102201

Multi-instance DeepSeek Harness

Unlimited isolated instances
on one desktop

Support, engineering, and testing cannot share one dsh. Each instance is a complete, isolated Harness — in one workspace, ready to export when an instance is configured.

Download for desktop View repository

Windows · macOS · Linux · GitHub Releases

DSHHelper workbench with multiple DeepSeek Harness panes
  • Support
  • Code
  • Tests
  • Orchestrate
  • Build
  • Docs
  • Test
  • Review
  • Frontend
  • Backend
  • Tests
  • Package

Isolated plugins · one workspace · export when ready

01 Isolate

One instance cannot host multiple workloads

Unrelated plugins in one preset make the model call the wrong tools. Isolate the workloads first — then collaborate, then deliver.

The problem

Customer messages, code, and tests share one DSH_HOME. The tool list grows until the model misroutes. Three capability sets cannot coexist in one preset.

Agent presets do not resolve this: only one mounts per session, and switching after a turn is locked.

Per DeepSeek Harness: Agent Presets and Personas

The approach

Each instance is a complete, isolated dsh — own process, port, and DSH_HOME. Create as many as you need. Plugin environments never conflict. Isolate, then run in parallel, then package.

02 Collaborate

Specialist instances, one workspace

Plugin sets stay isolated. Project files stay in view. You orchestrate from the panes — a configured instance is ready to export.

Each instance is a specialist: support, engineering, testing. Plugins are its toolchain, so they must be isolated.

Spreading them across three machines is three separate environments — multiplied hardware, files copied back and forth, no shared state. Place them on one computer, against the same project folder. Not three unrelated windows — one workspace advancing one release.

Specialist instances: one workspace versus three separate machines
Dark shell with a 2 by 2 quad-pane layout
Orchestrate from the panes — shared folder, isolated plugins

03 Monetize

When the instance is ready, export the pack

You deliver a runnable specialist instance — isolated and coached in one workspace, not another week of remote setup.

After configuring support, engineering, and testing, delivery often fails not on “can they use it”, but because every handoff feels like rebuilding the project.

  1. Cannot reproduce Days later even you cannot confirm which plugins and which preset constitute the product.
  2. Cannot hand off A README never guarantees a clean install; shipping the folder invites redistribution.
  3. Cannot scale The next customer still requires the same remote setup. Delivery time dwarfs the time spent configuring.

A .dshpack encapsulates that configured instance as a file. They import and run. Bind a machine code or password; forbid re-export. Configure several instances, export several packs.

Export and import flow for .dshpack packages
Import package dialog in DSHHelper
Import a pack — they receive a runnable instance, not a setup ticket

Get started

Install the app. Create instances.

Windows, macOS, and Linux. The wizard pulls the runtime. Then you split panes.

  1. Install Download the build for your platform from Releases.
  2. Initialize Choose a data folder, pull the runtime, create the first instance.
  3. Split Create more instances. Drag tabs to the edges. Run them in one workspace.
Download for desktop View repository

Free download · not code-signed yet · runtime downloads on first run

First-run wizard in DSHHelper
First-run wizard

Next

From isolation to delivery,
on one desktop.

Isolate the workloads. Collaborate in one workspace. Package a configured instance as a .dshpack.

Download for desktop