Portland, OR
Joined December 2006
Mattt retweeted
Pain is information As a child, you touched something hot, and it burned you. That pain gave you a piece of information: be careful touching hot things. When you sign up to run a marathon, you are signing up for pain. But whether or not you keep running is up to you. In Haruki Murakami’s words “pain is inevitable but suffering is optional”. You can choose pain without choosing suffering. Pain is information, and information is painful. Rewiring your brain is not a frictionless process. Some knowledge can only be discovered the hard way. The sooner you can convert pain into knowledge, the sooner you can experience the next useful pain. Don’t put yourself through the same pain too many times. To access new information you must experience new pains. They say “knowledge is power”. If pain is information, then pain can be converted to power. To do so, you must learn to control your suffering, accept that scars are beautiful. If you felt pain, ask what information it gave you. If that information was useful, seek the next pain. You are learning.
29
131
16
1,221
122,851
AnyLanguageModel turns one, and it's 1.0 🎂 Our drop-in replacement for Apple's Foundation Models framework lets you choose the model and provider your app uses. Here's what changed this year, and what's next: huggingface.co/blog/anylangu…
2
4
37
3,452
Hosted providers now include OpenAI, Anthropic, Gemini, Ollama, any Open Responses endpoint, Private Cloud Compute on OS 27, and Hugging Face Inference Providers. AnyLanguageModel runs on Linux, too, so the same code can work on a server.
1
2
1
900
Along the way, we also built swift-huggingface and swift-xet, for accelerated model downloads from 🤗 Hub. Thanks to everyone who contributed this year, and to @pcuenq for his help with all these projects. Excited to share more projects soon, so stay tuned!
1
530
Mattt retweeted
Beam is a token-efficient, frontier-scale model for your coding agents, comes with its own harness 🔥 A23B/501B model, also comes with NVFP4 & FP8 and day-0 transformers support 🤗 check out our blog ⤵️
Beam is now available across platforms, with open weights under the Apache 2.0 license - in collaboration with an ecosystem of 40+ partners. We're releasing: - Weights, tokenizer, and reference implementation - BF16, FP8 and NVFP4 weights and a DFlash2 drafter - The open source harness we built and trained Beam on: Mirror CLI Weights and model card on Hugging Face: huggingface.co/reflection
12
9
112
5,335
And of course Muse Spark Open works great with llama.cpp 🦙🔥 Behold this amazing demo courtesy of @mervenoyann, running on a Mac (M3 Ultra 512). This is a frontier model that runs on your (beefy) desktop. Local AI is only getting better 🚀
4
4
1
22
4,793
iMCP 1.7.0 is here, with a new Home service for HomeKit, Voicemail, and editable Reminders and Calendar events. Good timing, with Apple announcing HomeView yesterday — now your Mac's assistant can help run the house, too 🏠 github.com/mattt/iMCP/releas…
7
12
4
202
17,620
Also in 1.7.0: Voicemail lists visual voicemails and fetches transcripts and audio, Reminders can repeat and manage lists, Calendar gains events_update, and Messages can read attachments. Thanks to @bradhallett for much of this!
1
1
810
iMCP is free and open source. Existing installs update automatically, or grab it fresh: brew install --cask mattt/tap/iMCP Can't wait to see what your assistant does with the house keys 🔑 github.com/mattt/iMCP
5
803
Mattt retweeted
if you want to get every last bit of performance out of your local setup, you need to know a bit about inference 🥵 but we've got you covered, shipping conceptual guides 🔥 here's the first one about prefill, decode, KV cache what you should optimize for 🙌🏻
24
63
3
440
18,158
Thanks, Julien! Microduck turned out to be the perfect desktop companion. Pollen Pet is now open source. For now it's mostly the puppetry, but I had a blast hooking it up to Qwen3.8-27B. Can't wait to see what y'all build with it! github.com/mattt/PollenPet
Clippy and ChatGPT Pets << Microduck This creature needs to be integrated into our lives at the OS level kudos @mattt
19
5
3
53
15,677
Alert: Apple just dropped a new model on Hugging Face. It's a Qwen3.5-9B finetune that turns long documents into small page images to save tokens, then pulls up the full text of only the pages relevant to your question 💡 huggingface.co/apple/LensVLM…
74
276
50
3,452
284,200
What I like about Jev: for years, large generalist models sucked almost all the oxygen out of AI. But there are huge opportunities in much more specialized and customized models, built for specific tasks and languages and as a result, orders of magnitude cheaper, faster and more optimized. There are 3 million of them publicly available on @huggingface. Let's build a much more diverse AI ecosystem!
82
95
12
1,063
57,942
When my coding agent starts a long job, I usually end up scrolling Hacker News. I wanted a smaller distraction, so I built trifle.day It's daily programming trivia, served by your assistant over MCP while it works. Like Wordle, everyone gets the same question.
11
5
2
75
12,775
Back in the day, I ran @NSHipster pub quizzes at meetups and a few years of WWDC. I'd been going to weekly pub trivia at the Ginger Man in Austin (RIP), and wished they'd ask about obscure corners of Cocoa and Objective-C instead. Trifle is that quiz, one question a day.
3
6
1,410
Once it's connected, your agent can also offer the quiz at a natural pause (e.g., while a build or tests run) and it asks before starting. Give today's question a try, and let me know how you do. There's always room for a little trifle 😋
1
2
949