Getting Started with Aether

Installation

Download the installer for your platform from the Aether product page. Run the installer and follow the prompts. Aether ships in three formats, and the installer places each in the standard system-wide location. Installing requires an administrator password on both platforms.

The Headroom Piano is the sampled instrument Aether uses to audition your generations. It is installed alongside the plugin; if it is missing, previews fall back to a plain test tone.

First Launch

• ●Open Aether in your DAW as a virtual instrument, or launch the standalone app. Aether is an instrument plugin, so it will not appear in a MIDI effect or MIDI FX slot.
• ●Select an AI provider from the dropdown at the top of the chat panel.
• ●For cloud providers: click the gear icon, paste your API key, and click Save.
• ●For local providers (Ollama or LM Studio): enter your server URL and click Refresh Models.
• ●Type a description of the MIDI you want and press Enter or click Send.

API Keys

Cloud AI providers require an API key, which you create on each provider's website. A ChatGPT Plus or Claude Pro subscription does not include API access, so you need a key even if you already pay for one. At list prices, a generation is usually about 3 to 15 cents on a mid-tier model such as Claude Sonnet, depending on how many notes it writes. Lighter models such as Gemini Flash or DeepSeek are often a cent or two. The strongest models, such as Claude Opus or the top GPT and Grok models, usually write better parts, but each generation costs more and takes longer. Several providers offer free tiers. A local model through Ollama or LM Studio costs nothing.API keys are stored securely in your system's keychain (macOS Keychain, Windows Credential Manager).

To add or update a key: click the gear icon next to the provider dropdown, paste your key, and click Save. Your key is stored by your operating system rather than in a settings file, and Aether sends it only to the provider it belongs to, over HTTPS, to authenticate your own requests. It is never sent to Umbral Audio. Local providers need no key at all, so nothing leaves your machine when you use them.

Local Models (No API Key Required)

Aether generates entirely offline through Ollama or LM Studio. Nothing leaves your machine, there is no API key, and there is no per-request cost. The setup has one step that is easy to miss: the model and the server are separate things, and Aether talks to the server.

• ●Install Ollama or LM Studio and download a model. We recommend at least a 7B parameter model for usable results.
• ●Load the model with a context window of at least 32,768 tokens. Aether's instructions alone run to roughly 8,000 to 11,000 tokens before your prompt or any MIDI you attach, and it asks for up to 8,192 tokens back, so a model loaded at a 4,096 or 8,192 default has no room to answer. Both apps let you raise this when you load the model, and both default to well below what the model supports.
• ●Start the local server. In LM Studio that is the Status toggle in the Developer tab: loading a model in the chat window does not start it. Ollama serves whenever its app or background service is running.
• ●Turn on the option to start the server when the app launches, if the app offers one. Otherwise the server stops when you quit and has to be started again next time.
• ●In Aether, pick Ollama or LM Studio from the provider dropdown, click the gear icon, and check the server URL against the table below.
• ●Click Test Connection. It tells you how many models it found.
• ●Click Refresh Models, then choose your model from the list. Until you do, the dropdown only says Click Refresh Models: Aether cannot know what you have installed without asking your server.

Inside a DAW, Aether reaches your server through its network helper rather than connecting directly, because a DAW can block a connection the plugin opens for itself. The helper starts on its own a few seconds after the plugin window opens. If Test Connection fails and the message mentions the helper, start it from Settings and try again.

If Test Connection still fails against a server you know is running, put 127.0.0.1 in the URL in place of localhost. Both local servers listen on IPv4 only, and on some systems localhost resolves to an IPv6 address first.