Evidence snapshot reviewed Sep 16, 2026GitHub checked Aug 21, 2026
Source-reviewedApplicationAutomation & AgentsDesktop App Users ProfileLocal First Ai Users ProfileDocument Workflows Profile

Penglai

A local-first desktop distribution of DeepSeek Harness with document tools and workspace-scoped memory.

At a glance

What it does

A local-first desktop distribution of DeepSeek Harness with document tools and workspace-scoped memory.

Capabilities
Automation & AgentsAgentsFilesMemory

Before you choose it

Penglai packages DeepSeek Harness as a desktop app with an installer and guided setup. The public 0.6.1 release supports Apple Silicon Macs, Windows x64, and UnionTech UOS 20 LoongArch. Office and workspace memory start enabled; messaging, speech, voice generation, and Companion are opt-in.

Best for

People who want a desktop AI workspace without assembling an agent stack, especially for file-based office work and separated project contexts.

Common tasks

  • Set up a model key and verify setup by receiving a real model reply.
  • Read, create, preview, edit, and export DOCX, XLSX, PPTX, and PDF files.
  • Keep project memory within the selected Workspace and choose personal memory separately.
  • Optionally connect supported messaging services or enable local transcription and voice features.

Permissions and data

Penglai has no stated account, telemetry backend, or cloud-memory service, but model requests send task context to the provider you select.

Permissions
  • Document writes require confirmation for the specific action.
  • Optional iMessage support on macOS requires Full Disk Access and Messages automation.
  • Plugins share the local DeepSeek Harness process; review catalog permissions before installing plugins.
Data handling
  • Project memory is scoped to the current Workspace.
  • Personal memory requires a separate choice.
  • Default uninstall preserves user data and external Workspaces.
External services
  • A user-selected model provider receives the context needed to complete model requests.
  • Messaging integrations are optional and disabled by default.
Credentials
  • You provide your own model key.
  • Credentials are stored in app-private DSH YAML, not Keychain or hardware-backed isolation.

Limitations

  • The source snapshot describes a 0.6.2 release candidate; no public 0.6.2 download is available.
  • macOS is not notarized and Windows installers are not Authenticode-signed.
  • UOS functional testing remains marked post-release, and its bundled Electron/Chromium runtime is no longer maintained.
  • Intel Macs, Linux amd64, and Windows ARM are not targets for the public 0.6.1 release.

What DSHub checked

  • The repository documents public 0.6.1 installers and SHA-256 values for the stated platforms.
  • The repository is MIT licensed.
  • The snapshot is pinned to commit 8e4fa41053e5eda7746677df425bf399511472eb.

What DSHub did not check

  • Installation, model connectivity, document editing, messaging, speech, and plugin behavior were not executed during this curation.
  • Runtime compatibility with a particular DeepSeek Harness version is not declared in the supplied evidence.

Pinned install

Primary action

This application does not have a DSH Plugin install action. Use its source documentation for the delivery method.

Visit the source project

Maintainer source

Project README

View at commit 8e4fa41
Maintainer-authored contentCaptured from README.md on Sep 12, 2026. The text and repository-relative media are fixed to commit 8e4fa41053e5 with content hash 88482ad5539d; provider-hosted badges may update independently. README commands are upstream documentation; use the type-correct primary action above and verify it against this pinned source.
<p align="center"> <img src="website/art/readme-hero-060.png" width="100%" alt="Penglai 蓬莱: DeepSeek Harness, on your computer"> </p>

Penglai · 蓬莱

Your own AI workspace, ready to install.

English · 中文 · Website · 中文网站 · Security

Penglai 0.6.1 is the current public release. It brings official DeepSeek Harness to Apple Silicon, Windows x64, and UnionTech UOS 20 LoongArch in a normal desktop package. Office and Memory are ready from the first launch. Messaging, speech, and Companion wait until you turn them on.

Download 0.6.1 · Release notes · Security

Source on this branch is the 0.6.2 release candidate, built around the complete official DSH 0.1.5-rc.2 cohort. There is no public 0.6.2 download yet. Its acceptance record is in docs/0.6.2/.

<p align="center"> <img src="website/shots/0.5.5/welcome.png" width="78%" alt="Penglai first-run welcome"> </p> <p align="center"><sub>Installed 0.5.5 first-run screen, kept as a UI reference.</sub></p>

<a id="english"></a>

English

Meet Penglai

Penglai is a desktop distribution of DeepSeek Harness. DSH remains the one agent core: it owns models, tools, approvals, Workspaces, Sessions, and the conversation. Penglai gives that core an installer, a guided first launch, careful local paths, updates you approve, and a small set of practical plugins.

In other words, you do not need to assemble an agent stack before you can ask it to do useful work.

Start with one real reply

Install the package for your computer, bring your own model key, and follow the seven-step guide. Setup is not called finished until the model has answered. You can go back, retry a failed key, change the Workspace, or close the app and continue later.

Once inside, the pieces have straightforward jobs:

Included surface Starts What it is for
Penglai Office On Read, create, preview, edit, and export DOCX, XLSX, PPTX, and PDF
Penglai Memory On Keep project memory inside the current Workspace; personal memory is a separate choice
Mobile Messaging Off Connect Weixin, Feishu, DingTalk, WeCom, QQ, Slack, Telegram, or Discord
Speech Recognition Off Local SenseVoice transcription
Voice Generation Off Local MOSS-TTS on supported Mac and Windows systems
Companion Off Optional scheduled contact with quiet hours and daily limits

Attach a file in chat when you want the assistant to read it with tools. Use Office when you want a document created, previewed, or exported. Writes wait for confirmation tied to that action.

Memory follows the Workspace you chose. One project cannot quietly search another. Personal memory is never folded in without a separate choice.

<p align="center"> <img src="website/shots/0.5.5/plugin-center.png" width="48%" alt="Penglai Plugin Center"> <img src="website/shots/0.5.5/office.png" width="48%" alt="Penglai Office"> </p> <p align="center"><sub>Installed 0.5.5 Plugin Center and Office screens, shown as product references.</sub></p>

Download 0.6.1

GitHub Releases is the authoritative source. Choose the package that matches the computer and verify it against SHA256SUMS on the v0.6.1 release. All three installers were built from source 7ad7c29ecda5d9e867fdde0fe3fefd5c42d3ab1b.

Apple Silicon, macOS 13+

Penglai_0.6.1_macos_aarch64.dmg

284,369,412 bytes · SHA-256 91393e2e760871694e836c2582b903f6fee509e9fa11391724567e472f9946e9

Windows 10+ x64

Penglai_0.6.1_windows_x64_setup.exe

371,266,017 bytes · SHA-256 d4fb670edea847024abcb2a1ac760cc2f7c5814d0c60f77f80ec647c59f51791

UnionTech UOS 20 LoongArch

Penglai_0.6.1_uos_loong64.deb

293,074,470 bytes · SHA-256 cd38c06e37151f226e7f9eb2519dba0800fc9ed7f4449d2b7edd5a282efb5934

A few things worth knowing

  • There is no Penglai account, Penglai telemetry backend, or cloud memory service. Model requests still send the context needed for the task to the provider you chose.
  • Credentials live in app-private DSH YAML. This is not Keychain or hardware isolation.
  • macOS is ad-hoc signed and not notarized. Windows has no Authenticode. Do not disable system security to install.
  • Plugins share the local DSH process. Install only reviewed catalog entries and read their permissions.
  • iMessage is optional private text on macOS only, default off, and still LIVE_NOT_RUN. It requires Full Disk Access and Messages automation.
  • The UOS package is complete, but native install, startup, UI, file picker, sleep/resume, and functional use remain OWNER_POST_RELEASE. UOS requires libatomic1; bubblewrap is recommended. MOSS-TTS is unavailable on LoongArch. Its Electron 31.7.7 / Chromium 126 runtime is no longer maintained.
  • Default uninstall removes the app and cache while preserving user data and external Workspaces. Updates are never silent.

Why Penglai

The name comes from the Eight Immortals crossing the sea, each relying on a different skill. Office, memory, messaging, and voice have different jobs too, but they meet around one DSH core.

I spent more than ten years around networking, security, and operations, but I was not a software developer when this project began. Penglai started with one stubborn idea: an ordinary person should be able to reach a capable assistant from the computer already on the desk.

Version 0.5 made the project simpler and stronger. Instead of building another agent, Penglai became a distribution of a good open-source core. That remains the promise.

Build and contribute

This repository uses Node 22.23.2 and pnpm 11.7.0.

corepack enable
pnpm install --frozen-lockfile
pnpm typecheck
pnpm test:unit
pnpm test:contract
pnpm test:integration
pnpm test:e2e
pnpm test:security
pnpm verify:contracts
pnpm verify:dependencies
pnpm audit:secrets

Native packages must be built on their matching hosts. Start with CONTRIBUTING.md. Give an AI coding tool AGENTS.md before it works in this repository.

Penglai is created and maintained by Kevin Chen / 陈克文. It is built with DeepSeek Harness, Electron, Node.js, TypeScript, pnpm, SenseVoice, sherpa-onnx, MOSS-TTS-Nano, Mnemon, Lark SDK, openclaw-weixin, PPTFast, ExcelJS, pdf-lib, Noto CJK, and the work of many contributors. DSH-IM and the QQ Bot SDK informed reviewed rewrites inside Penglai; neither runtime is bundled wholesale. Every dependency keeps its own license.

<a id="中文"></a>

中文

认识一下蓬莱

蓬莱是 DeepSeek Harness 的桌面发行版。DSH 始终是唯一的 Agent 核心,负责模型、 工具、审批、Workspace、Session 和会话;蓬莱把安装、首次引导、本地目录、升级、 卸载和一组实用插件收拾妥当。

你不必先学会拼一套 Agent 工程,才开始让它做事。

从第一条真实回复开始

下载适合这台电脑的安装包,填入自己的模型 Key,跟着七步引导往下走。模型没有真的 回复,蓬莱就不会把设置算作完成。Key 不对可以重试,Workspace 可以重选,关掉窗口 以后也能继续。

办公和记忆默认开启。办公可以读取、创建、预览、修改和导出 DOCX、XLSX、PPTX、 PDF,写入前会针对这次动作确认。项目记忆只留在当前 Workspace;个人记忆要你另外 选择。微信、飞书、钉钉、企微、QQ、Slack、Telegram、Discord、语音和主动陪伴都 默认关闭,需要时再打开,不影响普通会话。

聊天里可以直接附上文件,让助手用工具读取。能不能理解图片内容,取决于你选择的 模型是否支持视觉。

<p align="center"> <img src="website/shots/0.5.5/memory.png" width="48%" alt="蓬莱记忆"> <img src="website/shots/0.5.5/privacy.png" width="48%" alt="蓬莱首次隐私说明"> </p> <p align="center"><sub>0.5.5 安装版的记忆与隐私界面,作为产品参考。</sub></p>

下载 0.6.1

当前公开版是 v0.6.1。 请按电脑选择 Apple 芯片、Windows x64 或统信 UOS 20 龙芯安装包,下载后用发行页 里的 SHA256SUMS 核对。三个安装包来自同一源码 7ad7c29ecda5d9e867fdde0fe3fefd5c42d3ab1b。Intel Mac、Linux amd64 和 Windows ARM 不是本版目标。

安装前值得知道

  • 没有蓬莱账号、蓬莱遥测后台或云端记忆同步。模型请求仍会把完成任务需要的上下文 发给你选择的模型供应商。
  • 凭据保存在应用私有的 DSH YAML 中,不是 Keychain,也不是硬件隔离。
  • macOS 安装包是 ad-hoc 签名,未经公证;Windows 没有 Authenticode。不要为了 安装而关闭系统安全功能。
  • iMessage 只支持 Mac 私聊文本,默认关闭,需要完全磁盘访问与 Messages 自动化, 真机 live 仍是 LIVE_NOT_RUN
  • UOS 是完整安装包,但真机安装、启动、界面、文件选择器、休眠恢复和功能仍是 OWNER_POST_RELEASE。系统需要 libatomic1,建议安装 bubblewrap; 龙芯不提供 MOSS-TTS。Electron 31.7.7 / Chromium 126 已不再维护。
  • 默认卸载会移除应用和缓存,保留用户数据与外部 Workspace;升级不会静默进行。

0.6.2 正在做什么

当前源码已经切换到完整的官方 DSH 0.1.5-rc.2 依赖组,并恢复 Apple 芯片和 Windows x64 从 0.6.1 升级的安装版验收。0.6.2 还没有公开下载;只有不可变发行 附件完成发布并回读后,这里才会换成真实的大小、哈希和下载地址。

更多资料: 产品 · 架构 · 插件中心 · 安全说明 · 0.6.2 验收增量

Operate deliberately

Install and manage

Prerequisites and target Profile

Target Desktop App Users Profile, Local First Ai Users Profile, Document Workflows Profile

Delivery External App — https://github.com/kevinchennewbee/PenglaiAgent/releases/tag/v0.6.1

Compatibility and access

Public 0.6.1 installers target Apple Silicon, Windows x64, and UOS LoongArch Not declared in supplied evidence

Review compatibility evidence

Risk facts

Credentials

Model credentials are stored in app-private DSH YAML rather than Keychain or hardware-backed storage.

Evidence
Installer Signing

The macOS installer is ad-hoc signed and not notarized; the Windows installer has no Authenticode signature.

Evidence
Evidence and editorial reviewManifest, Bundle patch, distribution and freshness

Immutable evidence

Review status and source activity

AI reviewed

Use the public 0.6.1 release rather than treating this repository snapshot as a downloadable 0.6.2 release.

AI reviewed Sep 12, 2026, 1:54 PM UTCGitHub facts last checked Sep 12, 2026, 1:54 PM UTC

No material source change has been recorded since this evidence baseline.

Next step

Compare ecosystem artifact types

Subscribe to material changes for Penglai