[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"article-deerflow-2-0-agent-orchestration-framework-zh":3,"tags-deerflow-2-0-agent-orchestration-framework-zh":37,"related-lang-deerflow-2-0-agent-orchestration-framework-zh":46,"related-posts-deerflow-2-0-agent-orchestration-framework-zh":50,"series-ai-agent-8593e416-9a55-4911-bbf1-d1d4319075a6":87},{"id":4,"title":5,"content":6,"summary":7,"source":8,"source_url":9,"author":10,"image_url":11,"keywords":12,"language":21,"translated_content":10,"views":22,"is_premium":23,"created_at":24,"updated_at":24,"cover_image":11,"published_at":25,"rewrite_status":26,"rewrite_error":10,"rewritten_from_id":27,"slug":28,"category":29,"related_article_id":30,"status":31,"google_indexed_at":32,"x_posted_at":10,"tweet_text":10,"title_rewritten_at":10,"title_original":10,"key_takeaways":33,"topic_cluster_id":10,"embedding":10,"is_canonical_seed":23},"8593e416-9a55-4911-bbf1-d1d4319075a6","DeerFlow 2.0：把 Agent 變工作流","\u003Cp data-speakable=\"summary\">DeerFlow 2.0 是開源 AI a\u003Ca href=\"\u002Fnews\u002Fwhy-prompt-engineering-is-dead-for-ai-agents-zh\">gent\u003C\u002Fa> 編排框架，把複雜工作拆成多個子代理來協作完成。\u003C\u002Fp>\u003Cp>說真的，這類工具很多。能真的跑長任務的，很少。\u003Ca href=\"https:\u002F\u002Fgithub.com\u002Fbytedance\u002Fdeer-flow\" target=\"_blank\" rel=\"noopener\">DeerFlow\u003C\u002Fa> 這次衝到 \u003Ca href=\"\u002Ftag\u002Fgithub\">GitHub\u003C\u002Fa> Trending 第一名，原因不難懂。它不是做一個聊天殼，而是做一套工作流。\u003C\u002Fp>\u003Cp>2.0 版本主打長時間執行。它有沙盒、持久記憶、子代理拆工，還能把中間結果壓縮。講白了，就是想讓 AI 不要一忙就失憶。\u003C\u002Fp>\u003Ctable>\u003Cthead>\u003Ctr>\u003Cth>項目\u003C\u002Fth>\u003Cth>內容\u003C\u002Fth>\u003Cth>意義\u003C\u002Fth>\u003C\u002Ftr>\u003C\u002Fthead>\u003Ctbody>\u003Ctr>\u003Ctd>版本\u003C\u002Ftd>\u003Ctd>2.0 rewrite\u003C\u002Ftd>\u003Ctd>整個 agent 流程重做\u003C\u002Ftd>\u003C\u002Ftr>\u003Ctr>\u003Ctd>授權\u003C\u002Ftd>\u003Ctd>MIT\u003C\u002Ftd>\u003Ctd>可自由修改與部署\u003C\u002Ftd>\u003C\u002Ftr>\u003Ctr>\u003Ctd>執行環境\u003C\u002Ftd>\u003Ctd>AioSandboxProvider \u002Fmnt\u002Fuser-data\u003C\u002Ftd>\u003Ctd>把檔案與 shell 操作隔離\u003C\u002Ftd>\u003C\u002Ftr>\u003Ctr>\u003Ctd>模型支援\u003C\u002Ftd>\u003Ctd>OpenAI-compatible LLMs\u003C\u002Ftd>\u003Ctd>可接多家模型供應商\u003C\u002Ftd>\u003C\u002Ftr>\u003Ctr>\u003Ctd>記憶\u003C\u002Ftd>\u003Ctd>Persistent local profile\u003C\u002Ftd>\u003Ctd>保留使用者偏好與上下文\u003C\u002Ftd>\u003C\u002Ftr>\u003C\u002Ftbody>\u003C\u002Ftable>\u003Ch2>DeerFlow 到底是什麼\u003C\u002Fh2>\u003Cp>\u003Ca href=\"https:\u002F\u002Fgithub.com\u002Fbytedance\u002Fdeer-flow\" target=\"_blank\" rel=\"noopener\">DeerFlow\u003C\u002Fa> 是一個開源 \u003Ca href=\"\u002Ftag\u002Fai-agent\">AI agent\u003C\u002Fa> orchestration framework。它的核心想法很直白。不要把所有事丟給一個模型硬扛。先拆任務，再分派給子代理，最後把結果整合回來。\u003C\u002Fp>\n\u003Cfigure class=\"my-6\">\u003Cimg src=\"https:\u002F\u002Fxxdpdyhzhpamafnrdkyq.supabase.co\u002Fstorage\u002Fv1\u002Fobject\u002Fpublic\u002Fcovers\u002Finline-1778188249264-yjzp.png\" alt=\"DeerFlow 2.0：把 Agent 變工作流\" class=\"rounded-xl w-full\" loading=\"lazy\" \u002F>\u003C\u002Ffigure>\n\u003Cp>這種設計很適合研究、寫程式、資料整理。因為這些工作不是一問一答就能結束。你會需要查資料、比對內容、跑工具、存中間狀態，還要記住前面做過什麼。\u003C\u002Fp>\u003Cp>DeerFlow 的做法比較像一個小型專案經理。它有工作流程、工具、記憶、沙盒。它不是在陪你聊天。它是在幫你把事情做完。\u003C\u002Fp>\u003Cul>\u003Cli>用 \u003Ccode>.skill\u003C\u002Fcode> 檔定義工作流。\u003C\u002Fli>\u003Cli>主代理負責拆工與協調。\u003C\u002Fli>\u003Cli>每個任務跑在隔離容器裡。\u003C\u002Fli>\u003Cli>中間步驟會被摘要壓縮。\u003C\u002Fli>\u003Cli>本地記憶可跨 session 保留。\u003C\u002Fli>\u003C\u002Ful>\u003Ch2>為什麼 2.0 重寫很重要\u003C\u002Fh2>\u003Cp>2.0 不是小修小補。它是重寫。這通常代表一件事。前一版的架構，撐不住更長、更雜的任務。這在 \u003Ca href=\"\u002Ftag\u002Fagent\">agent\u003C\u002Fa> 領域很常見。\u003C\u002Fp>\u003Cp>很多 demo 在 3 分鐘內很漂亮。真的跑到 30 分鐘，問題就來了。上下文爆掉、工具輸出太吵、狀態亂掉，最後整個流程開始飄。\u003C\u002Fp>\u003Cp>DeerFlow 對這件事下手很直接。它會把完成的步驟摘要化，把中間資料丟到磁碟，還會清掉不重要的歷史紀錄。這種設計很務實。因為長任務最怕的，不是算力不夠，而是流程失控。\u003C\u002Fp>\u003Cblockquote>“The future of AI is not just about scaling models, but also about building systems around them that can handle complex tasks reliably.” — Demis Hassabis\u003C\u002Fblockquote>\u003Cp>這句話出自 \u003Ca href=\"https:\u002F\u002Fwww.nobelprize.org\u002Fprizes\u002Fphysics\u002F2024\u002Fhassabis\u002Flecture\u002F\" target=\"_blank\" rel=\"noopener\">Demis Hassabis\u003C\u002Fa> 的 2024 諾貝爾獎演講。拿來看 DeerFlow 很剛好。因為它的重點不是模型多大，而是系統怎麼把模型管好。\u003C\u002Fp>\u003Cp>我覺得這才是現在 agent 工具的分水嶺。不是誰會講故事，而是誰能把雜事收拾乾淨。\u003C\u002Fp>\u003Ch2>實際功能有哪些\u003C\u002Fh2>\u003Cp>DeerFlow 的功能清單很像給工程師看的。它支援任何 \u003Ca href=\"\u002Ftag\u002Fopenai\">OpenAI\u003C\u002Fa>-compatible LLM，也能接 MCP server 或 Python functions。再加上內建 web search、檔案操作、bash 執行，基本上已經不是單純聊天框了。\u003C\u002Fp>\n\u003Cfigure class=\"my-6\">\u003Cimg src=\"https:\u002F\u002Fxxdpdyhzhpamafnrdkyq.supabase.co\u002Fstorage\u002Fv1\u002Fobject\u002Fpublic\u002Fcovers\u002Finline-1778188248993-r3gr.png\" alt=\"DeerFlow 2.0：把 Agent 變工作流\" class=\"rounded-xl w-full\" loading=\"lazy\" \u002F>\u003C\u002Ffigure>\n\u003Cp>它還有幾個很實用的入口。你可以從終端機操作，也可以在 Python 裡直接呼叫 \u003Ccode>DeerFlowClient\u003C\u002Fcode>。這對要把 agent 整合進現有軟體的人很重要。\u003C\u002Fp>\u003Cp>另一個亮點是 sandbox。它把檔案與 shell 行為關在隔離環境裡。這點很土，但很必要。因為 AI 一旦亂寫檔案或亂跑指令，整台伺服器就會很精彩。\u003C\u002Fp>\u003Cul>\u003Cli>可用 \u003Ca href=\"https:\u002F\u002Fgithub.com\u002Fbytedance\u002Fdeer-flow\" target=\"_blank\" rel=\"noopener\">claude-to-deerflow\u003C\u002Fa> 走 CLI。\u003C\u002Fli>\u003Cli>支援串流輸出，適合命令列工作流。\u003C\u002Fli>\u003Cli>可直接嵌入 Python 應用。\u003C\u002Fli>\u003Cli>可擴充自訂 API、檔案與 MCP 工具。\u003C\u002Fli>\u003C\u002Ful>\u003Ch2>跟一般 agent 工具有什麼差別\u003C\u002Fh2>\u003Cp>很多 agent 工具還是單輪思維。丟 prompt，回一段答案，頂多再呼叫一次工具。這種模式適合問答，不太適合長流程工作。\u003C\u002Fp>\u003Cp>DeerFlow 想處理的是另一種情境。像是研究專題、整理多份資料、掃 c\u003Ca href=\"\u002Fnews\u002Fsoderbergh-ai-lennon-doc-meta-zh\">ode\u003C\u002Fa>base、寫出一段可執行的程式，再把結果回填成報告。這些事都需要狀態管理。\u003C\u002Fp>\u003Cp>差別很明顯。單一聊天型工具重速度。DeerFlow 重流程。前者像速食。後者像你真的在開專案會議。\u003C\u002Fp>\u003Cul>\u003Cli>單輪聊天工具適合快問快答。\u003C\u002Fli>\u003Cli>DeerFlow 適合研究與多步驟任務。\u003C\u002Fli>\u003Cli>封閉平台常把流程藏起來。\u003C\u002Fli>\u003Cli>DeerFlow 讓 workflow 可見，也可改。\u003C\u002Fli>\u003C\u002Ful>\u003Cp>這裡還有一個現實問題。用這類框架，你要學會管技能、管工具、管沙盒。門檻比較高。可是一旦流程跑順，穩定度通常比亂接 prompt 高很多。\u003C\u002Fp>\u003Cp>從產業角度看，這類開源框架正在往基礎設施走。大家已經不太滿足於「能 demo」。大家要的是「能交付」。\u003C\u002Fp>\u003Ch2>開發者該怎麼看這件事\u003C\u002Fh2>\u003Cp>如果你現在也在做 agent，DeerFlow 很值得看。尤其是你已經被 context 爆掉、工具亂飛、流程中斷這些問題修理過好幾次。它的思路很像把 agent 拆成可管理的零件。\u003C\u002Fp>\u003Cp>你可以把它想成一個可編排的 AI 工作台。不是一個模型單打獨鬥，而是一組角色分工。這對研究團隊、資料團隊、內部自動化，都很有吸引力。\u003C\u002Fp>\u003Cp>接下來真正重要的，不是它功能列多長，而是社群能不能快速做出實用 \u003Ccode>.skill\u003C\u002Fcode>。如果技能生態起得來，DeerFlow 才會從「很會講」變成「真的能用」。\u003C\u002Fp>\u003Cp>我會先看兩件事。第一，長任務穩不穩。第二，整合成本高不高。只要這兩項表現不差，它就有機會變成很多團隊的預設 agent 層。\u003C\u002Fp>\u003Ch2>這波對 AI 開發代表什麼\u003C\u002Fh2>\u003Cp>\u003Ca href=\"\u002Ftag\u002Fai-工具\">AI 工具\u003C\u002Fa>現在很像經歷分工期。聊天模型負責對話。工作流框架負責執行。沙盒負責安全。記憶負責延續。這四個東西分開後，系統才比較像樣。\u003C\u002Fp>\u003Cp>DeerFlow 2.0 正好卡在這個位置。它不是要取代 LLM。它是要把 LLM 放進一個比較能做事的架構裡。這種方向很務實，也比較像工程世界會接受的樣子。\u003C\u002Fp>\u003Cp>所以，如果你在找下一個 agent 實驗專案，這套值得試。不是因為它最花俏，而是因為它把很多實際問題都先想到了。這才是\u003Ca href=\"\u002Fnews\u002Fwhy-solana-developer-hiring-should-stop-treating-skills-as-s-zh\">開發者\u003C\u002Fa>最省時間的地方。\u003C\u002Fp>\u003Cp>如果你要我下注，我會說接下來比的不是誰的 prompt 最會寫，而是誰的 workflow 最耐跑。這題，DeerFlow 已經先交了一份像樣的答案。\u003C\u002Fp>","DeerFlow 2.0 是一個開源 AI agent 編排框架，把長任務拆成子代理、沙盒執行與持久記憶，適合研究、寫程式與多步驟工作流。","medevel.com","https:\u002F\u002Fmedevel.com\u002Fdeerflow\u002F",null,"https:\u002F\u002Fxxdpdyhzhpamafnrdkyq.supabase.co\u002Fstorage\u002Fv1\u002Fobject\u002Fpublic\u002Fcovers\u002Finline-1778188249264-yjzp.png",[13,14,15,16,17,18,19,20],"DeerFlow","AI agent","workflow","open-source","orchestration framework","LLM","sandbox","MCP","zh",1,false,"2026-05-07T21:10:32.997956+00:00","2026-05-07T21:10:32.982+00:00","done","776ece3c-f8d6-4332-98ba-920f6bf56215","deerflow-2-0-agent-orchestration-framework-zh","ai-agent","5f4631b3-8bdb-45b4-9f1c-1ea6dfde1185","published","2026-05-08T09:00:14.68+00:00",[34,35,36],"DeerFlow 2.0 把 agent 從聊天工具改成工作流框架。","它靠子代理、沙盒與持久記憶處理長任務。","對研究、寫程式、資料整理這類工作更實用。",[38,40,41,42,44],{"name":39,"slug":16},"open source",{"name":14,"slug":29},{"name":15,"slug":15},{"name":13,"slug":43},"deerflow",{"name":17,"slug":45},"orchestration-framework",{"id":30,"slug":47,"title":48,"language":49},"deerflow-2-0-agent-orchestration-framework-en","DeerFlow 2.0 turns agents into workflows","en",[51,57,63,69,75,81],{"id":52,"slug":53,"title":54,"cover_image":55,"image_url":55,"created_at":56,"category":29},"38406a12-f833-4c69-ae22-99c31f03dd52","switch-ai-outputs-markdown-to-html-zh","怎麼把 AI 輸出改成 HTML","https:\u002F\u002Fxxdpdyhzhpamafnrdkyq.supabase.co\u002Fstorage\u002Fv1\u002Fobject\u002Fpublic\u002Fcovers\u002Finline-1778743243861-8901.png","2026-05-14T07:20:21.545364+00:00",{"id":58,"slug":59,"title":60,"cover_image":61,"image_url":61,"created_at":62,"category":29},"c7c69fe4-97e3-4edf-a9d6-a79d0c4495b4","anthropic-cat-wu-proactive-ai-assistants-zh","Cat Wu 談 Claude 的主動式 AI","https:\u002F\u002Fxxdpdyhzhpamafnrdkyq.supabase.co\u002Fstorage\u002Fv1\u002Fobject\u002Fpublic\u002Fcovers\u002Finline-1778735455993-gnw7.png","2026-05-14T05:10:30.453046+00:00",{"id":64,"slug":65,"title":66,"cover_image":67,"image_url":67,"created_at":68,"category":29},"e1d6acda-fa49-4514-aa75-709504be9f93","how-to-run-hermes-agent-on-discord-zh","如何在 Discord 執行 Hermes Agent","https:\u002F\u002Fxxdpdyhzhpamafnrdkyq.supabase.co\u002Fstorage\u002Fv1\u002Fobject\u002Fpublic\u002Fcovers\u002Finline-1778724655796-cjul.png","2026-05-14T02:10:34.362605+00:00",{"id":70,"slug":71,"title":72,"cover_image":73,"image_url":73,"created_at":74,"category":29},"4104fa5f-d95f-45c5-9032-99416cf0365c","why-ragflow-is-the-right-open-source-rag-engine-to-self-host-zh","為什麼 RAGFlow 是最適合自架的開源 RAG 引擎","https:\u002F\u002Fxxdpdyhzhpamafnrdkyq.supabase.co\u002Fstorage\u002Fv1\u002Fobject\u002Fpublic\u002Fcovers\u002Finline-1778674262278-1630.png","2026-05-13T12:10:23.762632+00:00",{"id":76,"slug":77,"title":78,"cover_image":79,"image_url":79,"created_at":80,"category":29},"7095f05c-34f5-469f-a044-2525d2010ce9","how-to-add-temporal-rag-in-production-zh","如何在正式環境加入 Temporal RAG","https:\u002F\u002Fxxdpdyhzhpamafnrdkyq.supabase.co\u002Fstorage\u002Fv1\u002Fobject\u002Fpublic\u002Fcovers\u002Finline-1778667053844-osvs.png","2026-05-13T10:10:30.930982+00:00",{"id":82,"slug":83,"title":84,"cover_image":85,"image_url":85,"created_at":86,"category":29},"10479c95-53c6-4723-9aaa-2fde5fb19ee7","github-agentic-workflows-ai-github-actions-zh","GitHub 把 AI 代理放進 Actions","https:\u002F\u002Fxxdpdyhzhpamafnrdkyq.supabase.co\u002Fstorage\u002Fv1\u002Fobject\u002Fpublic\u002Fcovers\u002Finline-1778551884342-8io7.png","2026-05-12T02:11:02.069769+00:00",[88,93,98,103,108,113,118,123,128,133],{"id":89,"slug":90,"title":91,"created_at":92},"4ae1e197-1d3d-4233-8733-eafe9cb6438b","claude-now-uses-your-pc-to-finish-tasks-zh","Claude 開始幫你操作電腦","2026-03-26T07:20:48.457387+00:00",{"id":94,"slug":95,"title":96,"created_at":97},"5bede67f-e21c-413d-9ab8-54a3c3d26227","googles-2026-ai-agent-report-decoded-zh","Google 2026 AI Agent 報告解讀","2026-03-26T11:15:22.651956+00:00",{"id":99,"slug":100,"title":101,"created_at":102},"2987d097-563f-46c7-b76f-b558d8ef7c2b","kimi-k25-review-stronger-still-not-legend-zh","Kimi K2.5 評測：更強，但還不是神作","2026-03-27T07:15:55.277513+00:00",{"id":104,"slug":105,"title":106,"created_at":107},"95c9053b-e3f4-4cb5-aace-5c54f4c9e044","claude-code-controls-mac-desktop-zh","Claude Code 也能操控 Mac 了","2026-03-28T03:01:58.58121+00:00",{"id":109,"slug":110,"title":111,"created_at":112},"dc58e153-e3a8-4c06-9b96-1aa64eabbf5f","cloudflare-100x-faster-ai-agent-sandbox-zh","Cloudflare 的 AI 沙箱跑超快","2026-03-28T03:09:44.142236+00:00",{"id":114,"slug":115,"title":116,"created_at":117},"1c8afc56-253f-47a2-979f-1065ff072f2a","openai-backs-isara-agent-swarm-bet-zh","OpenAI 挺 Isara 的 agent swarm …","2026-03-28T03:15:27.513155+00:00",{"id":119,"slug":120,"title":121,"created_at":122},"7379b422-576e-45df-ad5a-d57a0d9dd467","openai-plan-automated-ai-researcher-zh","OpenAI 想做自動化 AI 研究員","2026-03-28T03:17:42.090548+00:00",{"id":124,"slug":125,"title":126,"created_at":127},"48c9889e-86df-450b-a356-e4a4b7c83c5b","harness-engineering-ai-agent-reliability-2026-zh","駕馭工程：從「馬具」到「作業系統」，AI Agent 可靠性的終極密碼","2026-03-31T06:42:53.556721+00:00",{"id":129,"slug":130,"title":131,"created_at":132},"e41546b8-ba9e-455f-9159-88d4614ad711","openai-codex-plugin-claude-code-zh","OpenAI 把 Codex 放進 Claude Code","2026-04-01T09:21:54.687617+00:00",{"id":134,"slug":135,"title":136,"created_at":137},"96d8e8c8-1edd-475d-9145-b1e7a1b02b65","mcp-explained-from-prompts-to-production-zh","MCP 怎麼把提示詞變工作流","2026-04-01T09:24:39.321274+00:00"]