Your thing looks amazing in black!
Modifications more than welcome.
I think i know an easy fix!
The front edge of the device Body should be filleted! Use a fillet finder to measure what the radius must be, then add the chamfer. It will make the device lock into the 3D print perfectly.
something like rails to snap divice in.
1 Like
I think here -
extend the back up in some way that makes short clips that snap into the holes the feet used to be in.
Part of it is the lack of anchor - but the other part is the actual support ends exactly behind the button so levers happen. If i hit the button juuuuust too high i push it right off the back over the top of the base. Extend its back support higher and create a snapin at the same time. Id do it myself by my slicer-fu sucks and is still very baby-level.
Yes indeed. Needs a lip or rail.
EDIT:
I have released a version two of the stand. It supports the back of the device now. Take a look:
Looks ok!
Old version is now a temporary phone stand:
1 Like
That is a very good idea!
Edit: it was so good, i just went ahead and implemented it.
You can do this directly inside freeCAD, the source file that the STL comes from is right there.
Edit: see above for new version.
Hi, I hope the feedback been received by the development team?
Question about PE:
Can I enable Bluetooth proxy on it without ill effects?
FYI: With not too many entities exposed, and qwen 7b thru ollama, i get pretty reliable home smarts.
No, the audio process is too real-time intensive for multitasking with Bluetooth. I believe there is a note in the docs about this.
2 Likes
I am struggling to change a number entity with Assist. Should it be basically possible?
I tried to increase a number entity with a sentence trigger an get this error
ValueError: ('Amount 0.5 not valid for number.walkingpad_c2_speed, it needs to be a multiple of 0.1',)
Taregt was to increase the speed of my walkingpad about 0.5 each time i trigger the sentence.
Only steps like 0.2 or 0.4 are working but not 0.5 although using the original remote changes the speed in 0.5 steps
I haven’t tried qwen 7b ollama model. Gonna test it, thanks for the info.
I see several qwen models: qwen2, qwen2.5, qwen2.5-coder. Which one to try?
This is not the first time I’ve heard it mentioned in context with HA… It’s supposed to be pretty good. I’ve got a box that should push 100 TOPS im trying to build right now for it. (Intel ARC based so I have to get through the it’s not a 40xx hurdle first)
2.5 non coder should do - you can prob fit the 14b model on something like a 4090 but the 7b would prob be comfy.
2 Likes
Plain qwen2.5:7b in my case, 16K context size 12 GB VRAM on my old 2080Ti so no larger model for me.
Cozy nonetheless.
1 Like
Create a script that does it for you, name it “faster walk” and then expose to LLM. Bam, solved. This is obviously a workaround for the LLM not being able to infer the valid increments, but it works.
You can also make an automation with a trigger sentence “faster walk”. This will execute faster as no LLM will be involved, and no need to expose the script either.
I played a bit with qwen2.5:7b, looks like it can turn on/off my lamps pretty good compared to llama3.2:3b model.
1 Like
Even 8b llama3:1 was absolutely dumbfounded by my smart home.
2 Likes
One thing I don’t like about qwen2.5:7b is that with response that it turned something off it always asks like “what else I can do for you?” 
Did anyone try this device with Sonos Era 100? If so, how does it work together? That means if they are connected via line-in, Sonos is playing Spotify, and you start talking to HA assist - does it switch to line in fast enough so you don’t miss first words of the response, or any other quirks?
1 Like