[ dt ] // desktop app · windows · macos · linux

Local AI, no terminal required_

Everything the ArcFlare CLI does, in a window: chat, a studio for images, 3D and speech, flows that chain them, and your coding agents, all running on your own GPU.

Download for Windows· 10 / 11 · x64 installer

Or from PowerShell on Windows. It checks the download against the release's SHA-256 and skips the SmartScreen warning, which only appears because the builds aren't code-signed yet:

irm https://arcflare.net/install-desktop.ps1 | iex

Or from a terminal on Linux or macOS. It picks pacman, apt or the AppImage for you, checks the download, and opens without a Gatekeeper warning on a Mac:

curl -fsSL https://arcflare.net/install-desktop.sh | sh

On Arch, pacman can install the release directly:

sudo pacman -U https://github.com/Hakeperty/Arcflare-Desktop/releases/latest/download/ArcFlare-linux-x64.pacman

Free, MIT licensed, no account. All releases and notes · Source

The ArcFlare desktop app: machine readout, chat model and the four areas — chat, studio, models and harnesses

arc-01 // ascii

The same engine as the CLI, so a model you set up in one works in the other, and the llama.cpp server is shared instead of loaded twice. The panda is watching your cursor; click it.

[ 01 ]// inside

What's in the window.

Chat

Pick a model and talk. ArcFlare sizes its context to your free VRAM and reloads it for you after a studio job.

Images

stable-diffusion.cpp built in, or connect your own ComfyUI and run any workflow from the app.

3D, edit by asking

Turn a picture into a mesh, then brush over a part and type “make this bigger” or “smooth the ears”. The change happens on the model.

Speech

Qwen3-TTS, Kitten TTS 2, Kokoro, Chatterbox and more. Pick a voice, set the tone or emotion, or clone one from a short clip.

Voice lab

Record a voice with a passage to read, save it by name, and hear it on every cloning model side by side. Only voices you have permission to use.

Flows

Chain steps node by node, ComfyUI-style: prompt → image → 3D → render, or text → script → voice.

A coding agent

Open a folder and ask. It reads, writes and runs commands there, asking before anything risky, with Stop and a per-turn summary. Same agent as arcflare agent.

Any harness

Open OpenCode, Hermes, Codex or the ArcFlare agent pointed at your local model, in one click.

From your phone

Press phone in the title bar and scan the QR code. Chat with the model on your computer from anywhere.

Updates itself

New versions download in the background; restart when it suits you. Windows and AppImage update in place.

A model shop

ArcFlare's picks, plus every GGUF model on Hugging Face. Each file shows its real size against your free VRAM, and a suggested quantisation.

Shares with the CLI

Models, settings and the llama.cpp server are shared with arcflare, so nothing is downloaded or loaded twice.

[ 02 ]// setup

Three steps.

01
Download
Pick your system above. Installers are about 80 MB; models are downloaded separately, only the ones you choose.
02
Open it
The builds aren't code-signed yet, so Windows and macOS may warn you. The terminal installers above check the download and open without the warning. From the browser: on Windows choose SmartScreen → More info → Run anyway; on macOS open the app once, then System Settings → Privacy & Security → Open Anyway.
03
Get a model
The first run downloads llama.cpp for your graphics card by itself, checked against the SHA-256 GitHub publishes. Then the Shop lists hundreds of models marked against your GPU. One click downloads one.

Prefer the terminal? Install the CLI. Source code: github.com/Hakeperty/Arcflare-Desktop.