llama.cpp now ships a built-in GUI — run local LLMs, no terminal
The most-used local inference engine (123K stars) now boots a web chat UI by default and auto-discovers Pi — a fully offline dev stack is a one-liner.
Copy markdown
The GUI ships by default now
llama.cpp was terminal-only; now `llama-server` boots a built-in web chat UI at localhost:8000. The most-installed local inference engine (123K+ GitHub stars) is finally usable without memorizing flags.
One command, then a browser tab
Install via a single curl script (or Homebrew/Nix), run `llama-server`, and open localhost:8000. Pull models straight from the built-in browser — Qwen 3.6, Gemma 4, and GPT-OSS are a click away.
Build this weekend: a fully offline coding agent
llama.app wires in Pi, a local coding agent that auto-discovers your running model. Point it at llama.cpp and you get a private coding agent with no API keys, no telemetry, and no rate limits.
The catch: update before you grab a model
Newer models like Gemma 4 need a current build; stale Homebrew or Nix installs fail on them. Pull the latest release — commenters confirmed build ~8770 runs the new models cleanly.
Elsewhere: a text-to-3D-world paper, not a download
Tencent's WorldClaw turns one prompt into an editable, game-engine-ready 3D open world — but it's a paper for now, and the repo ships no code yet. A preview of where text-to-world is headed, not a weekend build.