Aether Desktop beta · Apple Silicon

Local AI without the local AI ceremony.

Aether Desktop turns a Mac into a private OpenAI-compatible gateway for local models, cloud providers, API keys, usage logs, and a built-in chat tester. Install it, choose a model, point clients at one endpoint.

YOLO install
curl -fsSL https://aether-models.ufrik.com/desktop/macos/install.sh | sh

Downloads the latest DMG to ~/Downloads, verifies the checksum, installs Aether.app, and launches it. Requires macOS on Apple Silicon.

One gateway.
Every model.
Endpoint
:8181/v1
Mode
Local
Release
Beta
Qwopus 3.6 35B-A3BLocal
Gemma 4 26B-A4B32 GB
NVIDIA · OpenAI · OpenRouterCloud
Manage the Inference Server from Aether. No llama.cpp jargon in the user workflow.
What ships

A desktop control plane for adopting local AI.

Aether Desktop is the lightweight single-user version of Aether: no Postgres, no multi-user admin console, no server ceremony. It packages the gateway, companion app, local model catalog, Inference Server controls, and dashboard into a macOS app.

01

OpenAI-compatible gateway

Expose local and cloud models through one `/v1` endpoint for OpenCode, Claude Code, Codex, curl, Python, and other OpenAI-compatible clients.

02

Local model catalog

Download curated GGUF models from Cloudflare R2, including Qwopus coder models and larger 32 GB Apple Silicon options.

03

Inference Server management

Start, stop, restart, and validate the local Inference Server from the provider workflow, with logs exposed in the dashboard.

04

Provider routing

Add local servers like llama.cpp, Ollama, LM Studio, MLX-LM, and vLLM, plus cloud providers like NVIDIA, OpenAI, OpenRouter, DeepSeek, Kimi, and MiniMax.

05

Models and client snippets

See enabled local and cloud models, rename client-facing model IDs, and copy ready-to-use client configurations.

06

Usage and request detail

Track requests, token usage, model breakdowns, API keys, latency, raw prompts, and responses for debugging and tuning.

07

Built-in chat tester

Open Chat directly from the main navigation and test configured local or cloud models without leaving the app.

08

Desktop-native safety signals

Surface local/network binding, real endpoint URLs, system logs, and Mac power-state warnings that affect local inference performance.

Workflow

From zero to a model endpoint in four moves.

The product goal is simple: narrow the gap between a human with a Mac and useful local AI.

InstallRun the command, download the DMG, install Aether.app into Applications.
ChooseSelect a local model or add an existing provider.
RunAether downloads the right Inference Server package and starts it.
ConnectCopy an API key and use `http://127.0.0.1:8181/v1` from your client.
Coverage

Local first. Cloud when it helps.

Aether keeps local model adoption frictionless while still giving users a single place to route cloud models.

Provider
Use it for
Notes
llama.cpp
Primary local Inference Server path
Use llama-server base URL, no `/v1` suffix.
Aether local models
Curated GGUF downloads from R2
Aether manages model files, runtime package, and start/stop.
Ollama · LM Studio · MLX-LM · vLLM
Existing local tools
Register their base URL and route through Aether.
NVIDIA · OpenAI · OpenRouter
Hosted model access
Store provider keys locally and expose models through the same client workflow.
DeepSeek · Kimi · MiniMax
Additional OpenAI-compatible APIs
Base URLs are stored without `/v1`; Aether appends protocol paths.

Install now

Aether Desktop Beta is a one-command Mac install.

Use the YOLO installer for the latest DMG. For signed Sparkle updates, the appcast path is prepared and waiting on the release Ed25519 key.

Terminal
curl -fsSL https://aether-models.ufrik.com/desktop/macos/install.sh | sh