Last modified September 12, 2026

Domain allowlist

Giant Swarm clusters need outbound access to the external domains listed here to function. Use this page to configure your firewall, proxy, or egress policy.

How to read this page

Everything in the required domains, on-premise installations and Agent Platform tables is HTTP or HTTPS traffic over TCP on port 443, unless specified otherwise.

Two flows aren’t HTTP, so a conventional firewall can’t match them by hostname. They have their own section, non-HTTP traffic.

Because a wildcard never covers the domain it belongs to, every base domain we also need access to gets its own row next to its wildcard.

Required domains

On-premise installations

These domains are only required for on-premise installations. They’re HTTP or HTTPS traffic, like the required domains.

Agent Platform

The Agent Platform is an opt-in addition to a management cluster. Its own components need nothing beyond the required domains: images and charts come from the Giant Swarm registry and agent skills from GitHub. What it adds is the model provider you choose for your agents. The platform ships without a model credential and without a preferred provider, so which of these domains you need depends on the provider you bring, as described in Bring your own model.

Amazon Bedrock is served from *.amazonaws.com, which is already a required domain. A model endpoint you host inside your own network needs no public egress at all.

A firewall that blocks the provider doesn’t show up as a firewall error: the model configuration is accepted, the agent starts, and every request fails with a connection reset. See the troubleshooting section of the model guide.

Non-HTTP traffic

Clusters open two kinds of connection that don’t carry an HTTP request, so there’s no hostname for a proxy to read and match. Handle them like this:

• If your firewall can resolve and match hostnames itself on non-HTTP traffic, put the hostnames below on the allowlist, exactly as you would for an HTTP domain.
• If it can’t, use the fallback described for each flow.

SSH to ssh.github.com

Fallback: We can tunnel this connection through the HTTP proxy with a CONNECT request to ssh.github.com:443. Apply the following patches to Flux’s Source Controller:

- op: add
  path: "/spec/template/spec/hostAliases/-"
  value:
    ip: "127.0.0.1"
    hostnames:
      - "ssh.github.com"
- op: add
  path: "/spec/template/spec/containers/-"
  value:
    name: proxy
    image: gsoci.azurecr.io/giantswarm/socat:1.7.4.4
    securityContext:
      allowPrivilegeEscalation: false
      readOnlyRootFilesystem: true
      runAsUser: 100
    command:
      - /bin/sh
      - -c
      - |
        # Fallback to static argument if injected env var doesn't exist
        export HTTP_PROXY=$${HTTP_PROXY:=${proxy_hostname}:${proxy_port}}

        export PROXY_HOSTNAME=$(echo $${HTTP_PROXY#http://} | cut -d":" -f1)
        export PROXY_PORT=$(echo $${HTTP_PROXY#http://} | cut -d":" -f2)

        socat TCP4-LISTEN:8081,fork,reuseaddr PROXY:$${PROXY_HOSTNAME}:ssh.github.com:443,proxyport=$${PROXY_PORT}

NTP time synchronization

Flatcar nodes synchronize their clock with the Flatcar NTP pool by default. Each pool.ntp.org hostname resolves to a rotating set of servers, so don’t translate these names into static IP rules, which go stale.

Fallback: permit all outbound NTP traffic on port 123. If that’s too broad for your policy, point the nodes at an NTP server you run yourself, through the cluster app’s timesyncd values, and restrict egress to that server.