Notes from the workshop.
Product updates, engineering notes, and thinking on privacy-first AI tooling.
The best on-device AI apps for Linux (2026)
On Linux, local AI isn't a setting you switch on. It's the default. The desktop picks, plus the self-host stack that makes your box a server.
The best on-device AI apps for Windows (2026)
Your PC hides an idle NPU or a gaming GPU. Which one you have decides your best local AI app. The picks, split by silicon.
What is quantization? How a giant AI model fits on your laptop
A 70-billion-parameter model is a 141 GB download. You run the 43 GB copy and can't tell them apart. That's quantization.
Run OLMo locally: the only AI model that's open all the way down
Every 'open' AI model hands you a sealed engine. OLMo hands you the factory: the data, the code, the recipe, the weights.
Run Phi locally: Microsoft's small models that beat far bigger ones
A 14B model you can fit on a laptop beat a 671B one at competition math. For Phi, small isn't a compromise. It's the design.
Government can switch off a frontier AI model. In June, one did, twice.
A letter went out at 5:21pm; two frontier models went dark worldwide. Cloud AI access is now a policy variable. The one kind that isn't.
Run Mistral locally: Europe's open family, and the license lines you can't cross
One command runs the best open code model in the world. Shipping what it writes breaks the license. Europe's open family, mapped.
Run DeepSeek locally: the reasoning model in a laptop-sized package
It wiped $600B off Nvidia in a day and you still can't run it. But DeepSeek bottled that reasoning into a 9 GB download. Here's which one your laptop handles.
Run Qwen locally: one open family for chat, code, vision, and audio
Most labs hand you one good model. Alibaba's Qwen hands you a toolbox (chat, code, vision, audio, search), and almost all of it is a free download.
Meta Llama, explained: which model is for what, and how to run it
It started the open-weights era and still sets the baseline everyone benchmarks against. Now its creator is hinting it might stop. The map, before that happens.
What is an embedding? How AI turns meaning into numbers
Spotify hands you thirty unheard songs and a third are keepers. The trick: it turns meaning into coordinates, then measures the distance.
What is prompt injection? The flaw every AI agent ships with
One email, no click, and Copilot mailed a stranger your files. The bug behind it can't be patched: the AI can't tell orders from text.
What is MCP? The standard that lets AI actually do things
A year ago your AI invented the numbers. Now it opens the file and reads them. The thing that changed wasn't a smarter model. It was a plug.
Production got free. Taste got expensive.
By April 2026, 44% of the music uploaded to Deezer every day was machine-made. When everyone can make anything, making it stops being the job.
What is RAG? How AI looks things up instead of guessing
Your chatbot just quoted a refund policy that doesn’t exist. RAG is the one-line fix: make it read the document before it answers.