The workspace
Layout
The workspace is organized around a session chat plus tabs for the active work surface:
- Files. Browse generated files and workspace artifacts.
- Research. Review plans, runs, metrics, and experiment state.
- Compute. Inspect live GPU, provider, and session status.
- Terminal and SSH. Read command output and connect from your own terminal.
- Activity. Follow the high-level stream of what changed.
The exact layout can change with viewport size, but these tabs are the durable navigation model.
File explorer
Shows the pod's filesystem. Click a file to preview it. Right-click to download, rename, or delete. The agent can also create and modify files. You'll see changes appear in real time.
Terminal
Live output from everything the agent runs. Commands, stdout, stderr. It's all here. You can scroll back through the full history. This is read-only; to run commands yourself, use SSH or ask the agent.
GPU monitor
Real-time GPU utilization, VRAM usage, temperature, and power draw. Updates every 2 seconds. Useful for spotting when a training run is actually using the GPU vs. being bottlenecked by data loading.
Switching tabs
Use the workspace tab bar to switch surfaces:
| Tab | Use it for |
|---|---|
| Files | File explorer and generated artifacts |
| Research | Plans, runs, metrics, and experiment details |
| Compute | GPU status, runtime, provider, and cost context |
| Terminal and SSH | Terminal output and SSH access |
| Activity | Recent session events |
Keyboard shortcuts
Press ? anywhere in the workspace to see the current shortcut reference. Press Cmd+K to open the command palette, which lets you search for actions by name.