Eight.ly Stick
A workspace you can carry.
Eight.ly Stick explores a portable AI workspace that lives on a USB drive. Models, settings and conversations travel with the workspace rather than being rebuilt on every computer.
- 3Desktop platform setup paths
Windows, macOS and Linux setup paths.
- Shared dataPortable configuration and history
Platform executables remain separate from the shared workspace.
- Local or remoteModel choices
Local execution depends on installed model files and host hardware.
- PrepareSet up the platform files and selected models.
- CarryKeep settings and conversations with the USB workspace.
- LaunchStart local chat or the coding dashboard.
- ChooseUse a local engine or a configured remote provider.
The problem
A useful local AI setup has more parts than a model file. It needs an engine, a way to talk to it, configuration and a record of the work. Moving to another computer often means assembling those pieces again and discovering which settings stayed behind.
Hardware adds another complication. A machine with an appropriate graphics processor needs a different execution path from a CPU-only computer. A portable workspace has to account for the computer it is plugged into while keeping the person’s data in a predictable place.
Who it’s for
People who move between Windows, Mac and Linux machines and want a portable local AI and coding environment.
What we built
The project combines platform-specific setup and launch scripts with a shared data folder, browser-based chat and a coding dashboard. The workspace can use local models or connect coding tools to a selected external provider.
Settings, model files and conversation history live within the portable workspace. Optional development tools extend that environment. The local chat interface includes model management and file attachment, so adding a model does not require rebuilding the whole stick.
How it works
Each supported desktop platform has its own setup path and executable folder. Shared data remains together, allowing the same workspace to move between machines after the necessary platform setup. Local chat connects to a model engine running on the host computer.
The setup includes hardware detection and different acceleration options, with a CPU fallback. A local model can work without a hosted chat service once its required files are present. Choosing a remote provider instead means that provider receives the relevant request; portability and local execution are separate choices.
What was hard
Portability does not make hardware identical. Memory, available acceleration and the chosen model still determine what can run well. The setup needs to select a suitable engine and leave room for the user to choose a model their computer can support.
The work assembles existing model engines and coding tools into a portable setup, shared workspace and launch experience. The challenge is making those parts behave as one environment across different desktop systems. That integration remains in the lab.