Assist_satellite.start_conversation shows no target, what kind of magic do i need?

I have 2 Voice PE installed.
They have installed the latest version 25.3.0. This version has an entry " * Allow continued conversations if supported by the conversation agent in Home Assistant". I use the OpenAI integration, this integration works with the normal wakeword.

But the action “assist_satellite.start_conversation” in the developer tools shows no target in the ui.

What is missing here?

Regards,
Moses1234

The action currently works for VOIP only.

1 Like

I just resolved using conversation.process and agent_id

I knew this, but this solution is with wakeword and restricted to a few answers.

It requires code on the vpe device which isn’t there yet.

They really need to stop testing with the grandstream and move to primary on the vpe. :rofl:

If this was available on the satellite and not on VOIP I sincerity doubt anyone would complain and we wouldn’t see this question almost daily now.

2 Likes

The last Voice PE update 25.3.0 contains this " add support for continue conversation (#346)"
So why is this not working, if it is included in the actual version. There is something missing…

it says add support for - doesn’t say make it work now.

there’s a lot of pieces to the puzzle. As big as this feature is I’m sure we’ll ALL know the minute it even remotely works.

edit: look what literally just popped on one of my VPEs

^^^ That last line…

1 Like

It is quite easy to implement continue conversation, several conditions and flags allow you to do it. I’ve been testing the code on regular esp32 for some time now. For stable work on VPE it will require more effort, as more components need to be coordinated, but the team managed it, as we can see.

Start_conversation (which is a sub-variant of conversation.process) requires more complex code rework, as it needs an initial trigger. For a phone, the trigger is when the user picks up the phone, after you start transmitting data.
For VPE, a new sound action will need to be added, during which the system should expect a button press or wake-up word.

Perhaps an announcement action with a continued conversation (I have implemented this behavior) should give a similar result, since all data is in a conversation with an identical id.

For just chatting with the system and getting information it works fine. Management involves the use of cloud-based LLMs, which is not my area of interest. I hope Nathan will continue his research, it is quite useful and interesting.