kortexa.ai

[Franci Penov @francip]

Porch - the little sidecar app for your @openhome devkit, just got JSON Render support so your abilities can show rich visual UI in addition to the voice message.

[Franci Penov @francip]

Just finished @NousResearch Hermes agent gateway support for @livekit, working with both Livekit Cloud and a local deployment. You can now have a webrtc call with your agent. Well, after the PR lands, that is... :-))

Sadly, it seems the sandbox app doesn't support avatars properly, so my agent is just a faceless pawn.

But I am so getting my agent a phone number through Livekit Cloud and calling Avery on the phone. Maybe even have her call me instead? :-))

[Franci Penov @francip]

[sigh] Do not give me new toys... It'll not end well.

[Franci Penov @francip]

KRTX-AI radio skill kicks ass on @openhome

Time to polish it and ship to the community 🚀

[Franci Penov @francip]

Walking sound created by IMU encoder + audio decoder combo. Baby knows how to make sounds as its IMU status changes...

Both the IMU and audio were trained separately without knowing anything about each other. And yet, I can combine them on the fly pretty easily.

[kortexa.ai @kortexa_ai]

RT @francip: Qwen 3.5 0.8B now understands its movement and orientation.

[kortexa.ai @kortexa_ai]

@mrmrs_ Lovely! Would be nice if we could search by a base color maybe?

[kortexa.ai @kortexa_ai]

Llama Lab IDE has a new icon. Also, bunch of improvements and new functionality coming later tonight.

[kortexa.ai @kortexa_ai]

Github repo: https://t.co/2kaHfiaWey

Feel free to use (it's very alpha right now), contribute, clone, fork, or print out on A4 and tape to your window.

[kortexa.ai @kortexa_ai]

We're open sourcing v0.1.0 of the Llama Lab research IDE. Github link below.

[kortexa.ai @kortexa_ai]

24x7 music. 0 pay-per-token.

[Franci Penov @francip]

Introducing KRTX-AI

Inspired by @eltokh7 radio WRIT-FM, I decided to vibecode a small radio of my own.

Tech stack:
- DJ - Qwen 3.5 4B
- TTS - Qwen 3 TTS
- Music - ACE-Step 1.5

All running locally on my Mac Mini M4 Pro in real time.

https://t.co/7mLeMszDEB

[Franci Penov @francip]

@AkshayNarisetti and the team put so much work in this little device and packed tons of functionality in a beautiful form factor.

[kortexa.ai @kortexa_ai]

Should we open source this tool or just open access it?

[Franci Penov @francip]

Hacked some small stuff last night. Needs a bit more work, so internal access only for now. But the idea is that this is a fully in-browser "situation monitoring" console, complete with an AI to help understand stuff.

[Franci Penov @francip]

FastAPI wrapper around Qwen3 TTS using mlx-lm. This is another one in the series of local servers I use to switch quickly between different models and cloud/local for the various parts of my pipeline.

Qwen3 TTS is not as good as Cartesia or ElevenLabs, and the of the nine voices it comes with out of the box, only one sounds good in English. But, it's running locally and it's free, so there's that.

Besides, they did ship the VoiceDesign model, so I am definitely going to play with some custom voices and/or clones.

https://t.co/F3A0PaUCof

[Franci Penov @francip]

The FastAPI wrapper for ACE_Step 1.5 I use to run it locally. My radio station app is running against this service.

https://t.co/3jA433NZWF

[Franci Penov @francip]

This... is genuinely fucking good!!!

Maya Angelou's "Still I Rise", aggressive pop-rock, female singer, guitar solo

ACE-Step 1.5 + Nanobanana

[Franci Penov @francip]

Qwen 3.5 08B in browser using transformers.js v4 and webgpu/

https://t.co/on3u1uqtuU

[kortexa.ai @kortexa_ai]

@BosonJoe I am still debating whether I want to release all the free tools as open source, but here are the sources for this one - https://t.co/mzf82X5G0V

[kortexa.ai @kortexa_ai]

Another single-shot tools - https://t.co/dSKK7UgTyq

No ads, no subscriptions, no cloud, no analytics, no tracking. Your QR codes, in your browser, on your machine.

[kortexa.ai @kortexa_ai]

If you still think AI has no impact on software...

#1 - a day to lay up the build and deploy pipeline.
#2-#4 - half a day to polish the skills.
#5-#6 - about two hours.
#7-#8 - half an hour to Coming Soon, got distracted..
#9 - one hour, mostly to get Transformers.js v4 up.

[kortexa.ai @kortexa_ai]

By now the factory is fully setup. We are churning free tools #6 - #8 in the lineup.

Most manual work is literally registering and setting up the domain, and then testing the final product. Everything else is fully automated.