MIT 授權、本地優先的跨平台內容發現 Agent；透過伴生 DSH 用戶端外掛（dsh-openbiliclaw）以第四欄面板 + 22 個 Agent Bridge 工具的方式接入 DeepSeek Harness。
DSH 整合
相容
作者聲明
安全稽核
未稽核
最後核實
2026-08-21
授權條款
MIT
01它能幫你完成什麼？
Use OpenBiliClaw's cross-platform personalized content discovery directly inside DeepSeek Harness
The OpenBiliClaw panel (recommendations / content library / chat / profile / settings) embedded as DSH's 4th column, plus 22 Agent Bridge tools that let DSH agents read recommendations, answer interest probes and close the learning loop
DeepSeek Harness users who want a personalized, local-first content feed alongside their agent work
Discover content across Bilibili, Xiaohongshu, Douyin, YouTube, X, Zhihu, Reddit, Linux.do, Bangumi, V2EX, Weibo and the open web from a single local agent
Cross-platform recommendations with friend-style explanations, unified into one candidate pool with diversity quotas and dedup
Users whose interests are fragmented across platforms and who want one recommendation agent instead of each platform's own feed
Keep the recommendation profile and all data fully local while tuning it via chat
A five-layer soul profile (MBTI, cognitive style, deep needs) and recommendation history stored only in local SQLite, updated by Socratic chat and feedback
Privacy-conscious users who refuse cloud recommendation profiles and want to own their data
02如何將外掛接入 DeepSeek Harness？
先決條件
- DeepSeek Harness installed (the DSH client plugin runs inside it)
- OpenBiliClaw backend deployed locally — desktop installers for macOS / Windows from Latest Release, or Python 3.11+ for script / manual deployment
- Your own LLM API key (LLM calls use your key by default); optional local Ollama + bge-m3 embedding
- A Chrome-compatible browser (Chrome / Edge / Brave / Arc / Vivaldi / Opera) or Safari (macOS) for the browser extension that connects platform login sessions
安裝步驟
- 01
Deploy the OpenBiliClaw backend: download the desktop installer (macOS .dmg / Windows .exe) from the Latest Release page, or paste the one-line prompt from the README to an AI coding assistant for a customizable deployment
- 02
Install the DSH client plugin from the companion repository github.com/whiteguo233/dsh-openbiliclaw to bring OpenBiliClaw into DeepSeek Harness
- 03
In the browser with the extension installed, log in to Bilibili (default source) or opt in to Xiaohongshu / Douyin / YouTube / X / Zhihu / Reddit / Linux.do / V2EX / Weibo to initialize your profile
- 04
The DSH panel then works through the local backend API, with the 4th column and 22 Agent Bridge tools available in DSH
驗證整合成功
作者未說明
03DSH 整合程度與能力邊界
Installed inside DeepSeek Harness via the companion DSH client plugin (dsh-openbiliclaw), which embeds the OpenBiliClaw panel as DSH's 4th column and registers 22 Agent Bridge tools
Cross-platform content discovery
Your logged-in platform sessions (via the browser extension) and opt-in source selection→Candidates from Bilibili / Xiaohongshu / Douyin / YouTube / X / Zhihu / Reddit / Linux.do / Bangumi / V2EX / Weibo / open web unified into one evaluated pool
Reads content from external platform sites using your existing browser login sessionsSends relevant content to the LLM / embedding providers you configureFive-layer soul profile with interest probes
Your cross-platform behavior, feedback and chat→A deep profile (MBTI, cognitive style, deep needs) plus proactive interest / avoidance probes that only become filters after confirmation
Writes profile and memory data to local SQLiteAgent Bridge for agent hosts (incl. DSH)
An agent host such as DeepSeek Harness / OpenClaw / Claude Code calling the versioned Agent Bridge→Proactive recommendations, interest probes, multi-turn durable chat, profile reads, on-demand multi-source recommendations and idempotent feedback write-back
Local-first storage & BYO LLM
Your own LLM API key (or experimental Codex CLI OAuth reuse); optional local Ollama + bge-m3 embedding→All behavior, recommendation, chat and profile data kept in local SQLite; no OpenBiliClaw-operated cloud account
Slim installers auto-download the bge-m3 vector model (~1.1GB) on first launchRequests are sent to the LLM / embedding endpoints you configure
04適合誰？何時不該用？
適合
- DeepSeek Harness users who want a personalized, local-first content feed alongside their agent work
- Users whose interests are fragmented across platforms and who want one recommendation agent instead of each platform's own feed
- Privacy-conscious users who refuse cloud recommendation profiles and want to own their data
不適合
- The DSH integration is not built into this repository: it ships as a separate client plugin in github.com/whiteguo233/dsh-openbiliclaw, and the OpenBiliClaw local backend must be running for the DSH panel to work.
- Desktop installers are experimental pre-releases: ad-hoc signed and not notarized, so macOS may block first launch and Windows SmartScreen may warn; the project itself notes this channel suits quick trials, not development.
05相容性、維護與安全提醒
- The DSH integration is not built into this repository: it ships as a separate client plugin in github.com/whiteguo233/dsh-openbiliclaw, and the OpenBiliClaw local backend must be running for the DSH panel to work.
- Desktop installers are experimental pre-releases: ad-hoc signed and not notarized, so macOS may block first launch and Windows SmartScreen may warn; the project itself notes this channel suits quick trials, not development.
- The system relies on your own LLM key by default (built-in Gemini / DeepSeek / OpenAI / Claude / OpenRouter / Ollama or any OpenAI-compatible service), so model quota and cost are on you; background LLM budget caps and embedding circuit breaking exist precisely because unattended runs can burn quota.
MIT · actively maintained (latest release openbiliclaw-v0.3.208, 2026-08-18)
06常見問題
OpenBiliClaw 如何接入 DeepSeek Harness？
透過獨立的 DSH 用戶端外掛（github.com/whiteguo233/dsh-openbiliclaw）實現：它把 OpenBiliClaw 裝進 DeepSeek Harness，介面常駐第四欄（推薦 / 內容庫 / 對話 / 輪廓 / 設定），並註冊 22 個 Agent Bridge 工具，DSH 裡的 Agent 可以讀推薦、答探測、把回饋寫回學習閉環。
是原生整合還是 MCP？
屬於 DSH 用戶端外掛，而非內建原生執行階段：DSH 面板只呼叫本機 OpenBiliClaw 後端 API，平台登入工作階段與 Cookie 同步仍由瀏覽器外掛承擔。
接入需要什麼前置條件？
需要已安裝 DeepSeek Harness、本機部署完成的 OpenBiliClaw 後端（macOS / Windows 桌面安裝包，或 Python 3.11+ 腳本 / 手動部署）、你自己的 LLM API Key，以及裝有 OpenBiliClaw 瀏覽器外掛的 Chrome 系瀏覽器（或 macOS Safari）用於連結內容平台。
我的資料會流向哪裡？
預設資料流是：瀏覽器外掛 → 你設定的本機 OpenBiliClaw 後端 → 本機 SQLite，不會傳送到開發者營運的伺服器。如果你設定了雲端 LLM / embedding 服務，相關內容會依你的設定傳送給對應服務商。
哪些平台需要登入？
B 站、小紅書、抖音、YouTube、X、知乎、Reddit 透過外掛複用你瀏覽器裡的現有登入工作階段；Linux.do、Bangumi、V2EX 與微博可匿名進行公開發現，個人訊號初始化則需要已登入的工作階段或公開使用者名稱。
07相關的 DSH 工作流程
open-design
作者 nexu-io
DeepSeek Harness 的開源 AI 設計外掛，本地優先桌面應用，讓編碼智慧體變身設計引擎，生成原型、落地頁、幻燈片、圖片與影片，支援多種格式匯出。
picgo
作者 molunerfinn
PicGo：面向高效創作者的圖片上傳神器，支援 Obsidian、Typora、VS Code 等編輯器，以及 S3、GitHub、Cloudflare R2、Imgur、阿里雲 OSS 等 60+ 圖床，貼上即上傳。
dsh-desktop
作者 anywhere-labs
為 DeepSeek Harness (DSH) 外掛生態打造的現代化桌面端解決方案。萬物皆「外掛」，桌面本身也是「外掛」。
deepseek-harness-desktop
作者 dsh-tauri-desk
DeepSeek Harness Tauri 桌面版 | Only 5mb installer, zero environment setup, preset plugins, Windows / macOS / Linux.
08資料與來源
新增 **DSH 客户端插件** —— 把 OpenBiliClaw 装进 [DeepSeek Harness](https://github.com/deepseek-ai/DeepSeek-Harness)：DSH 界面常驻第四栏（推荐…
以及把同一套面板搬进 DSH Web 界面的 [DSH 客户端插件](https://github.com/whiteguo233/dsh-openbiliclaw)（第四栏 + 22 个 Agent Bridge 工具）。桌面端、移动端…
此頁面根據專案公開文件、儲存庫中繼資料與 DSH Plugins 的結構化解析所產生；最後核實於 2026-08-21。發現錯誤？提交更正。
最佳 DeepSeek Harness 外掛
從全目錄挑出的 12 個值得優先安裝的外掛，涵蓋各個分類。