A self-hosted landing page for everything running on the server: media, security, home automation, apps and infrastructure. Each service shows a live status dot, and a built-in AI agent ("Miggles") can check what's up, find services and read server load from a chat panel.

• One YAML file lists your services (config/services.yaml). Edits apply on refresh.
• Live status comes from server-side pings over your LAN or Docker network, so checks never go through Cloudflare.
• Routing badges show whether each service goes through Cloudflare, NPM or LAN only.
• AI agent runs on Claude through the Anthropic API. Its tools are read-only: list services, check status, read host stats.
• Search (press /), category filters, dark and light mode, and a mobile layout.

git clone https://github.com/MisterMiggles/MisterMiggles.git hub && cd hub
cp config/services.example.yaml config/services.yaml   # then edit it
cp .env.example .env                                   # add ANTHROPIC_API_KEY
docker network ls                                      # find NPM's network, set PROXY_NETWORK in .env
docker compose up -d --build

In Nginx Proxy Manager, add a Proxy Host:

So that the agent's replies stream instead of arriving all at once, add this under Advanced → Custom Nginx configuration:

proxy_buffering off;
proxy_read_timeout 300s;

The app also sends X-Accel-Buffering: no, so this is a belt-and-braces setting.

Pick one of these:

1. Proxied DNS → NPM. Point an orange-cloud A/CNAME record for hub at your home IP. Set SSL/TLS mode to Full (strict), and give NPM either a Let's Encrypt cert (DNS challenge with a Cloudflare API token) or a Cloudflare Origin cert.
2. Cloudflare Tunnel. In cloudflared, route hub.yourdomain.com to http://mistermiggles-hub:3000, or to NPM if you want NPM to stay the single entry point.

The dashboard shows your internal layout, and the agent spends API credit, so don't leave it open to the internet:

• Cloudflare Access (Zero Trust → Access → Applications). Add a self-hosted app for hub.yourdomain.com with a policy that allows only your email. The hub shows the signed-in email in the header.
  • If the hub is reachable only through Cloudflare (a Tunnel, or a firewall that allows only Cloudflare IPs), also set REQUIRE_CF_ACCESS=true. The app will then reject any request that arrives without Access's identity header.
• Or an NPM Access List. Use basic auth or allow LAN IPs only, and attach it to the proxy host.

See config/services.example.yaml for every option. Example entry:

categories:
  - name: Media
    icon: "🎬"
    services:
      - name: Jellyfin
        url: https://jellyfin.yourdomain.com      # what you click
        check: http://192.168.1.10:8096/health    # what the hub pings (LAN/Docker address)
        icon: https://cdn.jsdelivr.net/gh/homarr-labs/dashboard-icons/svg/jellyfin.svg
        description: Movies and TV
        via: cloudflare                           # cloudflare | npm | tunnel | lan

How status is reported:

• up: the app answered with any status below 500. A 401 or 403 still counts, because the app is running and just wants you to log in.
• degraded: the app answered with a 5xx.
• down: timeout, refused connection or DNS failure.
• insecure: true skips TLS verification. Use it for self-signed internal endpoints like Portainer on :9443.
• check: false turns off the status dot, which suits external links.

Use internal addresses for check. A check through Cloudflare Access gets the Access login page back, so a dead app would still show as "up". Container names work too, e.g. http://jellyfin:8096, as long as the hub shares a Docker network with that container.

The icon set at homarr-labs/dashboard-icons covers nearly every self-hosted app. Plain emoji also work as icons.

Set ANTHROPIC_API_KEY in .env (get a key from https://console.anthropic.com). Without a key the chat button stays hidden and the dashboard works as normal.

• Model: claude-opus-5-5 by default. You can override it with AGENT_MODEL.
• AGENT_EFFORT sets how hard it thinks: low is cheapest and fastest, medium is the default, and high and above suit more complex questions.
• Refusal fallback is on. If a request is declined by Claude's safety classifiers, the API re-runs it on Anthropic's recommended fallback model (fallbacks: "default"). To turn this off, remove the fallbacks and betas lines in server/agent.js.
• Conversations are kept in memory on the server and disappear on restart. The New button starts over.

Tools the agent can call (all read-only):

To give the agent more reach, add a tool in server/agent.js: a JSON schema in tools, a zod schema in schemas, and a function in handlers. Good next steps would be Home Assistant entity states (GET /api/states with a long-lived token), Docker container status from a read-only socket proxy, or the NPM API's proxy-host and cert-expiry list. Keep new tools read-only unless you really want the agent to make changes.

npm install
cp config/services.example.yaml config/services.yaml
ANTHROPIC_API_KEY=sk-ant-... npm start     # http://localhost:3000

Requires Node 20 or newer.

server/index.js    HTTP server: static files, /api/config, /api/status, /api/agent/*
server/config.js   Loads services.yaml and strips internal URLs before they reach the browser
server/status.js   Health checks (cached 20 s)
server/agent.js    Claude agent loop and tools
public/            Dashboard UI (no build step)
config/            Service catalog