Archive

Run open models on the GPUs you own, with native Metal and CUDA inference.

Run a fleet of coding agents while keeping branches, reviews, and CI failures manageable.

Every project gets a main agent. Tell it what you want built and it plans the work, starts a worker for each piece in its own worktree, and comes back only when something needs you.

Sessions place themselves from live facts: what the agent is doing, its pull request, CI, and reviews. You open what needs you and leave the rest alone.

LLM serving, synthesized for your Mac.
Up to 2.7× as fast as the best other server we measured on the same Mac.
We used SkyDiscover to build specialized engines for a 24 GB M2 and a 48 GB M4 Pro. Each engine targets a specific model and Mac configuration, but we benchmarked all of them on both machines.
Set your Mac and memory, then choose a model below to get its commands.

Nix has been single-handedly the best thing I have added to my dev toolbox when I found out about it 6 years ago. More so after AI.
Yes, the learning curve is (Was?) hard. But these are complaints of the past. With agents, you have everything you need now to work with it. -- [X](https://x.com/ahmedelgabri/status/2119373110411452618

I have been using NixOS on my RTX5090, without writing a single line of nix config by hand. You can see see it here -- https://github.com/btbytes/nixos-dotfiles

Auto Ontology is a Python framework for building semantic, graph-based reasoning layers over data. It enables deterministic query generation, better context retrieval, and more explainable AI for LLM and agent workflows.

looks interesting. the frontpage copy is too claude-y.

maybe there are just too many software like this now?

by the creator of MyGo -- Write native GUI apps in #golang. See also https://nsh.dev

Community created roadmaps, guides, and courses.

The poetry of a certain South Indian childhood means that you have bathed in at least three waterfalls and been blest by more than one elephant. You know with a knowing that predates language: the scents of jasmine, of camphor, coconut oil, and filter coffee. Know them the way you know the particular sound of your mother’s bangles. The way you know the sound of the latch on your front gate, and the sound of wet laundry slapping stone. You belonged to an off-key choir of schoolchildren who chanted morning lessons in unrecognizable English and ear-splitting unison. Your to-go meals were eaten aboard trains and came wrapped in banana leaf and newsprint, neatly secured with twine. All your uncles rode motorcycles.

-- Pavithra Mehta

See also Part II

Build a small eval on your own data to choose a Claude model by accuracy first and cost second, fix the prompt before you upgrade, and keep checking after launch.

Supercharge your AI agents with data from the web and beyond. Building the library for superintelligence. 🔥

Desktop apps in Go. Tiny, fast, fully typed. One small binary for macOS, Windows and Linux. Native UI or a web frontend in Go, with installers and auto updates included.

The motivation:

What I want, alongside Claude Code, is a coding agent that:

• Does what I tell it to do. Sandboxing is critical, but it’s something the operating system should enforce, not the coding agent.
• Runs locally, even offline. My code stays on my machine, there are no usage limits, and I can use it whenever I need it.
• Gets the job done. A local coding agent has to get real work done.

Similar tools are: Agentsview, MCP Agent Mail, Entire, and Specstory

OpenAI- and Anthropic-compatible APIs, GGUF + safetensors + MLX, FP8/NVFP4/MXFP4/Q8/Q4, built-in Studio

look into this to borrow ideas to kaas