No code editor: the desktop app with a free local model | DATASPHERES AI - Dataspheres AI

What you will do Use Dataspheres AI from a desktop app instead of a code editor, with a language model that runs on your own computer. There is no assistan...

What you will do Use Dataspheres AI from a desktop app instead of a code editor, with a language model that runs on your own computer. There is no assistant subscription in this path and no key to paste: the app is a free download, the model is a free download, and the Dataspheres account is the same free account as the previous lesson. Once the model file is on the disk, the local model answers without an internet connection. Download the app The desktop app is open source and published as an installer for each platform on its releases page . Pick the file for your computer: the arm64.dmg for a Mac with Apple silicon, the plain .dmg for an Intel Mac, the Setup .exe for Windows, or the .AppImage for Linux. Open it and follow the installer. The app checks the same page for updates afterwards, so there is nothing to reinstall by hand. Choose the free local model The first time the app opens it asks one question: how should Ari think on this computer? The first card, Free local model , downloads Llama 3.2 1B Instruct at Q4_K_M quantisation, a 0.8 GB file published by bartowski on Hugging Face under the Llama 3.2 Community License. Choose Download and use it . A progress bar shows the megabytes arriving; on an ordinary connection the wait is a few minutes. The file lands in the app's own data folder, so removing the app removes the model. A laptop with 8 GB of memory runs this model comfortably; it needs about 2 GB free while answering. The second card, Bring your own key , is for someone who already pays for an Anthropic API key and wants Claude to answer instead. The key is encrypted by the operating system and stays on the machine. Skip that card for this lesson. Not now opens the app with hosted chat only and asks again next time. Sign in with the free account After the model lands, the app opens your Dataspheres workspace. Sign in with the account from the previous lesson; the sign-in happens in your browser and the app picks it up when it finishes. Your dataspheres, pages and planner boards appear exactly as they do on the web. The app keeps the model loaded and offers it to the workspace through its desktop bridge, so a feature that asks the desktop for a local answer gets one from your own processor. Chat in the workspace itself still uses the hosted models on the free plan today; the local route is being wired into it feature by feature, and the app's releases page lists each step as it ships. What a 1B model is and is not A one-billion-parameter model is small on purpose. It fits on a student laptop, it answers quickly, and it costs nothing per question, which makes it right for drafting, summarising, rephrasing and asking about a page in front of you. It knows less than the hosted models and reasons less carefully, so check facts against the source and move to the web app for research runs or long reports, which use the hosted models under your plan's allowance. Any other GGUF file from Hugging Face can replace it later, as the app's README describes, if the machine has the memory for it.