openma-ai/deepseek-harness-tui
272Last commit Aug 15, 2026
deepseek-harness-tui DSH plugin
dsh-tui is a terminal UI for DeepSeek Harness built with Rust/ratatui. It provides real-time streaming inference, tool calls, subagent lifecycle, token usage, and persistent session management. It can run as a dsh profile plugin or standalone with a JSON-RPC runtime.
How to install the deepseek-harness-tui DSH plugin
dsh plugin --profile tui add @openma/deepseek-harness-tuiCopying does not run this command. Review the repository and version before installing the deepseek-harness-tui DSH plugin.
deepseek-harness-tui DSH plugin data source
deepseek-harness-tui DSH plugin snapshot date: Aug 16, 2026
discovered
What the deepseek-harness-tui DSH plugin can do
- Real-time agent timeline displaying inference, tool calls, plugin context, subagent lifecycle, and token/cache metrics.
- Multi-image prompts: stage up to 8 images from file, clipboard, or paste, displayed as editable [image n] chips inline in the draft.
- High-density tool view showing progress, success, and failure states with collapsible outputs and independent scrolling.
- Long conversation controls: queue follow-ups, interrupt and send immediately, manage persistent JSONL sessions via /new, /resume, and --session-id.
- Cross-platform input: readline editing, context shortcuts, macOS physical ⌘/⌥ fix, Linux/Windows ctrl combos.
- Terminal-native UI: light/dark themes, narrow layout, mouse selection, clipboard support (native/tmux/OSC 52), and optional kitty graphics protocol image preview.
- Slash command menu: built-in commands and host skills share a searchable, scrollable menu triggered by /.
Where the deepseek-harness-tui DSH plugin fits
- Interact with AI agents in the terminal with real-time streaming and tool call visibility.
- Debug and monitor agent behavior, including subagent spawns, permission prompts, and token usage.
- Manage multi-image conversations with inline image chips and previews for visual tasks.
- Run long-lived agent sessions with persistent JSONL storage and session resume capabilities.
- Develop and test DeepSeek Harness skills and plugins in a convenient terminal UI.
Who the deepseek-harness-tui DSH plugin is for
- Developers and power users of DeepSeek Harness who prefer terminal-based interfaces.
- AI enthusiasts who want a native, keyboard-driven agent UI with streaming and tool call support.
deepseek-harness-tui DSH plugin limitations
- Plugin mode requires an installed and configured dsh (current baseline 0.1.0-rc.6) and Node.js 18+ with pnpm 10+.
- Official packages only cover macOS Apple Silicon, macOS Intel, Linux x64, and Windows x64; other platforms may not have native binaries.
- Standalone mode requires a separate DeepSeek Harness SDK installation (Python virtual environment) or a specified runtime binary.
- The image preview feature (kitty graphics protocol) is limited to compatible terminals; others fall back to half-block characters.
- The plugin is still in early development (version 0.2.x), and some features may be unstable or change.
deepseek-harness-tui DSH plugin: from the repository README
Quoted from the openma-ai/deepseek-harness-tui README, the upstream source of the deepseek-harness-tui DSH plugin. Copyright remains with the original authors.
在终端里运行 DeepSeek Harness：流式推理、工具调用、Skills、多图 prompt 与持久会话。 </p> <p align="center"> <a href="README.md">中文</a> · <a href="README.en.md">English</a> </p> <p align="center"> <a href="https://www.npmjs.com/package/@openma/deepseek-harness-tui"><img src="https://img.shields.io/npm/v/%40openma%2Fdeepseek-harness-tui?logo=npm&color=cb3837" alt="npm version" /></a> <a href="https://www.npmjs.com/package/@openma/deepseek-harness-tui"><img src="https://img.shields.io/npm/dm/%40openma%2Fdeepseek-harness-tui" alt="npm downloads" /></a> <a href="https://github.com/openma-ai/deepseek-harness-tui/actions/workflows/package-npm.yml"><img src="https://github.com/openma-ai/deepseek-harness-tui/actions/workflows/package-npm.yml/badge.svg" alt="CI" /></a> <img src="https://img.shields.io/node/v/%40openma%2Fdeepseek-harness-tui" alt="Node.js 18+" /> <a href="LICENSE"><img src="https://img.shields.io/badge/license-MIT-blue" alt="MIT" /></a> </p> --- `dsh-tui` 是 DeepSeek Harness 的终端原生 agent UI：在一个 Rust/ratatui 界面里查看流式推理、工具调用、subagent、token 用量和持久化会话。它既可以 作为官方 `dsh` profile 插件运行，也可以直接连接 SDK JSON-RPC runtime。  ## 快速开始 #
Read the full READMERepository license: MIT
deepseek-harness-tui DSH plugin questions
How do I install dsh-tui as a DSH plugin?
First ensure you have dsh installed (baseline 0.1.0-rc.6 or later) and Node.js 18+ with pnpm 10+. Then run `dsh plugin --profile tui add @openma/deepseek-harness-tui` followed by `dsh --profile tui`. The plugin will mount the TUI binary and connect to your dsh profile's agent, tools, and provider.
Can I try dsh-tui without an API key or runtime?
Yes, you can run the demo mode without any API key or runtime. Install the package globally with `npm install --global @openma/deepseek-harness-tui`, then run `dsh-tui --demo`. The demo simulates a conversation to showcase the UI features.
What platforms are supported?
The official npm package includes native binaries for macOS Apple Silicon (darwin-arm64), macOS Intel (darwin-x64), Linux x64, and Windows x64. If you see 'no native binary for ...' error, your platform may not be covered. You can build from source using Rust stable and Node.js 18+.
How do I switch between plugin mode and standalone mode?
Plugin mode is the default when you install via `dsh plugin` and run with `dsh --profile tui`. Standalone mode is used when you run `dsh-tui` directly (without dsh). In standalone mode, you need a DeepSeek Harness SDK runtime (Python venv) or specify a runtime binary via `--runtime-bin` or `DSH_RUNTIME_BIN`. The sessions directory also differs: `~/.dsh/sessions` for plugin, `~/.dsh-tui/sessions` for standalone.
What keyboard shortcuts are available?
Key shortcuts are listed in the README. Common ones: Enter to send, Ctrl+X to interrupt and send next, Esc to interrupt (keep draft), Ctrl+C to clear draft then interrupt. Slash `/` opens the command menu. Readline editing: Ctrl+A/E for line start/end, Ctrl+K/U to delete to end/start, Ctrl+W delete word. macOS also supports physical ⌘←/→ and ⌥←/→. Run `/keys` inside the TUI to see the full list.