#Not able to use openAI 5.x?

1 messages · Page 1 of 1 (latest)

weak knoll
#

Am i being dumb but it always responds with errors if i dont use 4o

untold bane
weak knoll
#

golden

#

ty

#

saves me messing around for ages aha

#

is 4o is the best we can use atm

untold bane
weak knoll
#

got link for beta?

untold bane
#

what type of install do you have? OS or container?

weak knoll
#

os

#

oh as in beta HA?

#

or beta openai

untold bane
# weak knoll oh as in beta HA?

yes HA, if you go to settings - system - updates. then in the top right in the overflow menu you can "join beta channel". this will then allow you to update to 2026.1 BETA

weak knoll
#

lovely

#

you legend

untold bane
#

if you find any issues in BETA then can discuss in #beta

weak knoll
#

would love to ❤️

#

been using HA for 6+ years now. not thought of beta before aha

untold bane
#

its usually just a week before release but this cycle its a little longer to give some extra time with the holidays etc...

weak knoll
#

ahh im impatient aha. wanna try 5.1 on the new voice assistant

#

gonna be swapping to a local LLM soon but just using this for now

untold bane
weak knoll
#

rip PI support?

untold bane
weak knoll
#

valid

#

oh no, updated to beta and she gone

untold bane
weak knoll
#

just web ui not loading and my voice no longer connecting

#

💀

untold bane
#

give it a few mins to see if it comes up, can try a hardware reboot of the host too if it still isnt coming up

weak knoll
#

damn shes back. that was the longest reboot of my life 💀

untold bane
#

nice

weak knoll
#

only thing is my voice is no longer responding even on 4o

untold bane
#

might need to restart your voice satellite

weak knoll
#

tried that

#

not even getting lights on it now

untold bane
#

which satellite are you using? is its firmware up to date?

weak knoll
#

i dont have any updates available

untold bane
#

on the device page what firmware version is it running?

weak knoll
#

Firmware: 2025.12.3 (Dec 30 2025, 19:48:58)

#

does that need an upgrade?

#

esp32-s3-devkitc-1

untold bane
#

your running custom firmware?

weak knoll
#

not that im aware of?!

#

just been trying to use esphome

untold bane
#

unless your customising the firmware you should not "take control" of it

weak knoll
#

oh

#

shiii

untold bane
#

its fine a bunch of people do it by accident. 1 moment

weak knoll
#

reflash to default firmware?

untold bane
weak knoll
#

already had to do that once lol

#

probably because i took control and it went wrong lmao

#

lovely, thats done

untold bane
#

unless you are customising firmware for an esphome device theres no need to even have the builder tool installed.

weak knoll
#

calm okay

#

i have it for my other esp devices so for some reason my mind just went straight to it

untold bane
#

can just delete its entry from the builder tool now you are back to stock

weak knoll
#

ah should i have deleted from builder before?

#

Firmware: 25.12.1 (ESPHome 2025.12.1)

#

still just listens for a sec then goes idel

#

light ring doesnt even show up anymore

untold bane
untold bane
weak knoll
#

I deleted, reflashed again and now getting:

Unable to connect to the ESPHome device. Make sure the device’s YAML configuration includes an api section.

untold bane
#

you may have to delete the device from the esphome integration and re-add it

weak knoll
#

not configued in there

untold bane
#

if you cant add in the integration then maybe something went wrong with the flash

weak knoll
#

just reflashed. same error

untold bane
#

try restarting home assistant, maybe its gotten confused with things happening

weak knoll
#

rebooting

#

power cycling the voice

#

worked

#

🤦‍♂️

#

i only rebooted the voice not ha lol

#

yup, still nothing. it responds to its wake word but doesnt give a response

#

not even getting anything in the openai voice assistant debug

#

even changing volume (via ha ui) doesnt update the devices LEDs live anymore

untold bane
#

usually when it responds to WW but doesnt "listen" its because theres something wrong with your pipeline setup.

weak knoll
#

I get this if i try and "setup voice assistant"

image.png
#

do i need to go down that route?

untold bane
weak knoll
#

i dont know if im being stupid, but if i believe what you mean by pipeline. mine WAS fine. I did have 4o responding previously

#

no vlans, mesh yes but other APS are just out of reach

#

well this is on my main network (no vlans on this one)

untold bane
#

that can show up when HA can start connections to the VPE but the VPE cannot start connections to HA. theres a few reasons this can happen

#

vlans/firewalls being a common one

weak knoll
#

going through setup up voice assistant worked when i "took control" with esphome

untold bane
#

something else may be causing the issue now

#

in settings system network. under the "home assistant URL" section what do you see?

weak knoll
#

internet or local

untold bane
#

the "local network" for thee most part should be set on automatic but sometimes people have it on manual then changes happen and the manual breaks.

weak knoll
#

im running behind an nginx proxy

#

nah mine is manual

untold bane
#

is that right?

weak knoll
#

yeah

#

im connected via my nginx proxy atm with no issues

untold bane
#

so just to explain the way it works
api connections are from HA --> VPE
when it wants to play audio it hosts the audio file on the web server and generates a link based on the url and its location on web server and sends to the VPE
the VPE then creates a new connection from VPE --> HA to stream the audio file.

#

hence the investigation

#

try setting it up manually

#

on the device page ensure an "assistant" is selected then check that assistants setttings in the voice assistants settings page to ensure its set up with STT / LLM / TTS correctly.

weak knoll
#

a 5th reboot and its workig again

untold bane
#

awkward... but i guess its eventually a win

weak knoll
#

still no response though

#

WW no light / instant idle

#

i dont know if im being an idiot and doing something wrong

untold bane
#

at the bottom of that screenshot cut off is tha assistant box

#

what is that set to?

weak knoll
untold bane
#

turn off assistant 2 and WW2

#

that is probably not helping with them being the same

#

but may not directly be the issue

weak knoll
#

it does respond to me but hangs. I have to press the button to let it listen to me again

#

still no luck

untold bane
#

ok so you changes from openai to full local.

#

what do those piplines look like?

weak knoll
#

im just trying different ones

#

my openai one worked before beta

#

its back on openai

#

but when i use WW i dont even get a light ring anymore

untold bane
#

can you show me the pipeline settings for the openai pipeline

#

settings - voice assistants - click on the pipeline

weak knoll
untold bane
#

ok so STT is missing it seems.

#

what are you using for that?

weak knoll
#

i did have it on none

#

was that an issue?

untold bane
#

speech is audio
STT converts to text
text is send to llm
llm sends text back
TTS converts text to speech
speech is send to device

#

thats the flow

weak knoll
#

yeaha i get that

#

so im stuck on STT

#

how does one fix

untold bane
#

given in thee abovve screnshot that STT section is missing that might be the issue

#

what were you using before? was it whisper?

weak knoll
#

i think it was just none as far as im aware

#

but i did take control via esphome so im not sure if that did something

untold bane
#

that has nothing to do with setting up STT

weak knoll
#

ok

#

i never changed it

#

it was on none

untold bane
#

when you call the device its looking to start a pipeline but sees there is no STT so it fails because it cant do anything with the speech

weak knoll
#

okay i get you

untold bane
#

the default option for STT locally is the whisper addon

#

which is then integrated via the wyoming protocol

weak knoll
#

okay i do have that

untold bane
#

if you look at your Wyoming integration page does it show whisper?

weak knoll
untold bane
#

ok so that is set up

#

you need to select whisper as the STT engine for the pipeline

weak knoll
#

done

#

faster-whsiper right?

untold bane
#

yeah

weak knoll
#

calm

#

and its just started showing lights

#

when i talk to it

#

🤦‍♂️

#

this thing has a mind of its own

#

not giving a response tho. rebooting it

#

im getting responses

#

it seems flip-flopping settings from updating to beta has fixed things

#

oh boy this has been frustrating

#

bar the duplicate WW and pipelines

untold bane
#

if you have 2 different pipelines set up you can use assistant/wake-word 2 to access a 2nd one

weak knoll
#

I love that

untold bane
#

if you set up a 2nd conversation agent you can use a different prompt so you can have a different personality

weak knoll
#

was just thinking that ahah

untold bane
#

I run a local LLM that has had most of its filters removed so i experiment with various personalities some might call "offensive"

#

I found a patrick stewart piper model recently and had captain picard personality semi working but the voice model was a bit crap.

#

lots of fun to be had.

weak knoll
#

I would like a somewhat offensive voice assistance aha

weak knoll
untold bane
weak knoll
#

(if it worked properly)

#

whats your best option for a locally hoested LLM to be abusive to me lol

#

ollama w/ ?

untold bane
#

i have a setup using ollama

#

its currently running huihui_ai/qwen3-abliterated:14b

weak knoll
#

Can you recommend any models

#

tyty

untold bane
#

i have been meaning to drop it to 8b to see if theres much difference. just so i can have another model loaded at the same time.

#

14b takes up most of the vram on the 5060ti inn the server ollama is running on.

#

i have a very dark sense of humour so my stuff gets pretty bad at times.

#

hence its gotta be local 😛

weak knoll
#

Love it

#

😩

weak knoll
#

Legend for your help. Should be alright from here. If it ain't working. I guess I just sit and wait for 10m for it to work itself out 🤣

untold bane