lewq

Hi, author of the post here. Just fixed up some formatting issues from when we copied it into substack, sorry about that. Yeah, I used Opus 4.5 to help me write it (and it actually made me laugh!). But the struggle was real. Something I didn't make clear enough in the post is that jpeg works because each screenshot is taken exactly when it's requested. Whereas streaming video is pushing a certain frame rate. The client driving the frame rate is exactly what makes it not queue frames. Yes, I wish we could UDP in enterprise networks too, but we can't. The problem actually isn't opening the UDP port, it's hosting UDP on their Kubernetes cluster. "You want to what?? We have ingress. For HTTPS"

Join our discord for private beta in January! https://discord.gg/VJftd844GE

(This post written by human)

Yes

Interesting, thanks!

Fascinating, wanna compare notes on a call some time?

Hey! Yeah we are working with partners on fully integrated hardware+software stack for this. We particularly like the RTX 6000 Pro Blackwell chips for this

Vision language models have been trained on how to operate human UIs though, so at least for a while, computer use will be an interesting area to explore. I think debugging web apps and building UIs is a particularly fruitful area for this

That's the next move :-D

I should have said creator. He seems to have a healthy community backing him, but we should ask him!

There's also value in being able to run multiple agents in parallel with their own isolated filesystems and runtimes. One agent won't tread on the toes of another whatever they do. You can let it loose and it doesn't matter if it breaks something, you can just spin up another one

Mainly so you can give the agent access to the desktop as well. Then it can debug your web app in Chrome Dev tools but also you can pair with it with streaming that is so good it feels local

Itself, more recently

I suppose but we got it working, and the primary interface is webrtc in the browser, and going via moonlight internally is just an implementation detail that got us here quickly. We are open to refactoring in the future of course :)

It's nice for unmodified moonlight clients to be able to connect - they have tons of them, you can even run it on a Nintendo DS

Wolf now supports multiple clients connecting to the same session via the wolf-ui branch that landed recently. After lots of stability work we are now running that mode in production (and in the latest release) https://github.com/helixml/helix/releases/tag/2.5.3

Join our discord for the beta https://discord.com/invite/VJftd844GE

Yeah that's what we did :) https://youtu.be/vVmnpcnLDGM?si=b6LxW6lmM7843LY0

Author of helix code here. Here's a demo of the full system working. https://youtu.be/vVmnpcnLDGM?si=b6LxW6lmM7843LY0

We're opening the private beta where we provide a hosted environment for testing, or you can install the latest Helix release and run the installer with --code to try it on your own GPUs

Docs here: https://docs.helixml.tech/kodit/

Yeah but then you have to fly Ryanair

Every. Day

Now you can create GPTs that run locally and version control them properly with GitOps, why would you send your data to OpenAI?

Launching this today. We're working on making it insanely easy to run open source models and get all the benefits you might expect from a hyperscaler solution like OpenAI: RAG, API integrations, fine-tuning but all on open source models, all running locally on a single 3090, or scaled in production on Kubernetes in your own private VPC.

This spec is an attempt to will a sensible, infrastructure-independent, declarative abstraction for LLM application patterns into existence

I'd love critical feedback and for more folks to get involved

Hi HN, we're bootstrapping Helix.ml. We're busy building out the core AI apps platform, but we're also available for hire to help people develop their own AI apps.

We've got experience deploying Ollama in production (HA) on Kubernetes, connecting it to API calling, plugging in RAG with pgvector, fine-tuning and so on. So I just thought I'd see if anyone is working on something (or planning a project) and could use our help?

For use cases from SMS chatbots for construction, to automatically extracting themes from marketing survey data and presenting them in a web app flow, we’re helping clients deploy GPUs, configure data sources, set up evals frameworks, and build interface layers to get their LLM projects rocking. We can work with your existing app teams or build projects from scratch.

Do you have an AI app you’re trying to get off the ground? Hit reply or email luke@helix.ml and tell me about it. I’d like to figure out if we can help you build it faster :)

Deck: https://docs.google.com/presentation/d/11bBUP8gBekmI7GkwvGdr...

Check out https://docs.helix.ml/helix/develop/getting-started/

But what I think is really interesting is the ability to define a helix app yaml like:

https://github.com/helixml/example-helix-app/blob/main/helix...

Then version control it and deploy your updated LLM app with a single git push. LLMGitOps?

oh, sorry, I didn't know that was a restriction. We legit wanted to show HN what we had built :-)

The product is ready - we can let people in straight away. In fact, secret tip, if you go to the searchbot docs, you can find the URL of the live app :-P

But we'd still like to chat to our users, hence the waitlist. We're setting up a slack channel for that. We'd love it if you'd like to be involved in trying it out and giving us feedback.

Not mentioned on the website (because it’s targeted at general website owners rather than a technical audience) but we are using a 100% open source AI stack for this, with llamaindex, pgvector and llama3:instruct running on ollama hosted on a stack of GPUs we have mostly in our houses.

Does it work with ipu6 webcams yet?