Plugins

你想蒸馏的下一个员工，何必是同事。蒸馏任何人的思维方式——心智模型、决策启发式、表达DNA。Distill how anyone thinks.

Let Claude Code or Codex do the thinking: understand the request, write a clear prompt for DeepSeek Harness to do the work, and review the result. dsh-baton runs your tests itself, lists what changed, and has a sub-agent check every requirement, so your main model only reads the result and decides. Codex waits in one call of up to 59 minutes, with no hook to trust and nothing to poll; you can change the request while it runs, and name the Harness model for a task. On big jobs it saves most of your plan's quota.

Google Antigravity (agy CLI) models for DSH — streaming chat with Gemini/Claude/GPT-OSS subscriptions, native tool cards, thinking turns, and in-GUI Google OAuth login.

Connect any ACP-capable agent to DSH and use DSH as a single entry point for all conversations and tasks. Models, thinking levels and permission modes come from the agent itself; tool calls appear in the main chat, diffs are collected on a dedicated NTR page, and approval requests appear in the composer. Requires an ACP-capable agent CLI installed locally.

Inline mini-game arcade for DSH streaming reasoning / 流式思考下方的轻量小游戏（7 款游戏，自动暂停/淡出，可继续上一局）

A Live2D whale-girl companion for the DeepSeek Harness Web UI: the model stays on screen and reacts to what the agent is actually doing (thinking, tool calls, streaming text, errors, turn completion); click her to send the agent a message and watch the reply stream into a speech bubble; right-click for a menu of expressions, dress-up items, desk scenes and one-shot animations, every action mapped from the original model author hotkey sheet so nothing overlaps and everything expires on its own. Non-commercial model assets (CC BY-NC-SA 4.0), MIT code, by Andersen216 (https://github.com/Andersen216).

Translate DSH reasoning, tasks, answers and review reasons into 9 languages. Ollama, custom OpenAI/Anthropic APIs or Google/Bing; ordered fallback, local model memory controls, originals preserved.

Keeps a local TinyTitan model server reachable from the harness: refreshes the llm-pi-ai route from the installed models and mounts a compaction backend that does not think.

Codex-style workflow auto-collapse: finished turns fold into a single "processed in Xs" row leaving only the final answer; tool calls and think blocks collapse into live-summary chips that follow streaming; expand level by level, fully reversible on uninstall.

Configure reasoning-effort levels and sub-agent defaults for custom DSH models.

Customize the "deep diving" thinking status label to anything you like.

Pins the model's reasoning language to a tier picked in Settings - General - Thinking language (Simplified Chinese, or a default tier that only constrains replies and leaves reasoning to the model), by injecting a system-prompt section at order -1000000 and re-hoisting it to first place at the end of the assemble waterfall, plus an optional per-turn copy placed ahead of each user message for models that follow static sections weakly.

Reflective self-evolution plugin for local Qwen3.8 27B/3.8 Flash and online APIs: a coding-loop convergence steerer curbs Qwen3.8 27B/Flash repeated-verification deep-thinking to converge faster and cut thinking time; a five-state delivery guard detects and corrects stuck turns and infinite loops; model-agnostic host logic, mostly compatible with online APIs; per-use continual self-improvement with counter-reason self-checks, critic escalation, inverted-index memory retrieval, coding-loop convergence steering, and a five-state delivery guard.

Side-panel arcade: 18 offline mini-games to play while the model thinks.

Let Your Ideas Flow — Tydora is a modern desktop Markdown editor combining WYSIWYG editing, bidirectional links, mind maps, and an infinite canvas — empowering deep thinking and effortless expression.

UI enhancement suite for the DSH web client: Chinese locale completion, one-row stats, auto-expanded thinking, chat width, session archiving and delete, plus an optional prompt-injection switch (off by default).

The marketing OS for AI agents — think like a marketer first, get tactics as the output.

Auto thinking-effort routing for dsh: a locally fine-tuned 7-question classifier (ModernBERT, 842MB, <1s CPU inference) picks low/high/max for no-intent messages; in-session retries escalate the tier (low to high to max) under exclusion constraints; the judge service fails safe to low if unreachable. A tier chip in the input bar shows the live tier and the last 20 judgments. All inference is local - task text never leaves the machine.

Live2D desk pet overlay for the DSH web GUI: mirrors agent state (thinking, idle, error, done, awaiting approval) with motions and speech bubbles, per-part touch reactions, drag-to-dock, switchable personas, and custom models via URL or local path.

H.D.D skin for the DSH Web UI: a voiced full-screen boot animation, an animated eye core (idle / thinking / done) you can drag or hide, a brand swap on the new-chat hero, a time-of-day greeting, a usage bubble with balance, today spend and the peak/off-peak countdown, the Fairy-mode agent preset, and optional local GPT-SoVITS voice.

Streaming fold for the DSH web transcript: the newest thinking window opens itself, older process rows fold into one summary, and a finished turn glides back to your prompt, which stays pinned while you read.

Per-request thinking budgets, a compaction backend, and a settings tab for the local Qwen3.8 line (llama.cpp or NInfer).

Installs Herta as a DeepSeek Harness agent: a preset carrying her persona and memory prompt sections, six memory and voice tools, a narrative layer that separates her thinking from what she says, 80 bundled voice clips, offline and MiniMax speech synthesis, and two conversation views that render the session with her own components.

Visual branch brainstorming that creates an isolated Session for each direction, automates parent, sibling, checkpoint, and return Handoffs, and provides a full tree tab with an optional Better Sidebar view.