Hermes: Meshtastic
Commander

Where MQTT is configured, and one exception to refusing its bridge (v1.12.0). A tracker publishing over MQTT already arrives as a device_tracker from Home Assistant's MQTT integration, which holds the address, password, TLS and reconnection, and Hermes reads it from there: there was no broker field because there should not be one, but nothing said so, and somebody went hunting for it. The mesh bridged over MQTT can be refused, and what was asked for was to trust your own broker and not the public one: that cannot be built, the packet carries a viaMqtt boolean and no topic. The sender is checked instead of the road: a message encrypted for this gateway alone was sealed with the sending node's own key, and no relay can produce one. The exception holds only for those, and the confirmation is asked on the way on, because it is the first guard that weakens rather than strengthens. The composer sits above the messages, newest first (v1.11).

Weather, trails and any tracker on the map (v1.9 and v1.10). Three layers over the base: Waymarked Trails (the OpenStreetMap route relations, with each network's colour and waymark shield), the RainViewer rain radar ported from Chronos, and the OpenWeather layers, greyed out with a reason when the key is missing; the first two need none, which is why they sit in front. The map also draws any Home Assistant device_tracker, as purple squares: an OwnTracks phone, a Teltonika through Traccar, a LoRaWAN tag, an inReach. Each already has an integration that knows how to talk to it, and an MQTT client inside Hermes would have duplicated one. Under the map a button asks Overpass which marked routes pass near the point: a pin on a ridge becomes "on the GTA, Piamprato to Ronco Canavese", a sentence you can read out over the radio. A polite client, with a one-hour cache and a short timeout, because Overpass runs on donated capacity. A sidebar entry on request, off by default.

Watching a walker carrying a node, without crying wolf (v1.7 and v1.8). A Home Assistant package in docs/hiking: hike mode, distances from the start point and from home, minutes since contact and six alarms, written in YAML rather than Python, because thresholds have to be read and tuned with a slider on the morning of the walk. The dead man switch has two stages and the first wakes nobody: the node buzzes and asks, whoever stopped on purpose answers ok and it ends there; only silence escalates. Standing still and out of contact are two alarms with their own words, and nothing disarms itself. Switching the mode off archives the walk in the Hermes store, because the recorder purges after ten days and "how was that walk in June" is a question asked in September.

Statistics, a log limit, and messages that no longer vanish in silence (v1.5 and v1.6). One intake branch counted a message and recorded it nowhere: a rising counter over an empty log looks identical to a dead radio. The log was capped at 200 rows and on the public channel that was hours: the cap is now a setting, the Log tab says how many rows it holds out of how many, and when full it says so and points at the setting. A Statistics tab computed from what the card already holds, with no new calls and therefore nothing that can drift out of step; ring colours validated for colour vision deficiency, and every ring carries a legend with the counts. The panel says who a node belongs to.

Base map on Esri, away from CARTO (v1.4.0). Since September 2026 CARTO stamps API KEY REQUIRED diagonally inside every tile it serves without a key, and the text is inside the image: no setting on this side could remove it, and the map was unreadable. The new backdrop is Esri's gray canvas, no key, light and dark with the theme, letting the node pins carry the colour. The selector keeps CARTO for whoever holds a key, OpenTopoMap showing the relief (and on a mesh map that is not decoration: a ridge between two nodes is usually the whole answer to why they cannot hear each other) and a custom XYZ template. You can also zoom further in: Esri's tiles stop at 16, and rather than capping the whole map there the last available tile is scaled up to 18, so two nodes a few metres apart can still be told apart.

Diagnostic sensors read by their real id (v1.3.2): the Status tab never showed the executed-commands counter or the last command, and it wasn't a refresh problem, those values were never read at all. The card looked for an entity id ending in 'last_command', but Home Assistant builds the id from the entity's translated name (sensor.gateway_last_command_received in English, different again on an Italian instance), so the search matched nothing. The backend now resolves them through the registry and tells the card. Also added a 'No group' button to take a command out of a group, and a row flash when a command is moved (respecting reduced motion).

Message hardening (v1.2): Hermes never used to look at how a message was encrypted, so the promise that 'a DM with PKC has a verified sender' held for the protocol but wasn't enforced (a sender on older firmware falls back to channel encryption and its identity becomes a claim again). Now, by reading the packet metadata, three switches (all off by default) let you accept only messages encrypted for this node, refuse ones bridged over MQTT and refuse ones too old; already-handled packets are dropped, and commands gain a condition entity and a cooldown as replay defence. The Status tab shows what the arriving messages actually carried, so you can confirm before relying on the rules.

Reorderable and groupable commands (v1.3): the order isn't decoration, because matching returns the first command whose keyword matches, so a specific 'lights kitchen' placed after a broad 'lights' would never fire. Each command now has arrows to reorder it, and several can sit under a group heading (ungrouped commands go last). The switch that requires PKC now states its cost next to it: turning it on stops channel commands, and on a gateway listening on a channel it stops them all.

Normalized group names (v1.3.1): a group exists by being named, and 'Lights' against 'lights' used to make two groups that look like a mistake and are one. Groups already in use now appear as buttons above the field (after the first command it's a click) and the name is stored uppercase, so there's only ever one spelling. Also improved state readability (states written in words) and closed an entity list that showed through where it shouldn't (v1.1).

First stable 1.0 release. Hermes leaves the pre-1.0 phase as a working, tested integration, with the split engine covered by tests.

1.0.1 and 1.0.2 hotfixes right after release: fixed a config flow that failed to import (Home Assistant raised an error at startup and the integration wouldn't configure) and made the card serve even when a gateway won't load, so dashboards no longer show an unexplained 'Configuration error'. Tests also run on Python 3.14.

Panel split into cards and reading the radio's own node database: the card no longer assumes it owns the whole screen, and the map shows every node the radio has heard (with its reported position) instead of only the ones imported by the base integration. Added two single-purpose cards for a dashboard you already use: Hermes summary (the state as a list of parameters that reflows to one value per line in a narrow column) and Hermes chat (reading and sending messages), both lightweight and neither changing a setting.

Radio configuration from Settings with read-modify-write: region, modem preset, hop limit, transmit power and node role are written straight to the device, in a dedicated section with its own save button and a warning, because most of these restart the node and a wrong region cuts it off from the mesh.

Mesh conversations read and written from inside Hermes: the Chat tab lists every channel (not only the ones already spoken on) and lets you reply directly. The Log tab shows what Hermes decided about each message, including the ones it ignored.

Install from My Home Assistant badges (one click opens HACS with the repository pre-filled, another starts the config flow), diagnostic sensors that persist across restarts, and the guide's security section rewritten around the recommended two-instance shape.

First version: commands from the mesh that run Home Assistant actions and reply, broadcast and send_direct services for notifications from automations, a byte-safe split engine, a node whitelist, and one card with the Status, Chat, Log, Devices, Map, Messages, Home Assistant and Settings tabs.