MQTT settings with several rhasspy


I have a problem getting two rhasspy working with Hermes Led Control

With snips, I can ssh the master and turn on or off leds like this:

mosquitto_pub -p 1883 -t 'hermes/leds/toggleOff' -m '{"siteId" : "salle"}'
mosquitto_pub -p 1883 -t 'hermes/leds/toggleOff' -m '{"siteId" : "studio"}'

So it will turn off leds on both devices

Now rhasspy:
I have master with MQTT enable, host = its own IP, site-id salle
I have a pi 0 with MQTT enable, host = its own IP, site-id studio

I can only turn on/off the device itself. If I run the command on ‘salle’ with site ‘studio’ nothing happens and vice versa.
I tried to set MQTT on the sat with the master IP as host, but it does the same.

With MQTT explorer, I can see that the mqtt message isn’t received by the other device.

Is it normal right now, or am I missing something ?
Should each rhasspy have its own IP as MQTT host, or all rhasspy devices should have the master IP as MQTT host ?

BUT: on salle with ssh, this work:

mosquitto_pub -h -t ‘hermes/leds/toggleOff’ -m ‘{“siteId” : “studio”}’

.141 is the IP of the sat (the other rhasspy pi0)
So working on port 1883, seems topics are not forwarded


I don’t know how next full hermes/serviced rhasspy will change all base/sat settings and such, so if all such settings and workflow will change with next version and such setup is premature, let me know :kissing_smiling_eyes:

This does not seem to be related with Rhasspy at all :thinking:

Anyway… All devices should be connected to the same broker (one single IP) then publishing to this broker IP will be broadcasted to all devices.

Use the IP of the master for MQTT broker for all devices and publish to this IP.

It should work as expected then.

Hope this helps.

Ok will try that.

Also I see in MQTT settings, siteId that we can use comma separated.
Should I put:
on master : salle, studio
on sat : studio, salle


mm… Looking at the code it seems that the satellite should only have their siteId and the master should have all the siteIds (his and all the satellites’).

I’d try:
on master : salle, studio
on sat : studio

@synesthesiam Can you confirm this? :slight_smile:

Doesn’t work either.

Same for this : /api/text-to-speech with site-id not working

Can’t send to /api/text-to-speech?siteId=studio will always play on the master :sleepy:

Ok my own on this one, now works perfect !

Like @fastjack said, both master and satellite have same MQTT host in setting.

But as HLC read rhasspy profile to know which mqtt ip/port to listen to, I just had to restart hlc service :yawning_face:


HLC wouldn’t work with comma separated siteid.

Could be nice to have a how to on this. If having all sat siteid in master sitedid is confirmed I can do a PR on HLC to support this.

@synesthesiam could you confirm or not that a master should have in mqtt settings siteid its id and all satallites ids comma separated ? Sort of siteId : master,sat1,sat2

Yes, this is how it’s supposed to work. The first siteId (“master” in your example) is what Rhasspy should use to publish new messages.

For responses, though, the same siteId is used. So if a hermes/nlu/query comes in, the hermes/intent/<intentName> response from Rhasspy will have the siteId of the query.

Hope this makes sense.

Hello everybody,
I folowed all this mqtt disscussions for a while and there is something that I really do not understand.

WHY ???

Rhasspy works with API and web sockets. This works perfekt with nodered. You want to implement a full hermes compatible mqtt comunication. Where is the advantage?
Even if mqtt is (full) implemented, web socket should be remain, as I read. In your plan, you also want to run an own Broker on the satellites. This takes recources (PI zero).
With websocket you have a direct connection between sat and master. No one else is involved. And you can keep wakeword and voice recognition on sat and the rest on master.

Pleas do not missunderstand. I do not want to put your idea down, but i want to understand it.

Thank you in advance


Would it be possible to disable the browser gui for zeros. Maybe this would save recources.

Hello Kay,
Well we need some kind of a message bus between the separate components inside Rhasspy. MQTT is a very easy way to do this and pretty common in home automation systems, but there are definetly other ways to achieve this aswell.
I do not really see a reason to run a mqtt broker on any satellite devices, since they can simply connect to the broker on the main device and broadcast their audio once the hotword was detected, at least thats how it worked with Snips.
Re Browser Gui. Just as with the MQTT Broker there is no reason to run this on the satellite devices, since satellite devices only need to detect the wakeword and broadcast the audio afterwards (You only have to create Intents on the Main Device anyways). This is basically a big reason to have separate services for everything starting v2.5 since this way it is pretty simple to deploy for example only the hotword/broadcast audio service to satellite devices and have the other services like the ASR that are more recource intense running on a central device like e.g. a normal Raspberry Pi.
I would not really worry about recource requirements for the MQTT Broker, because they are very low even when the broker is actively used.


I agree with you for mqtt (all connected to the master of course), but I would keep GUI on satellites and especially some endpoints. /api/text-to-speech and /api/listen-for-command for example are two I use on satellites and will really need in the future. Or find a way so we can use them with siteId on master and the master forward them through mqtt to the satellite. Which doesn’t work actually. Or even some /api/mqtt/{topic:’’, payload:} …

Also the GUI is very usefull to access settings (wakeword, mqtt for siteId and master ip, tts) and see if running, log to debug what happens, etc. Each satellite should have a lite version of the GUI.

Oh I simply forgot about this :sweat_smile:
But since it is modular I suppose this is something everyone can deploy the way he wants :slight_smile:
I do not think that there is any performance problem for satellite devices anyways, since both wakeword and GUI do not really require much recources
However when someone wants to have a gui but does not want to have it running all the time he could still simply start the gui container on a user interaction like e.g. a settings button on the satellite device I suppose, just as he might want to have a mute button on there aswell that could simply kill the wakeword container

1 Like

I have full venv rhasspy with mqtt as a satellite ( forward asr nlu etc) and no problem at all on pi zero. Runs fast.

Thanks for your answers. I also have zeros, connected by websocket. This works.
Of course there must be one kind of information transfer.
If you prefer mqtt and this would solve problems, (which really exist) and which are not solved with a method that actually exist, I also would change. But in this case I would change completly. If mqtt is the better system we could leave websocket and api behind.
Mqtt would be, as snips shows, the only system.
If you have a modular system and could chooce between mqtt and another messaging system, it would only makes sence if there is a real advantage in one.
I know mqtt and its great, useful and lightweight. But then there is no reason for another system, which makes the same. Only work to keep both systems up to date.
A mixture between both will end in trouble, I guess.

So I will build up a second docker nodered with mqtt rhasspy and see whats coming.

@ KiboOst
Do you send answers to zeros with mqtt?
How do you do that?
I think about creating answers with text2soeech on master and send wav to zeros

What do you mean by “answer” ?

I don’t use any mqtt commands, I use master and satellites exclusively with jeedom plugin, which use /api/texttospeech and /api/listenforcommand

I mean:
sat receives message
master “reads” intents and slots
nodered does somthing like switching lamps and sends answers back to sat
“I switched on the light”
or answerd
The time ist 13:15h
So the sat hast to receive text to say.

Yes for this the jeedom plugin use the siteId in the intent to call the right master/sat with /api/text-to-speech to say the text.