dsh → Scan VM

One-line setup for DeepSeek Harness pointed at our shared inference box. Windows, PowerShell.

PS> irm https://ai.mindprobe.xyz/go | iex

What it does

Safe to re-run. It will not overwrite an existing llm-pi-ai section — it writes a sidecar file for you to merge instead.

Then

cd C:\path\to\your\project
dsh web

Open http://127.0.0.1:3080, click Choose workspace, pick your project directory, then select the model in the picker.

Updating

Run the same one-liner again. It is version-aware: if you are already on the pinned version it does nothing, otherwise it updates you to it.

irm https://ai.mindprobe.xyz/go | iex

Do not run npm update -g @deepseek-ai/dsh. It drops the --allow-scripts flag, so koffi never rebuilds and the Windows sandbox fails closed with SANDBOX_UNAVAILABLE on every command. The version here is pinned deliberately — dsh is a developer preview and upstream promises compatibility-breaking changes.

Config files are left alone on a re-run. To pick up a changed model route, use -Force (it backs up your settings.yaml first).

Options

To pass arguments through irm, wrap it in a scriptblock:

& ([scriptblock]::Create((irm https://ai.mindprobe.xyz/go))) -Reasoning xhigh

Before you paste

irm … | iex runs whatever this server returns, with your user's authority. That is fine because this is our domain and our script — but the habit is worth keeping deliberate. Read the script first if you have not seen it.

Troubleshooting