Build with Atelier, native to your machine.
Studio ships as a real desktop app on macOS, Windows, and Linux. Same workspace as the web build, plus on-device inference, direct filesystem access, and the full @atelier/{project,skills,library} tool catalog.
All platforms
Pick the build for your hardware. Auto-updates are on by default and can be paused per workspace.
What the desktop gives you that the web can't.
Same Studio workspace. More permissions, more hardware, more privacy.
On-device inference
Run local LLMs through the desktop bridge. Metal on Apple Silicon, CUDA on NVIDIA. Tokens never leave the machine.
Filesystem-native
Open existing repos as projects, edit live, and let the agent run shell commands in their real working directory.
Offline-capable
Once a project and its providers are cached you can keep building on a flight. Sync to Base when you reconnect.
Same workspace, more permissions
Identical UI to the web build, plus OS integrations: drag-and-drop assets, system clipboard, native notifications.
System requirements
Studio is light at idle. The numbers below cover comfortable use with local inference workloads.
macOS
- macOS 13 Ventura or later
- 8 GB RAM (16 GB recommended for local inference)
- Apple Silicon strongly preferred for Metal acceleration
Windows
- Windows 10 (build 19041) or Windows 11
- 8 GB RAM (16 GB recommended)
- NVIDIA GPU with CUDA 12 for local inference (optional)
Linux
- Ubuntu 22.04+ / Debian 12+ / Fedora 38+
- glibc 2.31 or later
- NVIDIA driver 535+ with CUDA 12 for local inference (optional)
Prefer the browser?
Studio runs in any modern browser too. You lose local inference and direct filesystem access, but the rest of the workspace is identical and projects sync to Base either way.