Changelog Roundup for October 5, 2026

The most important change this week is for operators and customers alike: a visitor’s browser can now load a public site’s files from nearby operator nodes, not only from our gateways. It is what the network was meant to do from the start.

For node operators

Operator nodes can now serve a public site’s files. When a site is placed on operator nodes, browsers fetch its images, scripts, styles and fonts from them. Nodes behind home routers count, because the browser reaches them over WebRTC, with a relay when a direct connection isn’t possible.

If you run a standalone node, update to v0.0.107 or later. Before that, a Level 1 site served by a standalone node loaded but stayed encrypted, with no error shown to anyone. Docker nodes were never affected.

Nodes started with docker run now keep their identity when recreated. Without our Compose file, recreating the container could bring the node back as new, with its Active Node Hours, leaderboard place and badges gone. Pull v0.0.113 or later and keep your /data volume, where the ID now lives. If you use our Compose file, you need to do nothing.

You now get an email when your node goes offline for 15 minutes, one reminder after 24 hours, and another email once it’s back. We don’t email you about outages on our side. You can turn the emails off in Settings.

A node’s location now comes from its real connection address. A modified node could previously report a different address and so choose its city and metro. Nodes running our software always reported their real address, so nothing changes for them.

Node locations are now looked up in a database we host, so your node’s IP address is no longer sent to an outside service. Every node was located again on 2 October, so your city, map pin or metro leaderboard may have changed. A metro you set yourself in the portal still wins.

From v0.0.116, a new Linux node on a system with no machine ID takes its identity from the hardware, so it survives a wiped data folder. Existing nodes and containers are unchanged. This was tested against sample firmware values, not yet on real hardware.

The edge node now runs on 32-bit ARM, which covers the Raspberry Pi 2, 3 and 4 on a 32-bit operating system. It is a binary install only, as there is no container image yet.

Signing in now takes you to the part of the portal you use. Your view is saved with your account, so operators land on their dashboard on any browser or device.

For customers

Your site’s files can now come from operator nodes near your visitors. Every piece is checked against the record made at upload, and anything that fails or is slow falls back to the gateway. Pages always come from the gateway, and your addresses don’t change. It applies to Level 0 sites on their own domain. For now we place sites on request, so ask customer support with your domain.

A _headers file at the root of your build now sets response headers on Level 0 and Level 2 sites. Every matching rule applies, and a Content-Security-Policy works without breaking the page. Visitors can also no longer download your _headers or _redirects files.

_redirects now handles rewrites and per-section 404 pages on Level 0 sites. Status 200 serves another page at the same address, which is the usual single-page-app fallback. Check your existing rules, because a rule now gives way to a real file at its source path. Add ! to the status to make the rule win.

Add # ee:clean-urls on to _redirects and /about serves about.html directly, with no redirect. It applies to Level 0 sites only.

Portal deploys now check your _redirects file before publishing, as the command-line tool already did. Errors stop the deploy and name the line, and every problem is reported at once.

Videos and large files on your site can now be seeked and resumed. Before, a video couldn’t be seeked and Safari wouldn’t play it at all.

Encrypted workloads no longer reveal their unencrypted size in their readable header. Rebuild and redeploy Level 1 and Level 2 workloads to get the new header.

Deleted workload secrets now stop working across the fleet within five minutes. Before, a deleted secret could keep being handed to edge nodes until our systems restarted.

Level 1 sites now decrypt when served by a standalone node. You need to do nothing.

For everything else, see the full changelog.