Draw Things’ developers have launched Local Code, a free Mac app that runs downloaded AI models on the computer rather than sending prompts and files to a cloud provider. It can work with documents and spreadsheets, create presentations and media, and use a browser to carry out online tasks.
Watch Desk analysis
What happened
Local Code bundles model downloads with the tools needed to use them, so Mac users do not have to assemble that setup separately. With permission, it can read information from Mail, Messages and Notes, and make changes in Calendar and Photos. Its browser can navigate websites and complete online tasks.
The app offers Qwen and DeepSeek models, with the models available depending on the Mac’s memory. The developers’ website lists 16GB of memory as the minimum for Qwen 3.8 27B. Larger DeepSeek downloads are listed at about 89GB or 376GB, depending on the version. A direct Mac download is available; a Mac App Store version is promised soon. MacRumors reports that Local Code follows a public beta that began on 12 September.
Local use is free. Users can also choose a cloud model, with access through their own API keys or prepaid credits starting at $9.99. In that mode, prompts and relevant files go to the selected provider.
Why it matters
Local inference gives people a way to use AI with prompts and files kept on their Mac, rather than routinely handing them to a cloud service. But “on-device” does not mean the assistant can do nothing beyond the device: its browser can go online, and the app can access selected personal apps when permission is granted.
The practical trade-off is familiar, but now unusually concrete: privacy and local control on one side, model size and hardware demands on the other. A 376GB download is a fairly emphatic reminder that the word “local” still needs a storage plan.
Our read
Local Code looks useful for Mac users who want to try models on their own hardware without building the surrounding toolkit themselves. Before granting access to personal apps, check what permissions you are giving it; before downloading, check that your Mac has the memory and storage the chosen model needs. The headline feature is local processing, but the permission settings and model requirements are where the everyday experience will be decided.
What to watch
- Which Macs can run the available models smoothly, beyond the published memory requirements.
- Whether the promised Mac App Store release arrives.
- How clearly the app presents permissions for personal apps and online tasks.
- Whether users find local models capable enough to avoid sending work to a cloud provider.
Discussion spark: Would you trust a local AI assistant with access to Mail, Messages and Calendar, or would those permissions defeat the point of keeping the model on your Mac?
Sources and evidence
- Source update (30 October 2026, 13:38 UTC)
Watch Desk is operated by WittyWires as an independent cross-cutting AI news tracker. It does not speak for the organisations or people it covers.