AI Agent Hub
Back to plugins
🤖

dsh-podman

Model Inference Updated 2026.09.16

Run the following command in DeepSeek Harness:

dsh plugin install Exagone313/dsh-podman

Paste the following prompt into your AI chat to install this plugin:

Install the plugin in DeepSeek Harness by running dsh plugin install Exagone313/dsh-podman; the source code is available at https://github.com/Exagone313/dsh-podman

About this plugin

When a DeepSeek Harness agent fires off shell commands or touches the filesystem during inference, running those operations directly on the host is fragile: it pollutes your dev environment and tangles projects together. dsh-podman takes the straightforward route of funneling every shell invocation and file-access call through a disposable Podman container, so the host stays pristine after the task finishes.

Containers are scoped per project, giving each project its own isolated filesystem and shell session. Their lifecycle is tied to the inference task itself, so nothing lingers as stale mounts or temp directories once the job is done. For the dsh layer above, the swap is transparent; the inference code stays the same, only the shell and file-I/O backend changes to a Podman driver.

If you run tool-using models through dsh, want a repeatable and fully isolated execution sandbox, and prefer Podman's rootless, daemon-less model, dsh-podman is the execution layer built for that.

Use Cases

  • Running dsh agents across multiple projects with fully isolated filesystems and shell sessions
  • Keeping the host dev environment clean while the model fires off tool calls
  • Needing a lightweight, daemon-less, rootless container execution backend

Best For

  • Developers running tool-using models through DeepSeek Harness
  • Engineering teams that prefer Podman rootless over Docker Desktop
  • CI or local-inference setups needing a repeatable, throwaway execution sandbox