Steven Moon

Getting ready to onboard some brave souls onto PocketCloud -- privacy-first AI agents that run on your device. My goal is to democratize access to AI tools. Here's the gist: The problem: - Right now, $20/month barely gets you in the door with AI, and rate limits make it hard to do real work - Most developers actively using AI are paying ~$200/month at "Pro" tiers The majority of that money flows to a small handful of companies - Everything is a black box and getting visibility into what's happening under the hood is painfully difficult What PocketCloud does differently: - Capable of running completely offline once installed and models are downloaded all on your on hardware. - Telemetry, observability, and logging baked in at every level - Can tap into all major providers and their models when needed Workflows and logic are text-based Markdown files -- easy for humans and AI agents alike - Written in Swift 6.2, far more accessible than the Python-heavy alternatives Licensing: - Open source, licensed in the style of Linux - Business use: $5/machine/month - Personal use: low cost - Free for students, researchers, and those who can't afford it - Volume licensing for schools and governments Link to preview repo in comments

@DnuLkjkjh I’d love to jump on call and swap insights. As you know, just an insane amount of trial and error figuring out what works well and what doesn’t.

really interested in the Swift 6.2 choice — I build on-device voice transcription in Swift and the concurrency model in 6.x made a huge difference for managing parallel model inference without blocking the UI thread. curious how you're handling model downloads and versioning for the offline-first flow. that's been one of the trickier parts for me — users expect it to just work after install but some models are 1-2GB

Getting ready to onboard some brave souls onto PocketCloud -- privacy-first AI agents that run on your device. My goal is to democratize access to AI tools. Here's the gist: The problem: - Right now, $20/month barely gets you in the door with AI, and rate limits make it hard to do real work - Most developers actively using AI are paying ~$200/month at "Pro" tiers The majority of that money flows to a small handful of companies - Everything is a black box and getting visibility into what's happening under the hood is painfully difficult What PocketCloud does differently: - Capable of running completely offline once installed and models are downloaded all on your on hardware. - Telemetry, observability, and logging baked in at every level - Can tap into all major providers and their models when needed Workflows and logic are text-based Markdown files -- easy for humans and AI agents alike - Written in Swift 6.2, far more accessible than the Python-heavy alternatives Licensing: - Open source, licensed in the style of Linux - Business use: $5/machine/month - Personal use: low cost - Free for students, researchers, and those who can't afford it - Volume licensing for schools and governments Link to preview repo in comments

I have a default list of 300ish huggingface models and then there is code that attempts to decide what models are a good fit for the hardward (current works well on macs, iphones and ipads. There is a benchmarking process that can be run to assist with initial setup. Then benchmarking stats are keep in real time which are tracked to help an "AIRouter" decide how to route "task/commands" etc based on real results.

swift 6.2 is a great choice for this. i built my on-device transcription app entirely in swift and the strict concurrency model saved me from a class of async audio pipeline bugs that would've been brutal to debug otherwise. curious how you're handling model downloads — are you bundling or pulling from a registry on first run?

I’m in Denver interviewing for a few jobs, and I decided to swing by the OpenClaw Global Unhackathon downtown. Honestly, I loved it—just sitting in a room full of builders, watching what people are dreaming up. Two folks were playing with this open-source robot from Hugging Face called Reachy Mini. Go watch a couple videos if you’re curious, but picture a small, stationary robot you’d put at a receptionist desk or in a hospital room. It can see, hear, and talk/make sounds—and it comes with built-in expressive “personality” stuff. Like, you can trigger reactions such as scared or sad, and it layers these little lifelike behaviors on top of whatever AI it’s running. It’s been a long time since my brain got flooded with that many dumb (in the best way) business ideas all at once. What really hit me: to test their “apps,” the developers would just… interact with the robot and see if it behaved the way their code intended. Basically, it felt like wrapping a real, physical interface around ChatGPT (and more). Pretty wild—and kind of eye-opening for what’s about to show up in the world.

Fun find of the day. I just learned Claude Code has a folder ~/.claude/plans/ that gives random names to the plans it generates. I'm curious what random names other people are seeing? Here's mine:

@TheTrueVanguard ⬆️⬆️⬇️⬇️⬅️➡️⬅️➡️🅱️🅰️▶️

Prove to me you’re an old gamer in one sentence. I’ll drop a like if I’m convinced.

@elonmusk This is the obvious next move. Solar/Nuclear are way more efficient in space. One of the biggest limitations on compute GPUs/CPUs/AIChips/etc has always been being able to cool them. Space opens up a whole new frontier in computing.

Interesting

@BroberDave I so excited for Saturday. LJ Martin - healthy + Bear - more experienced + Master of chaos himself Jay Hill take 2. + The insane amount of national disrespect for a team that thrives as an underdog = They don’t know, but they’re about to find out Let’s go Cougars!!

Stoked for LJ. What a year he has had.

Elite Tweet by the official BYU Cougars app. They have been spitting 🔥and taking shots the last few days. Love it!

LFG BYU Cougars!! I have been a bleeding blue fan for 25+. All I have ever wanted was a good team with a legit path. I watch some amazing teams but prior to getting into the Big 12 a BYU team would have needed to be undefeated and then maybe. Now? Win and advance! LFG!

@tay_Dougie @BamBamsBBQ @espn Is it? I think there is enough grounds to at least file a class action on behalf of fans everywhere. The thing to remember is the fan bases are the revenue source. You’re right we have a path and I am beyond grateful. But let’s not pretend shady shiz isn’t happening.

@stevenmoon @BamBamsBBQ @espn This is ridiculous. Just beat Texas Tech. At least make it competitive.

What does @espn have against BYU? I’m actually asking. Every time I watch any other network BYU gets a lot more respect. ESPN really seems to go out of their way to make us disappear. #lawyerup

I remember when the Big 12 was formed and have watched as many Tech games as possible over the years. I am so excited to see them get the opportunity to play in their first Big 12 championship. Watching them play and have so much success this year has been so fun.

As a random side note I was born and raised in Lubbock. Grew up a die hard Tech fan. Attended my “lost” freshman year at Tech. During that time I joined the Church and soon after went on a mission to Salt Lake City and have been an Utah transplant ever since.

So excited for @BYUfootball Those who know me know I bleed blue. I have been to 90% percent of home games over the last 25 years. At my own expense my company developed and supported the original BYU Cougars iOS and Android app back in 2011.

@kelli_BYUFan This whole angle being pushed is the biggest nothing burger about the story. The Church Education Board which is chaired by the Prophet and has many members of the 12 involved was always going to have to approve any deal even when no Church funding is involved.

You mean the Apostle who attends games, interacts with players & oversees BYU Athletics as a member of the Church Education Board