thread of current projects among our members for August 2024
#What is everyone working on?
dense hemlock
Our early bird proposal was to make a proof-of-concept local AI version of a demo from Rosenfeld Media's Designing with AI conference: https://llewynpaine.consulting/avatars. We used tools like Wonder Studio and Elevenlabs to redact human bodies and voices from research videos in the demo, but it should all be able to be done locally if you really want to preserve privacy. The recent SAM 2 launch might have been a great fit!
My later proposal was to build a POC to allow local AI to run against locked-down hardware. Lots of workers from big enterprises to boiler rooms to secretive startups don't have access to USB drives or printers, have restricted network access, can't install applications, etc. They'll never be able to access online AI or run local AI. But even locked-down systems need to let you use keyboards and monitors, so why not have a hardware configuration that lets a local language or diffusion model appear as a keyboard and monitor to another computer? Support taking input by scraping the mirrored or extended screen, and respond by "typing" directly into the locked-down PC. OCR or multimodal models would have been real interesting here.
hard estuary
Happy to chime in. I received the rejection email today as well. I'm part of a small team of Swift and Python developers (one ex-Apple) creating Swift native tools for Apple's MLX.
We noticed that the quality of open source and local LLM apps for iOS and macOS is lacking the professional ones (e.g. ChatGPT and Claude). One reason is the lack Swift native tools. Even Apple's own MLX-Swift is incomplete.
This means local AI apps still have Python dependencies which makes not suitable to distribute through the App Store, harder to use/install, and even impossible to run on iOS.
Our proposal is called MLXKit, a 100% Swift-native abstraction layer for MLX that eliminates all Python dependencies and enables ChatGPT-quality apps that use local AI.
Part of the proposal was to 100% local vector database and local embeddings generation.
So if you're a Swift developer eager to create high-quality local AI apps, follow/star us on https://github.com/picoMLX while we figure out how to bootstrap the project
Edit: We also host a weekly livestream on Twitter and YouTube about Swift and AI. Every Friday around noon PST on https://twitter.com/rayfernando1337/ We previously had Chris Lattner (Swift creator), Rick Lamers (Groq), and the other members of the Pico MLX project on the show. This Friday our guest will be Lou Zell, the creator of AI Proxy, a multi LLM proxy for iOS and macOS apps.
Twitter
raven vessel
Happy to chime in. I received the rejection email today as well. I'm part of a s...
I got to catch one of your streams the other day, great! Let me know if you ever would like to host one on the Mozilla AI stage via our member events. 🙂
hard estuary
I got to catch one of your streams the other day, great! Let me know if you ever...
That's great you listened to one! We actually reached out to Justine for an interview a few times but we never heard back. How would the AI stage work? Can we stream to stage?
raven vessel
That's great you listened to one! We actually reached out to Justine for an inte...
Sure, this is part of what I'm working on this August so hopefully still relevant to the discussion. I hope to help give more exposure to people working with open source ai and inspire more collaboration, and idea exchange.
For more details take a quick read here: https://mattcool.tech/posts/member-organized-events-in-discord
stray pulsar
AI-at-Work is an project dedicated to simplifying the development of AI agents through a comprehensive chat management system. This suite of services provides developers with the tools they need to focus on implementing AI agents without getting bogged down in infrastructure management.
mint tulip
Hi everyone! 
Since my indie iOS projects about local ML are not selected for builder. I am sharing it here:
- Featured on Tech Crunch, a Local ML powered Fediverse app that brings a familiar experience like Twitter but with all algorithms running locally:
SoraSNS uses local ML to build weights on interests of the user and stores it locally. By using natural language analysis to analyze posts, Sora can curate a timeline with posts user likes to see.
It supports Mastodon, Bluesky, Misskey all in one.
Already published on App Store https://mszpro.com/sorasns/
- Llama-3.1 powered Matrix chat client
Adding chat summarization and reply suggestions to Matrix. And provides a modern interface for Matrix chat.
Already published on App Store https://mszpro.com/nil/
I am also working on indie developing many other things, including VisionOS games, AI powered Bugzilla client, and many more.
Congrats to those who got selected! I might try again in the future 
Happy coding everyone!
crystal tundra
I am working on a Nvidia sass assembler for Ada Lovelace (sm_89) and will be using it to write sass kernels as the backend for an inference library. The main idea is to get full control over your hardware as if you're using cuda/ptx you don't have control over register allocation (similar to using compiler intrinsics vs writing asm).
not an application of local AI but something more fundamental in that sense
dim trench
Who is selected for the builder project?
dim trench
Is the Mozilla has Linkedin channel?
raven vessel
#general message
hasty hawk
I am working on the following
Midori AI Subsystem
The Midori AI Subsystem is a robust platform designed to optimize AI workload management. Built on Docker, it provides a modular approach by isolating each AI system in its own container. This architecture ensures consistent deployment, simplifies configuration, and enhances overall system stability. Ideal for managing multiple AI programs.
PixelArch OS by Midori AI
PixelArch OS is a minimalist Arch Linux distribution fine-tuned for Docker environments. It's offered in four modular levels: Quartz, Amethyst, Topaz, and Emerald. Each level caters to different developer needs, from basic installations to advanced setups with preinstalled tools and remote shell support, making it perfect for both lightweight and intensive AI development workflows.
Midori AI Subsystem - https://io.midori-ai.xyz/subsystem/manager/
PixelArch OS - https://io.midori-ai.xyz/pixelos/