#Error talking to open ai

1 messages · Page 1 of 1 (latest)

ruby cloak
#

Hello so when I use the entity of open ai conversation it works perfectly, but when I try to use it through voice assistant all I get is error talking to openai

gentle salmon
#

Anthropic as well

south coyote
#

I got the same issue with OpenAi when upgrading to 2026.1 of Home Assistant. The logs for me showed an error like this:

"Invalid schema for function 'HassStartTimer': schema must have type 'object' and not have 'oneOf'/'anyOf'/'allOf'/'enum'/'not' at the top level."

I rolled back to the previous version to solve the issue for me, as it looks like a problem that can't be solved by config.

ruby cloak
#

yeah fixed it for me too, thanks man 🙏

pallid arch
#

I have the same issue, unfortunately I can't roll back so it looks like no OpenAI for me till it gets fixed. ChatGPT says this was a fixed issue so it looks like some old code got into the latest release (or ChatGPT is hallucinating again 😆 )

tawdry jackal
#

Hi not sure if this helps but I got around it by unchecking the assist option in the 'control home assistant' option.

97F71FAB-E522-4369-90B5-A8FD4E585F66.png
gentle salmon
#

That’s fine but then you can’t have it do anything

tawdry jackal
#

It will still follow some commands but you have to ask it directly. If you start a conversation with it then it won't do HA stuff.
I'm not sure what the issue is but something happens when the assistant is checked. Hopefully someone smarter than me knows what that means and to solve it 😅

rare jacinth
#

It's a known bug, think it's supposed to be fixed in core HA soon

ruby cloak
#

yeah it's an issue with the new update, also why would you want llm no assist

short scarab
#

Yesterday Paulus wrote:

Let's revert and include in patch tomorrow so we are at least not broken in production and pressure is off.

Is this for the 2026.1.1 minor release? When do those minor releases roll out?

mystic mortar
# short scarab Yesterday Paulus wrote: > Let's revert and include in patch tomorrow so we are ...

The Revert PR is set for the 2026.1.1 milestone.

In general the x.x.1 release is on the Friday after the release. which would have been yesterday.
sometimes they are delayed for various reasons. with CES going on and some other stuff the team is a bit thin on the ground compared to usual so this is likely one of the causes of the delay however i am not totally sure. Ill ask people more important than me though and see if anyone has any idea of the timeline.

mystic mortar
pallid arch
#

unfortunately I can't roll back so some of my stuff aint working 😭

fallen marlin
#

My issue is probably related but I get streaming tts errors suddenly with it saying "pipeline is busy" asking any question or giving command causes it to take forever. Using ollama though. I'm about to the point of taking ha completely offline and keep it in maint mode. The devs seem hell bent on staying relevant against homey and wanting things "easier" that they aren't debugging much before they throw out releases. Or at least that's what it feels like. With the amount of paid employees at HA now, there really is no reason how this was overlooked. Not whining as I get things happen but the frequency of it lately is really getting annoying.

merry rover
#

I rolled back to 2025.12.5 and still have the problem. Switched to Homeway Sage and it works fine. I'll cross my fingers for a quick release of 2026.1.1 .

ashen drift
#

2026.1.1 doesn't fix this. I had to rollback again.

merry rover
#

Damn. It's so weird. I can click on assist at the upper right of the screen and it works, but if I click on assistant on my dashboard, or talk to my voice assistants, it won't work.

gentle salmon
fallen marlin
#

i've noticed the stop command not working either. it's like the voice pe isn't negotiating properly to the llm

#

Something is borked bad here

#

This isn't just openai. I've tried the latest beta firmware via esp web and everything and i get incredibly delayed responses and delayed tts through sonos. This was never an issue before. I may have to downgrade HA and just wait. But yea none of the beta firmwares which supposedly address this are working

pallid arch
#

Trying again with OpenAI crashed HA and forces a restart for me... so worse

fallen marlin
#

But what i am seeing is a very delayed update from when voice pe switches from idle to listening to responding. The device itself lights up upon listening for command but the log is taking forever to update

#

Shit where can i find previous HA ver. my backups already overwrote it

#

well doesn't matter actually, i'd have to start from scratch. damn

#

yea not just us lol

#

changes coming

#

I think this is the root cause of a lot of it

#

the issues addressed there are basically everyone is having

#

but the changes won't be out until 2026.1.2 which i hope is quick as hell because i'm fked until then

gentle salmon
#

Yeah, I’ve made too many changes to lose to revert to before it broke. Hours of tedious cleanup that I would lose.

fallen marlin
#

actually i did manage to revert but not revert

fallen marlin
#

technically you're updating to a previous version so nothing is lost

gentle salmon
merry rover
#

I rolled back to 2025.12.3 and mine is still not working. If I set the conversation agent to OpenAI Conversation I get the error from the Dashboard as well as my voice assistants (Error talking to OpenAI). If I sety the conversation agent to Sage (ChatGPT through Homeway addon) it works fine.

merry rover
#

So here's something odd. I went into OpenAI Integrations and created a new service. I named it OpenAI Test, I used the same model as my original service GPT-4.1-mini. Then I set my Voice Assistant's Assist Model to use OpenAI Test, and it's working. Hope this works for someone else. I changed back to the original conversation agent and it was broken again. So... Settings-Devices and Integrations-OpenAI-Add Conversation Agent (name it something Unique) Then Settings-Voice Assistants-(choose your assist model) then change the conversation agent to the newly created to the new one you just created.

#

Now I'm going to update HA to 2026.1.1 and cross my fingers

merry rover
#

well... updating back to 2026.1.1 broke it again. I give up trying to figure out a work around.

#

the above only works apparently if your reverting didn't solve your problem.

fallen marlin
#

i'd wait until tmrws patch. hopefully it's addressed

merry rover
#

Has anyone confirmed whether 2026.1.2 fixes the problem with OpenAI? I read the notes and there's no mention of the issue being addressed.

gentle salmon
#

It is fixed for me.

lavish dagger
#

thanks guys I'll also update and test

#

as an aside is anyone using a custom openai compatible LLM?

merry rover
#

didn't fix it for me.. rolling back again.

wispy tendon
lavish dagger
#

also happy to pay any consultantant for initial setup if any freelancers want to tackle!

wispy tendon
#

As for the bridge I'm using it was custom made by a member of their discord specifically for 1min.ai

merry rover
gentle salmon
#

For me it is. I had issues with Claude and OpenAI. Now it’s resolved.

wispy tendon
#

It looks like Jekalmin is also now addiing the AI Task ability now in the one I use, I got an email this morning about it from the discussion there. which I think would make it easier to speak what you'd like to build to your Voice Assistant. 🙂

merry rover
#

Confirmed.. Extended OpenAI Conversation is working with 2026.1.2 !

#

Well, that didn't last long. I get a long winded message about some bloody error, and it's broken, even with extended OpenAI Conversation. This is using my voice assistants... it works fine if I use the dashboard icon.

merry rover
#

Changed the model to gpt-4o-mini and everything is working now. It seems the model is the problem, they're making some of the models obsolete or something.

lavish dagger
#

Annoying, im pretty comfortable with 5.2 is it still broken even with temp adjusted for you @merry rover

merry rover
ashen drift
#

I upgraded to 2026.1.2 and the error still persists. The only way that fixed is selecting "Recommended model settings" at the OpenAI configuration. It prevents selecting the model and adjusting the parameters, but it's a temporary solution (I hope). 😉

merry rover
red mist
merry rover
red mist
modest hatch
ruby cloak
#

has this been fixed 😔

merry rover
modest hatch
#

Thanks--I did that last week and recently had the "Error talking to OpenAI" error again, but last night it started working on its own

red mist
#

@merry rover @modest hatch @ruby cloak
it works!
I'm in 2026.2.1 and gpt-4.1-mini or gpt-5-mini is working now for me !

merry rover
misty briar
#

Antrophic was having a similar issue that seems to have cleared up in the recent 2026.2.x releases