arcflare_
[ log ] // writing

The blog.

Releases, measurements, and the reasoning behind the decisions — including the ones we got wrong the first time.

8 postsrss feedchangelog →

latest·4 min read

More models: six ways to make 3D, sixteen coding models, five voices

The hub is now 56 models from 26 labs. What is new in 3D, code and speech, which one fits your GPU, and what we are adding next.

read it
[ ?? ] // start here

New to ArcFlare? Here's everything you need to know

What ArcFlare is, what your computer needs, which model to pick for your GPU, and the dozen commands you will actually use, in plain words. Start here.

  1. 01 what your computer needs
  2. 02 pick a model that fits
  3. 03 the commands you'll use
9 min guide

earlier

·4 min

Scan to control from your phone, an app that updates itself, and a security review

/rc prints a QR code, the desktop app updates itself, verified updates and clearer MCP trust after a security review, harness updates, bug reports, and water on the website.

#release #remote #desktop #security

·3 min

ArcFlare Desktop: local AI without the command line

Chat, a studio for images, 3D and speech, 3D models you edit by pointing and asking, flows, and one-click coding agents, in an app for Windows, macOS and Linux.

#release #desktop #studio #3d

·4 min

ArcFlare 1.1: 3D meshes, your session from anywhere, and updates that work offline

arcflare gen turns a photo into a mesh with Hunyuan3D on your own GPU. /rc puts the session on your phone. And /update keeps it current, even on a machine that never sees the internet.

#release #3d #remote #cli

·5 min

The machine server is now something you choose

It runs commands, opens applications, reads the clipboard and photographs the screen. That is the point of it — and a good reason to be asked first.

#security #mcp #cli

·4 min

What fits, and what measurably runs best

Two questions that cost wildly different amounts to answer. arcflare fit answers the cheap one every time and the expensive one only when you ask.

#performance #vram #cli

·4 min

1.0: we stopped embedding llama.cpp

0.x hosted the engine in-process. 1.0 drives the stock llama-server binary as a child process — and that one change did most of the work.

#architecture #llama.cpp #release