built reference verified 503cd17 (2026-09-13)

Operate your node

The overlay is the management plane. Once a router has joined, you reach and update it over the mesh. Ethernet at 192.168.1.1 is only for recovery.

Status: built. Remote OTA over the internet with signed payloads is planned, not built.

Reach a router

Where you are	How
On the mesh (its Wi-Fi or a LAN port)	`ssh root@10.254.x.y`, routed directly with no jump host
On the network upstream of the mesh	Add `ProxyJump root@192.168.1.1` through one wired router
On someone else's network the router is plugged into	Press the router's WPS button (below)
Nothing works	Ethernet to its LAN port, `ssh root@192.168.1.1`

SSH on a router's WAN side is firewalled by default.

A matching ~/.ssh/config stanza for on-mesh use:

Host 10.254.*
    User root
    StrictHostKeyChecking accept-new

Copy files to OpenWrt with scp -O. Dropbear has no SFTP.

Open SSH from the upstream network with the WPS button

Do: press the router's WPS button once.

Expect:

The WPS LED blinks.
SSH (TCP 22) is accepted on the WAN side, but only from the address ranges on that WAN interface right now, never from the whole internet.
It closes by itself after 15 minutes (mjolnir.wan_admin.timeout), on a second press, a reboot, or a firewall reload.

You still need an authorized SSH key. The press doesn't do Wi-Fi WPS pairing.

Update

One router: re-run the installer. Config is kept, a running daemon restarts onto the new build, and a failed health check rolls it back.

deploy/openwrt/install-node.sh root@10.254.x.y

A fleet:

deploy/openwrt/update-fleet.sh                 # every node in fleet-nodes.conf, one at a time
deploy/openwrt/update-fleet.sh --wireless deploy/openwrt/fleet-secrets/wireless.env
deploy/openwrt/validate-fleet.sh               # read-only sweep afterwards

It halts on the first failure (that router has already rolled itself back), and skips unreachable routers. Unchanged binaries aren't restarted.

Rename routers and the network

Change	How
One router's display name	`uci set mjolnir.meshd.name='attic'; uci commit mjolnir; service mjolnir-meshd restart`
All router names from the inventory	`deploy/openwrt/seed-names.sh`
The Wi-Fi name for the whole fleet	Lightning Admin desktop app, or `deploy/openwrt/apply-network-name.sh`

Lightning Admin (admin/) is a desktop app that discovers routers and sets the fleet's network name. Guild membership in it is planned (bead st1.5).

option gateway 'auto' (the default) makes any router with a working WAN share internet with the whole mesh, and stop when the uplink drops. Set never on a router plugged into an untrusted, metered or captive network:

uci set mjolnir.meshd.gateway='never'; uci commit mjolnir; service mjolnir-meshd restart

Diagnose

Question	Command
Is everything up?	`service mjolnir-meshd diag` (read-only, safe on live nodes)
Same build everywhere?	`logread -e "mjolnir-meshd starting"` on each router
Did the last apply work?	`cat /root/mjolnir-stage/result` and `apply.log`

Recover

An experimental flag hung the daemon, for example lan_tunnels=1 (keep it 0):

ssh root@<node> 'uci set mjolnir.meshd.lan_tunnels=0; uci commit mjolnir; service mjolnir-meshd restart'

A bad build passed the health gate: re-run install-node.sh with a known-good build.
Unreachable over the mesh: ethernet to 192.168.1.1. Clear the host key first with ssh-keygen -R 192.168.1.1.

A dead router means a new identity. Its secret never leaves the box, by design. Add the replacement as a new node.

Next: Publish a service on the mesh.