Projection mode?

[asleepynerd]

Exactly what the title says. How do we use the projection mode?

Changine the line win.loadFile('index.html') to win.loadFile("indexProjection.html");

shows the correct file, but animations and everything doesn't work.

[small-cactus]

After you change the line, rename the following files to the correct names in the order they are given:

1. renderer.js -> rendererDefault.js

2. rendererProjection -> renderer.js

[asleepynerd]

Should I create a fork for a completely local version of M.I.L.E.S? I already have it completely implemented, running either Ollama, or LM Studio, along with using either ElevenLabs for a TTS api, or a local one.

[small-cactus]

Yes that would be amazing, I’ll put the link to your fork somewhere in the readme so it can be found.

[Shadowphoenix]

Should I create a fork for a completely local version of M.I.L.E.S? I already have it completely implemented, running either Ollama, or LM Studio, along with using either ElevenLabs for a TTS api, or a local one.

when you have it up, please send me a link, i would love to give it a shot. That being said your profile is private so i cant see if you already posted it

[asleepynerd]

https://github.com/letmefocus/M.I.L.E.S/tree/elevenlabs

Heres the version with ElevenLabs. No additional packages are required, and i have added the voiceid and apikey into the setup menu so no editing of files is required.

I will post the local version either sometime today or tomorrow.

[small-cactus]

Looks awesome. You should probably change the models in the code to cheaper models.

gpt-4-turbo => gpt-4o

gpt-4-vision-preview => gpt-4o or gpt-4o-mini (same image processing costs and speed, vision preview is only used for image scanning in Miles)

gpt-3.5-turbo => gpt-4o-mini

I’ll update my main repo with these changes tomorrow, but in your forks you should update those strings, you’ll also have to update the system prompts and other strings that say gpt-4-turbo and gpt-3.5-turbo etc in the code. As well as the price comparisons. Here’s prices for you to make it easy, I tried to follow the format I already had them in, so it should be a copy and paste with minor changes.

gpt-4o-mini is roughly 2x slower but is 30x cheaper than gpt-4o, it’s basically free.

gpt-4o is faster but much more expensive, 30x more expensive than its mini counterpart.

[asleepynerd]

Well i no longer use openai for MILES since i ran out of credits, i've switched to using LiteLLM and Google Gemini 1.5-flash because of the speed and the cost effectiveness. But will do.

[asleepynerd]

I'm also doing a rewrite of the main.js file for M.I.L.E.S to be usable on a raspberry pi. My idea is you could have a raspberry pi running the "miles server" (hosting the microphone and the whisper+wakeword model), and have a seperate device (like an ipad) with the frontend. I might even move the microphone handling to the ipad and implement a websocket request to transmit audio data to the server.

[asleepynerd]

If needed, i can always contribute directly to the main repo if you want.

[small-cactus]

I'm also doing a rewrite of the main.js file for M.I.L.E.S to be usable on a raspberry pi. My idea is you could have a raspberry pi running the "miles server" (hosting the microphone and the whisper+wakeword model), and have a seperate device (like an ipad) with the frontend. I might even move the microphone handling to the ipad and implement a websocket request to transmit audio data to the server.

This is awesome, when you're done with it please drop a link to it, and also which raspberry pi to use, I'm definitely trying this when you're finished. It was one of my goals that I never completed so it would be amazing to try out.

[small-cactus]

If needed, i can always contribute directly to the main repo if you want.

Sure, that sounds great! You could add a provider selector to the setup screen, so users can choose the model provider they want to use for everything. That would be awesome!

But it’s a bit of a lot of work. You’d have to manage tool management for non-OpenAI models and local models, which is super tricky.

If you don’t want to do that, no worries! Just disable those features by default or comment them out so people can edit the file to enable them. It just has to be user-friendly and easy to set up.

But if you want to do that, pull requests for even the smallest things are totally cool with me. You could probably include a toggle for tools that disables itself when using local. You’d just have to make sure the user knows tools wouldn’t be available.

And if you’re only developing on Mac or only on Windows, please consider a platform toggle to make it disable certain newly added features if you only tested on one platform so it doesn’t break for other users. I can test it on both when needed later.

[Audiopicasso]

@letmefocus Hey, im trying to build a Alexa-similar device running MILES on a RaspberryPI with speaker and far-field-microphone attached to it.

Therefore it would be a blessing to use your RaspberryPi version of MILES. Unfortunately your profile is private, so i cant see if you already posted a functioning version or not.

Would you mind shooting me a DM or providing a quick update on that Topic ?

Thanks in advance

[asleepynerd]

I'm still in the progress of making the raspberry pi version but it should be done soon and I'll let you know.