tagged release, use the nightly images. They are built every night from master (only when there are new commits), after the unit test suite passes, for both linux/amd64 and linux/arm64:

helm upgrade --install keel --namespace=keel keel/keel \
  --set image.repository="ghcr.io/keel-hq/keel" \
  --set image.tag="nightly"

Two tags are published: the rolling nightly tag and a date-stamped nightly-YYYYMMDD tag if you want to pin a specific night's build. Nightly images are pre-releases — they are not supported for production use.

Quick Start

A step-by-step guide to install Keel on your Kubernetes cluster is viewable on the Keel website:

https://keel.sh/examples/#example-1-push-to-deploy

Configuration

Once Keel is deployed, you only need to specify update policy on your deployment file or Helm chart:

apiVersion: apps/v1
kind: Deployment
metadata: 
  name: wd
  namespace: default
  labels: 
    name: "wd"
  annotations:
    keel.sh/policy: minor # <-- policy name according to https://semver.org/
    keel.sh/trigger: poll # <-- actively query registry, otherwise defaults to webhooks
spec:
  template:
    metadata:
      name: wd
      labels:
        app: wd        
    spec:
      containers:                    
        - image: karolisr/webhook-demo:0.0.8
          imagePullPolicy: Always            
          name: wd
          command: ["/bin/webhook-demo"]
          ports:
            - containerPort: 8090

No additional configuration is required. Enabling continuous delivery for your workloads has never been this easy!

Harbor registries

Harbor is supported through both polling and its native webhook. Harbor project names remain part of the repository path, so an image such as harbor.example.com/library/ai-rag:latest is polled at /v2/library/ai-rag/.... Public projects need no registry-specific configuration. For private projects, reference a Docker registry pull secret from the workload in the usual Kubernetes imagePullSecrets field or with the keel.sh/imagePullSecret annotation.

Use a semantic-version policy such as all, major, minor, or patch when the image tag is versioned. A mutable tag such as latest must use the force policy with tag matching so polling compares its manifest digest:

metadata:
  annotations:
    keel.sh/policy: force
    keel.sh/trigger: poll
    keel.sh/matchTag: "true"
spec:
  template:
    spec:
      containers:
        - name: ai-rag
          image: harbor.example.com/library/ai-rag:latest
          imagePullPolicy: Always

imagePullPolicy: Always tells the kubelet how to pull after a rollout; it does not make Keel treat latest as a semantic version. With keel.sh/policy: all and a Harbor repository whose only tag is latest, an empty event list is expected because there is no newer semantic-version tag.

For push-based updates, configure a Harbor project webhook with the default payload format and the PUSH_ARTIFACT event, targeting POST /v1/webhooks/harbor on the Keel service.

Tracking OCI image volumes (Kubernetes 1.31+)

Keel can also watch and update OCI image volume sources (spec.volumes[].image.reference). It is opt-in per resource, mirroring keel.sh/initContainers:

metadata:
  annotations:
    keel.sh/policy: minor
    keel.sh/imageVolumes: "true" # <-- also track image volume references
spec:
  template:
    spec:
      containers:
        - name: app
          image: karolisr/webhook-demo:0.0.8
          volumeMounts:
            - name: oci-config
              mountPath: /etc/config
      volumes:
        - name: oci-config
          image:
            reference: karolisr/webhook-demo:0.0.8
            pullPolicy: IfNotPresent

Image volumes require Kubernetes 1.31 or later and a container runtime that supports image volumes. Kubernetes compatibility is:

• 1.31-1.34: the ImageVolume feature gate must be explicitly enabled;
• 1.35: the feature is beta and enabled by default;
• 1.36+: the feature is stable (GA).

See the official Kubernetes feature-gate table and image-volume documentation for cluster configuration details.

Documentation

Documentation is viewable on the Keel Website:

https://keel.sh/docs/#introduction

Contributing

Before starting to work on some big or medium features - raise an issue here so we can coordinate our efforts.

We use pull requests, so:

1. Fork this repository
2. Create a branch on your local copy with a sensible name
3. Push to your fork and open a pull request

Developing Keel

If you wish to work on Keel itself, install Go 1.26.5 and use the module-based build from this repository checkout.

To test Keel while developing:

1. Launch a Kubernetes cluster like Minikube or Docker for Mac with Kubernetes.
2. Change config to use it: kubectl config use-context docker-for-desktop
3. Build Keel from cmd/keel directory.
4. Start Keel with: keel --no-incluster. This will use Kubeconfig from your home.

End-to-end tests

The PR smoke suite builds the checked-out Keel Docker image and runs it inside a temporary native k3s cluster:

make e2e

Prerequisites are Linux/amd64, Go 1.26.5, Make, Docker, curl, sudo, setsid, and iproute2 (ip/ss). Passwordless sudo is required for the task-owned k3s process. For safety, the command refuses to run when it detects an existing k3s installation, default kubeconfig, k3s network interface, or occupied test port. It never invokes the k3s installer or uninstaller.

k3s is pinned to v1.35.6+k3s1 and verified with its official checksum. Registry and workload fixtures are digest-pinned and isolated by run/test repository. The expected runtime is 6–8 minutes, with a hard 10-minute target. Diagnostics are retained under .test/artifacts/ and exclude Secrets, tokens, environment dumps, and kubeconfig contents.

make test remains the fast unit-test path. CI runs the packaged-artifact k3s path for pull requests, master, application tags, and manual dispatches.

Release candidates use the same harness with the packaged Helm chart and release Dockerfile. Run make release-validate for the complete non-publishing install/upgrade/rollback test, or see Release process and validation for package-only, published-artifact, diagnostics, and recovery commands.

Debugging Keel on Windows

# Ensure we have gcc and go
choco upgrade mingw -y
choco upgrade golang -y

# Move and build
cd cmd/keel
go build

$Env:XDG_DATA_HOME      = $Env:APPDATA; # Data volume for the local database
$Env:HOME               = $Env:USERPROFILE; # This is where the .kube/config will be read from
$Env:KUBERNETES_CONTEXT = "mycontext" #Use if you have more than one context in .kube/config

.\keel --no-incluster

Running unit tests

Get a test parser (makes output nice):

go get github.com/mfridman/tparse

To run unit tests:

make test

Running e2e tests

Prerequisites:

• configured kubectl + kubeconfig
• a running cluster (test suite will create testing namespaces and delete them after tests)
• Go environment (will compile Keel before running)

Once prerequisites are ready:

make e2e

Debugging keel inside the container against your remote cluster (Windows)

The repository contains a debug version of keel container ready for remote debugging.

You can start the debug container with powershell (docker desktop needed):

.\build.ps1 -StartDebugContainers

To connect to your cluster, copy the authentication details from within the keel pod in your cluster from:

/var/run/secrets/kubernetes.io/serviceaccount

to the serviceaccount folder at the root of the repository and make sure you set the environment variable for the K8S management API endpoint:

# This can be configured in envesttings.ps1 to be picked up automatically by the build script
$ENV:KUBERNETES_SERVICE_HOST = "mycluster-o5ff3caf.hcp.myregion.azmk8s.io"
$ENV:KUBERNETES_SERVICE_PORT = "443"

And make sure your API server is accesible from your client (VPN, IP whitelisting or whatever you use to secure your cluster management API).

Once started, simply use the debug option in a Go debugger, such as Jetbrains GoLand:

Debugging a Go application inside a Docker container | The GoLand Blog