[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"article-bytedance-deerflow-2-0-47k-stars-zh":3,"article-related-bytedance-deerflow-2-0-47k-stars-zh":28,"series-ai-agent-a93a9f12-827d-47a3-a2cc-9a602450b5ba":85},{"id":4,"slug":5,"title":6,"content":7,"summary":8,"source":9,"source_url":10,"author":11,"image_url":12,"cover_image":12,"category":13,"language":14,"translated_content":11,"related_article_id":15,"keywords":16,"key_takeaways":11,"views":25,"created_at":26,"published_at":27,"topic_cluster_id":11},"a93a9f12-827d-47a3-a2cc-9a602450b5ba","bytedance-deerflow-2-0-47k-stars-zh","ByteDance DeerFlow 2.0 衝上 47.…","\u003Cp>\u003Ca href=\"https:\u002F\u002Fgithub.com\u002Fbytedance\u002Fdeer-flow\" target=\"_blank\" rel=\"noopener\">DeerFlow\u003C\u002Fa> 在 GitHub 衝到 47.3K stars。這數字很誇張，真的不是隨便玩玩。它主打把 AI agent 的規劃、工具呼叫、長流程任務，整理成開發者能直接用的框架。\u003C\u002Fp>\u003Cp>講白了，就是把原本很散的 agent 工作流，收進同一套架構。你不用每次都自己拼 prompt、記憶、工具、重試邏輯。這對要做產品的人來說，省的不是一點點時間，是一堆髒活。\u003C\u002Fp>\u003Cp>這件事會紅，不意外。因為很多 agent demo 看起來很猛，一碰到搜尋、檔案、API、程式碼執行，就開始翻車。De\u003Ca href=\"\u002Fnews\u002Fanthropic-xero-ai-small-business-finance-zh\">er\u003C\u002Fa>Flow 2.0 想處理的，就是這種真實場景。\u003C\u002Fp>\u003Ch2>DeerFlow 2.0 在解什麼問題\u003C\u002Fh2>\u003Cp>\u003Ca href=\"https:\u002F\u002Fgithub.com\u002Fbytedance\u002Fdeer-flow\" target=\"_blank\" rel=\"noopener\">DeerFlow\u003C\u002Fa> 是 \u003Ca href=\"https:\u002F\u002Fwww.bytedance.com\u002F\" target=\"_blank\" rel=\"noopener\">ByteDance\u003C\u002Fa> 開源的 agent framework。它的核心不是聊天，而是長任務。像研究、整理、查資料、跑工具、再回頭修正，這類工作才是它的主戰場。\u003C\u002Fp>\n\u003Cfigure class=\"my-6\">\u003Cimg src=\"https:\u002F\u002Fxxdpdyhzhpamafnrdkyq.supabase.co\u002Fstorage\u002Fv1\u002Fobject\u002Fpublic\u002Fcovers\u002Finline-1775142408400-om8u.png\" alt=\"ByteDance DeerFlow 2.0 衝上 47.…\" class=\"rounded-xl w-full\" loading=\"lazy\" \u002F>\u003C\u002Ffigure>\n\u003Cp>你可能會想問，這跟一般 \u003Ca href=\"\u002Fnews\u002Faws-s3-sagemaker-unified-studio-fine-tuning-zh\">LLM\u003C\u002Fa> wrapper 差在哪。差在它把流程想得比較完整。模型負責推理，工具負責動作，控制層負責步驟。這三個東西如果沒分好，agent 很容易變成一個會講話的失控機器。\u003C\u002Fp>\u003Cp>官方範例裡，還能看到 \u003Ca href=\"https:\u002F\u002Fpython.langchain.com\u002Fdocs\u002Fintegrations\u002Fchat\u002Fopenai\u002F\" target=\"_blank\" rel=\"noopener\">LangChain\u003C\u002Fa> 的 ChatOpenAI 整合，以及 \u003Ca href=\"https:\u002F\u002Fopenai.com\u002Findex\u002Fgpt-4\u002F\" target=\"_blank\" rel=\"noopener\">GPT-4\u003C\u002Fa>。這很直白。它不是在做玩具，而是在對準已經有模型、也有工具需求的團隊。\u003C\u002Fp>\u003Cul>\u003Cli>GitHub stars：47.3K，討論熱度很高\u003C\u002Fli>\u003Cli>任務型態：研究、檢索、工具呼叫、多步驟執行\u003C\u002Fli>\u003Cli>模型範例：GPT-4 搭配 LangChain\u003C\u002Fli>\u003Cli>目標族群：做 agent app、內部自動化、研究助理的開發者\u003C\u002Fli>\u003C\u002Ful>\u003Ch2>為什麼開發者會買單\u003C\u002Fh2>\u003Cp>做 agent 最煩的，不是 prompt。是狀態管理。是 retry。是工具路由。是中間結果怎麼存。這些都很瑣碎，但少一個就會出事。\u003C\u002Fp>\u003Cp>DeerFlow 會讓人注意，就是因為它碰的正是這些髒活。它想把 orchestration 層、tool 層、model 層放在一起。開發者不用從零拼一套，至少可以少寫很多 glue code。\u003C\u002Fp>\u003Cp>我覺得這比那種「AI 什麼都能做」的說法實際多了。真正上線後，系統要能先查資料，再判斷要不要補資料，接著呼叫工具，最後把結果收斂成可用輸出。這才是有用的 agent。\u003C\u002Fp>\u003Cblockquote>“The future of work is going to be about working with AI that can do things for you.” — Sam Altman, OpenAI DevDay 2023\u003C\u002Fblockquote>\u003Cp>這句話很適合拿來看 DeerFlow。因為它不是在做純聊天。它在做可執行的流程。能不能把 intent 變成 action，才是 agent 框架的考題。\u003C\u002Fp>\u003Cp>ByteDance 也有一個優勢。大型產品公司開源這種框架，外界通常會先猜你們內部真的用過。這不代表品質一定好，但至少設計思路比較可能貼近實戰。\u003C\u002Fp>\u003Ch2>它跟其他 agent 框架怎麼比\u003C\u002Fh2>\u003Cp>這個賽道很擠。\u003Ca href=\"https:\u002F\u002Fwww.langchain.com\u002F\" target=\"_blank\" rel=\"noopener\">LangChain\u003C\u002Fa> 幾乎是很多團隊的預設起點。\u003Ca href=\"https:\u002F\u002Fgithub.com\u002Fmicrosoft\u002Fautogen\" target=\"_blank\" rel=\"noopener\">Microsoft AutoGen\u003C\u002Fa> 偏多 agent 協作。\u003Ca href=\"https:\u002F\u002Fwww.crewai.com\u002F\" target=\"_blank\" rel=\"noopener\">CrewAI\u003C\u002Fa> 則是角色分工很明確。\u003C\u002Fp>\n\u003Cfigure class=\"my-6\">\u003Cimg src=\"https:\u002F\u002Fxxdpdyhzhpamafnrdkyq.supabase.co\u002Fstorage\u002Fv1\u002Fobject\u002Fpublic\u002Fcovers\u002Finline-1775142408296-8wuv.png\" alt=\"ByteDance DeerFlow 2.0 衝上 47.…\" class=\"rounded-xl w-full\" loading=\"lazy\" \u002F>\u003C\u002Ffigure>\n\u003Cp>DeerFlow 的味道，介於通用 orchestration 跟偏工作流設計之間。它不像純框架那麼空，也不像封裝太死的工具那麼硬。這種位置很討喜，因為開發者通常想要彈性，但又不想從零開始。\u003C\u002Fp>\u003Cp>真正要比的，不是功能清單。是你要花多少時間，把模型、記憶、工具、回復機制全部接起來。少一天整合時間，就等於少一堆 bug。\u003C\u002Fp>\u003Cul>\u003Cli>\u003Ca href=\"https:\u002F\u002Fwww.langchain.com\u002F\" target=\"_blank\" rel=\"noopener\">LangChain\u003C\u002Fa>：生態大，整合多\u003C\u002Fli>\u003Cli>\u003Ca href=\"https:\u002F\u002Fgithub.com\u002Fmicrosoft\u002Fautogen\" target=\"_blank\" rel=\"noopener\">AutoGen\u003C\u002Fa>：多 agent 協作味道更重\u003C\u002Fli>\u003Cli>\u003Ca href=\"https:\u002F\u002Fwww.crewai.com\u002F\" target=\"_blank\" rel=\"noopener\">CrewAI\u003C\u002Fa>：角色式工作流比較直覺\u003C\u002Fli>\u003Cli>\u003Ca href=\"https:\u002F\u002Fgithub.com\u002Fbytedance\u002Fdeer-flow\" target=\"_blank\" rel=\"noopener\">DeerFlow\u003C\u002Fa>：偏 workflow，適合長任務\u003C\u002Fli>\u003C\u002Ful>\u003Cp>另外，範例直接放 \u003Ca href=\"https:\u002F\u002Fopenai.com\u002Findex\u002Fgpt-4\u002F\" target=\"_blank\" rel=\"noopener\">GPT-4\u003C\u002Fa> 也很有意思。這代表它不是只想服務本機玩具專案，而是把高品質模型輸出當成預設前提。這點對產品團隊很重要。\u003C\u002Fp>\u003Ch2>47.3K stars 到底代表什麼\u003C\u002Fh2>\u003Cp>\u003Ca href=\"\u002Fnews\u002Fphpstorm-20261-laravel-13-agent-tools-zh\">St\u003C\u002Fa>ar 數字不是 KPI，但也不是完全沒用。47.3K stars 代表它至少打到三群人：想做 agent 的開發者、愛追新工具的開源社群、還有看到 ByteDance 就先點進去的人。\u003C\u002Fp>\u003Cp>不過，star 多不等於真的會用。很多 repo 會紅一波，然後就沒下文。這在 AI 工具圈很常見。大家先收藏，再說。\u003C\u002Fp>\u003Cp>但 agent 這條線還在找標準答案。今天大家都在問同一件事：哪個框架能讓 agent 少翻車、少卡住、少靠手動補丁。DeerFlow 的熱度，某種程度就是這個問題的回音。\u003C\u002Fp>\u003Cp>如果看數據感，你也可以這樣理解：47.3K stars 已經不是小眾實驗。它代表這個專案進入高關注區。開發者會開始拿它跟現有工具做比較，而不是只當成新聞看。\u003C\u002Fp>\u003Cul>\u003Cli>47.3K stars：代表社群關注度很高\u003C\u002Fli>\u003Cli>熱度來源：ByteDance 品牌、agent 題材、開源傳播\u003C\u002Fli>\u003Cli>實際價值：看能否減少整合成本\u003C\u002Fli>\u003Cli>風險：星星很多，不代表上線穩定\u003C\u002Fli>\u003C\u002Ful>\u003Ch2>背後的產業脈絡\u003C\u002Fh2>\u003Cp>這波熱度，其實反映的是 AI 工具層正在重組。以前大家比的是模型參數。現在更常比的是 workflow、tool use、memory、eval、observability。說真的，這些才是產品落地時最痛的地方。\u003C\u002Fp>\u003Cp>在台灣，很多團隊已經不是在問「要不要接 LLM」。而是在問「怎麼讓 LLM 接內部系統後還能穩」。這就牽涉到 API、權限、資料流、日誌、回復機制。DeerFlow 這類框架，就是在吃這塊需求。\u003C\u002Fp>\u003Cp>所以它的意義，不只是又一個開源專案。它是在告訴大家，agent 的競爭已經往工程化移動。誰能把流程做得更可控，誰就比較有機會真的進到企業環境。\u003C\u002Fp>\u003Cp>從這角度看，ByteDance 開源 DeerFlow 也很合理。大公司通常最先遇到的是規模問題，不是 demo 問題。當任務變長、工具變多、失敗點變複雜，框架設計就會變得很重要。\u003C\u002Fp>\u003Ch2>我怎麼看這波熱度\u003C\u002Fh2>\u003Cp>我覺得 DeerFlow 值得看，但不要先神化。47.3K stars 很猛，沒錯。可是真正的考驗，是你把它接進產品後，會不會還是要自己補一堆例外處理。\u003C\u002Fp>\u003Cp>如果你現在就在做 agent、RAG、內部自動化，我會建議你直接跑一輪。看它的 tool routing、state handling、錯誤復原，能不能少掉你原本 30% 的樣板碼。這種東西，實測比看簡報準。\u003C\u002Fp>\u003Cp>接下來 3 到 6 個月，重點會是社群有沒有持續補文件、補範例、補 eval。只要這三件事有跟上，它就不只是熱門 repo。它會變成很多團隊評估 agent 架構時，會先丟進候選名單的名字。\u003C\u002Fp>\u003Cp>你如果問我一句話總結，我會說：DeerFlow 2.0 不是來講夢想的。它是來處理流程的。這種東西，才是真的工程師會在意的。\u003C\u002Fp>","ByteDance 的 DeerFlow 2.0 在 GitHub 衝到 47.3K stars。它想把 agent 的規劃、工具呼叫與多步驟工作流程，做成開發者更好接上的框架。","zhuanlan.zhihu.com","https:\u002F\u002Fzhuanlan.zhihu.com\u002Fp\u002F2020566695719256852",null,"https:\u002F\u002Fxxdpdyhzhpamafnrdkyq.supabase.co\u002Fstorage\u002Fv1\u002Fobject\u002Fpublic\u002Fcovers\u002Finline-1775142408400-om8u.png","ai-agent","zh","4ed4aff6-7cc8-49a7-aa35-3f68e984135e",[17,18,19,20,21,22,23,24],"ByteDance","DeerFlow","agent framework","AI agent","GitHub stars","LangChain","GPT-4","開源",7,"2026-04-02T15:06:32.612652+00:00","2026-04-02T15:06:32.471+00:00",{"tags":29,"relatedLang":44,"relatedPosts":48},[30,32,34,36,38,39,41,43],{"name":23,"slug":31},"gpt-4",{"name":22,"slug":33},"langchain",{"name":17,"slug":35},"bytedance",{"name":19,"slug":37},"agent-framework",{"name":20,"slug":13},{"name":21,"slug":40},"github-stars",{"name":18,"slug":42},"deerflow",{"name":24,"slug":24},{"id":15,"slug":45,"title":46,"language":47},"bytedance-deerflow-2-0-47k-stars-en","ByteDance's DeerFlow 2.0 Hits 47.3K Stars","en",[49,55,61,67,73,79],{"id":50,"slug":51,"title":52,"cover_image":53,"image_url":53,"created_at":54,"category":13},"83c2f8f6-3710-466e-b52c-473b811f0535","how-to-set-up-openclaw-safely-zh","如何安全架設 OpenClaw","https:\u002F\u002Fxxdpdyhzhpamafnrdkyq.supabase.co\u002Fstorage\u002Fv1\u002Fobject\u002Fpublic\u002Fcovers\u002Finline-1780549368665-1t2l.png","2026-06-04T05:02:21.26625+00:00",{"id":56,"slug":57,"title":58,"cover_image":59,"image_url":59,"created_at":60,"category":13},"0ba5b1a8-82c5-464a-bea5-9a2c8730da74","aws-devops-agent-turns-incident-chaos-into-triage-zh","AWS DevOps Agent 把事故排查變成三步","https:\u002F\u002Fxxdpdyhzhpamafnrdkyq.supabase.co\u002Fstorage\u002Fv1\u002Fobject\u002Fpublic\u002Fcovers\u002Finline-1780466689960-g1sv.png","2026-06-03T06:03:14.154923+00:00",{"id":62,"slug":63,"title":64,"cover_image":65,"image_url":65,"created_at":66,"category":13},"841eac88-b0f0-4a4c-9e1e-efc3b5c16281","kimi-k26-live-300-agent-workflows-zh","Kimi K2.6 上線：300 代理工作流","https:\u002F\u002Fxxdpdyhzhpamafnrdkyq.supabase.co\u002Fstorage\u002Fv1\u002Fobject\u002Fpublic\u002Fcovers\u002Finline-1780430574285-hqpn.png","2026-06-02T20:02:24.972179+00:00",{"id":68,"slug":69,"title":70,"cover_image":71,"image_url":71,"created_at":72,"category":13},"f0411957-bcdb-42d9-a267-3e90ae7d9cb1","how-to-take-a-sabbatical-at-openai-zh","怎麼申請 OpenAI sabbatical","https:\u002F\u002Fxxdpdyhzhpamafnrdkyq.supabase.co\u002Fstorage\u002Fv1\u002Fobject\u002Fpublic\u002Fcovers\u002Finline-1780398216422-8fi7.png","2026-06-02T11:02:25.74372+00:00",{"id":74,"slug":75,"title":76,"cover_image":77,"image_url":77,"created_at":78,"category":13},"37a5e429-4235-439c-9b05-bb377085462c","8-steps-build-production-rag-with-langchain-zh","8 步驟打造可上線的 LangChain RAG","https:\u002F\u002Fxxdpdyhzhpamafnrdkyq.supabase.co\u002Fstorage\u002Fv1\u002Fobject\u002Fpublic\u002Fcovers\u002Finline-1780178597493-4hz7.png","2026-05-30T22:02:48.14022+00:00",{"id":80,"slug":81,"title":82,"cover_image":83,"image_url":83,"created_at":84,"category":13},"e73c041b-852b-44c3-85aa-0f1e2e5848e3","ai-agents-hit-chaos-mode-claude-code-openclaw-zh","Claude Code＋OpenClaw 讓 AI 代理失控升溫","https:\u002F\u002Fxxdpdyhzhpamafnrdkyq.supabase.co\u002Fstorage\u002Fv1\u002Fobject\u002Fpublic\u002Fcovers\u002Finline-1780160576178-yqcs.png","2026-05-30T17:02:25.725767+00:00",[86,91,96,101,106,111,116,121,126,131],{"id":87,"slug":88,"title":89,"created_at":90},"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":92,"slug":93,"title":94,"created_at":95},"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":97,"slug":98,"title":99,"created_at":100},"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":102,"slug":103,"title":104,"created_at":105},"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":107,"slug":108,"title":109,"created_at":110},"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":112,"slug":113,"title":114,"created_at":115},"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":117,"slug":118,"title":119,"created_at":120},"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":122,"slug":123,"title":124,"created_at":125},"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":127,"slug":128,"title":129,"created_at":130},"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",{"id":132,"slug":133,"title":134,"created_at":135},"f2ca7720-b471-4ce5-9336-2a9ac2a876fd","amazon-bedrock-agents-multi-agent-workflows-zh","Amazon Bedrock Agents 進入多代理工作流","2026-04-01T09:30:29.945429+00:00"]