Terminal-native AI workbench
TrainShell
AI agents, real terminal, your GPU. Connect cloud or private hosts, then run Codex, Claude Code, recipes, and long jobs from one browser workbench.
$ trainsh connect gpu-4090-west
ok Host identity verified
✓ Browser vault unlocked locally
✓ Encrypted secrets injected into tmux
user@gpu-4090-west:~$ codex
Starting Codex CLI...
Why TrainShell
Built around the terminal, not around prompts.
The browser terminal, encrypted vault, datasets, and host state stay in one workspace instead of scattering across separate tools.
Connect a host
Use Vast.ai, RunPod, private EC2, Colab, Codespaces, or any Linux machine.
Open the workbench
Terminal, hosts, vault status, datasets, recipes, and outputs stay in one browser surface.
Run real agents
Launch Codex, Claude Code, OpenCode, and long-running jobs inside the actual shell session.
GPU supply
Bring your own host fleet.
TrainShell works with spot rentals, private machines, and cloud instances without forcing your training workflow into a separate hosted runtime.
Tooling
Use the tools you already trust inside the real shell session.
No chat wrapper, no second SSH client, and no separate secrets dashboard to keep in sync.
Native apps
macOS beta builds ship through the release channel. iPhone follows on TestFlight soon.