Skip to content
  • Home
  • Plugin manual
  • Liltloom

Adkid-Zephyr/Liltloom

30Last commit Aug 14, 2026

Liltloom DSH plugin

Liltloom is a DeepSeek Harness plugin that extracts and stores your writing style preferences from eligible user-authored text. It compiles a bounded Style Context Packet only when you explicitly want AI to mimic your style. It is Chinese-first, with full UI for rule management and data control.

skills-workflowsdeveloper-tools

How to install the Liltloom DSH plugin

dsh plugin --profile web add /absolute/path/to/Liltloom

Copying does not run this command. Review the repository and version before installing the Liltloom DSH plugin.

Liltloom DSH plugin data source

Liltloom DSH plugin snapshot date: Aug 16, 2026

discovered

What the Liltloom DSH plugin can do

  • Automatically learns writing style from user's own text, filtering out non-user content like tool outputs and quoted text.
  • Explicit take-out mode: style context is only applied when user activates it, not altering system prompt by default.
  • Full rule management: inspect, edit, lock, suppress, delete, and export rules as readable JSON with versioning.

Where the Liltloom DSH plugin fits

  • Maintain consistent writing style across multiple chat sessions without repeating style prompts.
  • Export style data for backup or migration to other compatible hosts.
  • Avoid re-copying style instructions when switching AI models or tools.

Who the Liltloom DSH plugin is for

  • Chinese content creators who want AI to mimic their personal writing style.
  • Users of DeepSeek Harness who need a privacy-respecting style memory layer.

Liltloom DSH plugin limitations

  • Currently only supports DeepSeek Harness Web profile with persistent storage (storage-domain).
  • Does not include embeddings, vector databases, fine-tuning, factual memory, or automatic personality inference.

Liltloom DSH plugin: from the repository README

Quoted from the Adkid-Zephyr/Liltloom README, the upstream source of the Liltloom DSH plugin. Copyright remains with the original authors.

> 你不应该每换一个模型、一个工具或一个新对话，就重新复制一遍“请模仿我的风格”。 Liltloom（语织）是一个**中文优先、由用户拥有的个人写作风格记忆层**。它从合格的用户原生表达中提炼语言偏好，保存为结构化、可审阅的规则；当你确实需要 AI 模仿自己时，再把与当前任务有关的内容编译成一份短小、有界的 **Style Context Packet**。 它不是另一段越来越长的万能 Prompt，也不会默认把所有聊天原文塞进数据库。你能看到它学到了什么，随时修改、锁定、忽略、删除，也可以导出为可读 JSON 带走。 名字里的 **lilt** 是语调与节奏，**loom** 是织机：平时安静收集表达中的细线，需要时再把它们织进回答。 ### 一眼看懂 | 能力 | Liltloom 的做法 | |---|---| | 自动学习 | 只观察通过资格检查的用户本人表达，过滤工具结果、助手内容、日志、代码、大片引用、疑似粘贴与秘密 | | 静默保存 | 基础模式只累计统计特征和结构化规则；普通对话不额外调用模型，也不改变 system prompt | | 显式取用 | 默认只有用户主动激活，或受控接口获得授权时，风格信息才进入写作任务 | | 人工控制 | 规则可查看、新增、编辑、锁定、忽略和删除；使用前可以预览完整上下文 | | 数据迁移 | 提供带版本的可读 JSON 导出；同一份数据协议可供不同 Agent / Harness adapter 复用 | | 精度升级 | 原文片段与小模型分析是两个独立的高阶开关，开启前明确提示存储、调用和成本影响 | ### 为什么是“中文优先” 中文写作的风格不只是一组形容词。标点是全角还是半角、单句多长、自然段如何推进、什么时候列点、标题密度，以及“直接”与“生硬”之间的距离，都会影响最终读感。 当前 MVP 已经把这些细节做进产品，而不只是翻译 README： - 确定性提取中文句末标点、全角标点占比、句长、段落、列表和标题倾向； - 自动生成可执行的中文风格规则，例如短句负担、自然分段和中文标点偏好； - 设置界面、授权说明、风险提示和数据控制以中文为主，减少术语堆砌； - 区分默认、技术、专业、社交和长文等文体，避免把一套语气强行用在所有场景； - 中文文本可以直接学习和使用，不要求先翻译成英文，也不要求用户自己编写风格 Prompt。 **中文优先不等于只支持中文。** Core 接受 Unicode 文本，Style Context 协议也不绑定语言；只是当前产品文案、默认规则和主要验收样本优先服务中文用户。英文与其他宿主可用，但成熟度暂时不与中文 + DeepSeek Harness 组合等同。 ## 界面预览 以下界面来自目前维护最完整的 DeepSeek Harness ad

Read the full READMERepository license: MIT

Liltloom DSH plugin questions

Does Liltloom save all my chat conversations?

No, in the base mode Liltloom does not save full conversation transcripts. It only stores aggregated statistical features and structured rules. Optional Deep Style mode can retain sparse excerpts from eligible user-authored text with clear storage and retention limits.

How do I enable Liltloom to automatically apply my style?

By default the take-out mode is 'explicit', meaning you need to manually activate style use via `/liltloom use` or through the UI. You can switch to 'auto' mode with `/liltloom mode auto`, but the current message's instruction 'do not use my style' will still take precedence.

Can I export my style data to use elsewhere?

Yes, Liltloom supports exporting all learned rules, settings, and data as a versioned readable JSON file. Use `/liltloom export` or the data control UI. The export format is documented in `liltloom-export.schema.json` and can be imported into other compatible hosts.

How do I completely delete my data from Liltloom?

Run `/liltloom clear confirm` in the chat, then remove the plugin with `dsh plugin --profile web remove liltloom`. This will delete the stored data in the storage-domain. If you want to keep a backup, export your data before clearing.

Does Liltloom work with languages other than Chinese?

Yes, Liltloom core accepts Unicode text and the Style Context protocol is language-agnostic. However, the product is Chinese-first: the default rules, UI, and deterministic extraction are optimized for Chinese. Other languages are supported but may have lower maturity.