Preface

The philosophy of DSH is “everything is a plugin”: capabilities like models, sessions, and tool calls can be combined by profile. If you prefer to interact directly with DeepSeek Harness models in the terminal rather than switching to a browser or an additional web interface, deepseek-harness-tui offers an optional TUI plugin.

It is a community project under the MIT license, independent of DeepSeek and TokenDance, not a reimplementation of the harness, but a thin UI plugin of approximately 800 lines.

What This Is

deepseek-harness-tui is an interactive terminal chat TUI for DeepSeek Harness (dsh), provided by the gxinxing/deepseek-harness-tui repository.

It is positioned as terminal-native: built with Ink, it brings session transcripts, tool calls, thinking processes, and Markdown rendering into the terminal interface. Through cordis.patch.yml, it routes llm-deepseek to the TokenDance gateway.

Core Features

  • Ink-based terminal chat UI
  • Bottom-anchored live transcript, with the tail always visible
  • Tool calls collapsed into cells, showing active/completed/error states, with truncated output
  • Terminal background detection via OSC 11, inferring message tints and code chips
  • Toggle thinking/reasoning trace with ctrl + t
  • Interrupt a running turn with esc
  • Markdown rendering preserving headers, fenced blocks, inline code, and handling CJK/emoji widths
  • Routing llm-deepseek to the TokenDance gateway via cordis.patch.yml

Installation and Enablement

You need Node.js ≥ 20 and the DeepSeek Harness CLI. First, install the harness, then clone and install the plugin dependencies:

npm install -g @deepseek-ai/dsh
git clone https://github.com/gxinxing/deepseek-harness-tui
cd deepseek-harness-tui && pnpm install

Then add this plugin to the tui profile. Use a local path here, not a remote plugin address formed directly from the repository name:

dsh plugin --profile tui add @deepseek-ai/dsh-headless
dsh plugin --profile tui add /path/to/deepseek-harness-tui

If the clone directory is not in /path/to, replace it with the actual path.

Typical Usage

First, prepare the TokenDance API key. You can export it temporarily or place it in ~/.dsh/.credentials.yaml (file permissions recommended as 0600):

export TOKENDANCE_API_KEY=sk-...

Start the TUI:

dsh --profile tui

After entering the TUI, common operations are:

  • ctrl + t: Toggle thinking trace
  • esc: Interrupt a running turn
  • /help: View all keys and commands

If you want to test with a fixed background color, set DSH_TUI_BG:

DSH_TUI_BG=#ffffff

Then run:

dsh --profile tui

Applicable Scenarios and Considerations

It is suitable for developers who need to use DSH conversations, tool calls, and Markdown output in the terminal, especially for scenarios where the interface should closely resemble the terminal itself, avoiding extra web wrappers.

Before use, please note:

  • It is an independent community project under the MIT license, with no official affiliation with DeepSeek or TokenDance.
  • The plugin runs with the current dsh process; before installation, it is recommended to check the repository source, package.json, and license.
  • package.json declares private: true and requires engines.node >= 20.
  • TokenDance tool call deltas require a patch to @deepseek-ai/dsh-llm-deepseek; the patch is in the global dsh installation and will be lost after upgrading dsh, requiring reapplication.
  • Do not interpret the installation commands here as official app store entries; the community directory is an independent site, and the actual installable content is subject to the current descriptions in the repository and directory page.

Conclusion

The value of deepseek-harness-tui lies in condensing the terminal experience of DSH into a thin plugin: bottom-anchored transcripts, tool call collapsing, thinking toggle, OSC 11 background detection, and Markdown rendering are all organized around terminal use cases. If you want to try it in your own dsh installation, you can start with the installation steps above and complete the profile binding based on your local path.

GitHub repository: https://github.com/gxinxing/deepseek-harness-tui
Directory page: Please refer to the current link in the community directory you are using.