LoRa Mesh

Findings

• Throughput: replicating even a todo list of three short entries is noticeably sluggish — and the radio is only half the explanation. The test setup used OrbitDB with network sync switched off: the courier carries the content-addressed blocks itself and side-loads them straight into the local blockstore, past the IPFS network entirely. First contact is the heavy part — manifest, access controller, identity and entries cross as a block bootstrap of a few kilobytes, roughly halved by gzip — and even afterwards, signatures and identity references keep every entry hundreds of bytes. On top sits the duty-cycle limit (EU 868: ~6 minutes of airtime per hour). The courier itself works — built and tested in funkpost, and on 4 September 2026 it first ran over a real LoRa mesh between two independent nodes, a desktop browser either side — but standard office applications are ruled out. The setup can be tried live: the demo replicates a todo list across the mesh — or, with ?mesh=bc, across two browser tabs, no radio required.
• As a fallback for internet or power outages the current adaptation is unsuitable: it ships whole OrbitDB blocks, which the data rate does not carry. A follow-up release is already being considered that works the way mesh-calendar does — a few bytes of text per event instead of block transfers. What no release removes: every participant would still need radio hardware.
• Platform: Web Bluetooth runs only in Chromium on Android and desktop — no iOS browser. An iOS device is not locked out, though: over Wi-Fi — a mobile hotspot is enough — it can reach a node through Meshtastic’s HTTP/TCP API; a native app would also work and sits outside this evaluation. Device variance is real: a GrapheneOS phone talks to the node noticeably more stably than Android Chrome.
• Adoption: hardly anyone carries a radio node today. That is less a verdict than the actual design problem: commercial applications need ideas that do two things at once — spread the nodes and offer a benefit that justifies carrying one. A use case with a clear USP distributes its own hardware.

What remains

Three planes share the one radio and nothing else. The first — OrbitDB replication — has run, and what it cost is in the findings above. These two are what remains:

A courier for small, rare, valuable payloads. The signalling handshake from design issue #161 fits in two frames; a single signed event — a booking, a handover confirmation — in a few. Where minutes of latency are acceptable and infrastructure is absent, the model still holds — and metadata that is never generated cannot leak.

Technical foundations

• funkpost — the byte courier itself: framing, ARQ and duty-cycle pacing over Meshtastic® radios — two data planes built (OrbitDB and Yjs), signalling plane designed; live demos.
• webrtc-qr — the library: two devices connect by scanning a code — no signalling server.
• qr01 — the tutorial chapter for this: a todo list moves from phone to phone by scanning a code — over a Wi-Fi hotspot, with no internet involved.
• the demonstrators — both live in the browser: a todo list replicating across the mesh (OrbitDB), and an appointment book for a shop around the corner (Yjs). With ?mesh=bc two browser tabs stand in for the radios; a Meshtastic® node over Web Bluetooth makes it real.

Test channel

Two radios only hear each other if they carry the same key — and if they do not, nothing arrives and nothing says so, which looks exactly like an empty room. This is ours, published on purpose: a meeting place for anyone wanting to try the demonstrators with someone else. It is not a secret, and nothing in them was protected by the channel key in any case.

le-space.de · EU_868 · LONG_FAST · ⌗3dd3

Both demonstrators print that fingerprint beside the channel selector. It is worth comparing across devices: a channel can carry a familiar name and a completely different key, and the name will not tell you.

Importing replaces the entire channel set — every other channel on the device is erased, including keys that exist nowhere else. Export what is on there first.

• Role CLIENT_MUTE — a default node rebroadcasts everything it hears, and on a busy mesh that is most of its airtime, spent on strangers’ packets. Muted, it carries only its own.
• Preset SHORT_TURBO for heavy testing — 21,880 bit/s against LONG_FAST’s 1,070, so the same frame costs a twentieth of the air, and range is irrelevant across a desk. It is deliberately not baked into the channel above: the preset is part of the air configuration, so everyone testing together has to switch at once or they stop hearing each other.

The band is shared and legally rationed — EU 868 gives each transmitter about six minutes of airtime per hour, and a community mesh may have a hundred neighbours in earshot. Bench etiquette has the rest, and channels explains how to make your own.