[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"article-openviking-agent-context-database-zh":3,"article-related-openviking-agent-context-database-zh":33,"series-industry-5976a9eb-d24d-4848-bcc5-b7fbb6afd529":78},{"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":25,"views":29,"created_at":30,"published_at":31,"topic_cluster_id":32},"5976a9eb-d24d-4848-bcc5-b7fbb6afd529","openviking-agent-context-database-zh","OpenViking 把代理上下文收進同一個資料庫","\u003Cp>OpenViking 怎麼把 \u003Ca href=\"\u002Ftag\u002Fagent\">agent\u003C\u002Fa> 的記憶、\u003Ca href=\"\u002Ftag\u002Frag\">RAG\u003C\u002Fa> 和技能收在一起？\u003C\u002Fp>\u003Cp data-speakable=\"summary\">OpenViking 是一個把 agent 記憶、檢索和技能整合成單一 context database 的系統。\u003C\u002Fp>\u003Ctable>\u003Cthead>\u003Ctr>\u003Cth>項目\u003C\u002Fth>\u003Cth>組織內容\u003C\u002Fth>\u003Cth>關鍵特點\u003C\u002Fth>\u003C\u002Ftr>\u003C\u002Fthead>\u003Ctbody>\u003Ctr>\u003Ctd>Filesystem paradigm\u003C\u002Ftd>\u003Ctd>Memory, resources, skills\u003C\u002Ftd>\u003Ctd>像本地檔案一樣管理上下文\u003C\u002Ftd>\u003C\u002Ftr>\u003Ctr>\u003Ctd>Tiered context loading\u003C\u002Ftd>\u003Ctd>L0, L1, L2\u003C\u002Ftd>\u003Ctd>按需載入，節省 token\u003C\u002Ftd>\u003C\u002Ftr>\u003Ctr>\u003Ctd>Directory recursive retrieval\u003C\u002Ftd>\u003Ctd>Search and positioning\u003C\u002Ftd>\u003Ctd>結合目錄定位與語意搜尋\u003C\u002Ftd>\u003C\u002Ftr>\u003Ctr>\u003Ctd>Visualized retrieval trajectory\u003C\u002Ftd>\u003Ctd>Debugging context flow\u003C\u002Ftd>\u003Ctd>讓檢索路徑可觀察\u003C\u002Ftd>\u003C\u002Ftr>\u003Ctr>\u003Ctd>Automatic session management\u003C\u002Ftd>\u003Ctd>Conversation memory\u003C\u002Ftd>\u003Ctd>把 session 壓縮成長期記憶\u003C\u002Ftd>\u003C\u002Ftr>\u003C\u002Ftbody>\u003C\u002Ftable>\u003Ch2>1. 檔案系統式上下文\u003C\u002Fh2>\u003Cp>OpenViking 的出發點不是把上下文塞進一個黑箱向量庫，而是把 agent context 當成檔案系統來整理。這讓開發者能直接理解 memory、resource、skill 各自放在哪裡，也更容易維護。\u003C\u002Fp>\n\u003Cfigure class=\"my-6\">\u003Cimg src=\"https:\u002F\u002Fxxdpdyhzhpamafnrdkyq.supabase.co\u002Fstorage\u002Fv1\u002Fobject\u002Fpublic\u002Fcovers\u002Finline-1784057563908-sadm.png\" alt=\"OpenViking 把代理上下文收進同一個資料庫\" class=\"rounded-xl w-full\" loading=\"lazy\" \u002F>\u003C\u002Ffigure>\n\u003Cp>這種做法解決的是「上下文散在各處」的問題。\u003Ca href=\"https:\u002F\u002Fgithub.com\u002Fvolcengine\u002FOpenViking\">OpenViking\u003C\u002Fa> 把資料結構收斂成可讀、可查、可搬移的層次，讓 agent 的工作集\u003Ca href=\"\u002Fnews\u002Fgpt-5-6-chatgpt-codex-one-workspace-zh\">更像\u003C\u002Fa>資料夾，而不是一堆難以追蹤的 embedding。\u003C\u002Fp>\u003Cul>\u003Cli>Memory：保存使用者與任務歷史\u003C\u002Fli>\u003Cli>Resources：保存知識、參考資料與外部內容\u003C\u002Fli>\u003Cli>Skills：保存可重用的 agent 行為\u003C\u002Fli>\u003C\u002Ful>\u003Ch2>2. L0、L1、L2 分層載入\u003C\u002Fh2>\u003Cp>OpenViking 用 L0、L1、L2 三層來控制上下文載入，不會一次把所有資料都灌進\u003Ca href=\"\u002Fnews\u002Frequential-coding-model-compression-self-generated-data-zh\">模型\u003C\u002Fa>。這樣可以減少 \u003Ca href=\"\u002Ftag\u002Ftoken\">token\u003C\u002Fa> 消耗，也避免 agent 被不必要的背景資訊干擾。\u003C\u002Fp>\u003Cp>對長任務特別有用。當對話越跑越長，單純截斷容易丟掉關鍵細節；分層載入則讓系統只在需要時往更深層取資料。\u003C\u002Fp>\u003Cul>\u003Cli>L0：當前立即需要的工作上下文\u003C\u002Fli>\u003Cli>L1：附近的支援上下文\u003C\u002Fli>\u003Cli>L2：較深層或歸檔內容\u003C\u002Fli>\u003C\u002Ful>\u003Ch2>3. 目錄遞迴檢索\u003C\u002Fh2>\u003Cp>它不是只做平面的 RAG \u003Ca href=\"\u002Fnews\u002Fcloudflare-openai-pilot-fresher-search-zh\">搜尋\u003C\u002Fa>，而是加入目錄感知的遞迴檢索。系統會先用路徑做定位，再用語意搜尋縮小範圍，對巢狀結構的上下文特別實用。\u003C\u002Fp>\n\u003Cfigure class=\"my-6\">\u003Cimg src=\"https:\u002F\u002Fxxdpdyhzhpamafnrdkyq.supabase.co\u002Fstorage\u002Fv1\u002Fobject\u002Fpublic\u002Fcovers\u002Finline-1784057565908-gv0c.png\" alt=\"OpenViking 把代理上下文收進同一個資料庫\" class=\"rounded-xl w-full\" loading=\"lazy\" \u002F>\u003C\u002Ffigure>\n\u003Cp>當專案裡同時有筆記、工具、記憶與技能時，這種方法比單純向量召回更好找。它能先進到對的資料夾，再在那個分支裡找最相關的內容。\u003C\u002Fp>\u003Ccode>範例流程：\u003Cbr>project\u002F\u003Cbr>  memory\u002F\u003Cbr>  resources\u002F\u003Cbr>  skills\u002F\u003Cbr>  notes\u002F\u003Cbr>先按目錄定位，再在目標分支內做語意匹配。\u003C\u002Fcode>\u003Ch2>4. 可視化檢索軌跡\u003C\u002Fh2>\u003Cp>OpenViking 會把檢索路徑顯示出來，而不是藏在系統內部。這種 trace 能告訴你查詢走過哪些目錄、抓回了哪些上下文，除錯時比猜測黑箱 RAG 鏈路有效得多。\u003C\u002Fp>\u003Cp>如果團隊正在調整 agent 行為，這個功能很實用。答案不對時，你可以直接看路徑，找出錯誤分支，再回頭調整結構或搜尋規則，而不是盲目改 prompt。\u003C\u002Fp>\u003Cul>\u003Cli>查看 recall 查詢走過的路徑\u003C\u002Fli>\u003Cli>檢查 prompt 注入與 tool-call 歷史\u003C\u002Fli>\u003Cli>追蹤某段記憶為什麼被選中或被略過\u003C\u002Fli>\u003C\u002Ful>\u003Ch2>5. 自動 session 管理\u003C\u002Fh2>\u003Cp>OpenViking 也會處理持續進行中的 session。它能把對話內容、工具呼叫與資源引用壓縮成長期記憶，讓 agent 在下一次執行時延續先前學到的東西。\u003C\u002Fp>\u003Cp>這讓它不只是儲存層，而是會累積經驗的上下文底座。隨著任務反覆發生，系統會保留常見模式、任務歷史與有用產物，讓後續運行更穩定。\u003C\u002Fp>\u003Cul>\u003Cli>對話壓縮\u003C\u002Fli>\u003Cli>工具呼叫捕捉\u003C\u002Fli>\u003Cli>長期記憶萃取\u003C\u002Fli>\u003C\u002Ful>\u003Ch2>怎麼挑\u003C\u002Fh2>\u003Cp>如果你要做的是需要記憶、檢索和技能整合在一起的 agent，OpenViking 很合適。特別是當你希望上下文流向可追蹤、資料結構像檔案而不是向量黑箱時，它的優勢會更明顯。\u003C\u002Fp>\u003Cp>如果你的需求只是簡單 RAG，輕量方案可能就夠了；但如果你要的是可累積、可除錯、可擴充的 agent 上下文系統，OpenViking 會更值得投入。\u003C\u002Fp>","5 個重點看懂 OpenViking：它把 agent 記憶、RAG 與技能整合成一個 context database，並提供分層載入、路徑追蹤與長期記憶。","github.com","https:\u002F\u002Fgithub.com\u002Fvolcengine\u002FOpenViking",null,"https:\u002F\u002Fxxdpdyhzhpamafnrdkyq.supabase.co\u002Fstorage\u002Fv1\u002Fobject\u002Fpublic\u002Fcovers\u002Finline-1784057563908-sadm.png","industry","zh","7900970f-473f-4208-91c4-248bab8870dc",[17,18,19,20,21,22,23,24],"OpenViking","context database","agent memory","RAG","retrieval tiers","session management","retrieval tracing","AI agents",[26,27,28],"把 memory、RAG、skills 收斂到同一個 context database，降低上下文分散問題。","用 L0\u002FL1\u002FL2 分層載入與目錄遞迴檢索，兼顧 token 成本與召回精度。","可視化檢索軌跡與自動 session 管理，讓 agent 更容易除錯，也能累積長期記憶。",0,"2026-07-14T19:32:20.226876+00:00","2026-07-14T19:32:20.215+00:00","b6c3293f-d6f9-49b5-b804-64f46dd2cd76",{"tags":34,"relatedLang":37,"relatedPosts":41},[35],{"name":20,"slug":36},"rag",{"id":15,"slug":38,"title":39,"language":40},"openviking-agent-context-database-en","OpenViking turns agent context into one database","en",[42,48,54,60,66,72],{"id":43,"slug":44,"title":45,"cover_image":46,"image_url":46,"created_at":47,"category":13},"e4ad584d-4247-465d-8adb-af42d62d8e25","gpt-5-6-chatgpt-codex-one-workspace-zh","GPT-5.6 让 ChatGPT 更像工作台","https:\u002F\u002Fxxdpdyhzhpamafnrdkyq.supabase.co\u002Fstorage\u002Fv1\u002Fobject\u002Fpublic\u002Fcovers\u002Finline-1784052188042-wlg0.png","2026-07-14T18:02:33.326107+00:00",{"id":49,"slug":50,"title":51,"cover_image":52,"image_url":52,"created_at":53,"category":13},"7a749f8e-1ca9-404c-98c3-b85ced5d8b32","cloudflare-openai-pilot-fresher-search-zh","Cloudflare 與 OpenAI 的試點，押對了 AI 搜尋最缺的「新鮮…","https:\u002F\u002Fxxdpdyhzhpamafnrdkyq.supabase.co\u002Fstorage\u002Fv1\u002Fobject\u002Fpublic\u002Fcovers\u002Finline-1784035974063-4edw.png","2026-07-14T13:32:23.061902+00:00",{"id":55,"slug":56,"title":57,"cover_image":58,"image_url":58,"created_at":59,"category":13},"d9af1534-8c92-4c22-84a7-07b2d7f4f223","cloudflare-487-partner-ecosystem-partnerbase-zh","Cloudflare 的 487 個合作夥伴地圖","https:\u002F\u002Fxxdpdyhzhpamafnrdkyq.supabase.co\u002Fstorage\u002Fv1\u002Fobject\u002Fpublic\u002Fcovers\u002Finline-1784034182919-dzef.png","2026-07-14T13:02:33.437045+00:00",{"id":61,"slug":62,"title":63,"cover_image":64,"image_url":64,"created_at":65,"category":13},"d5b8a8b6-9bf6-460c-b3aa-fb4fe927a462","anthropic-access-cut-speech-limits-ai-zh","Anthropic 斷權揭露 AI 言論限制","https:\u002F\u002Fxxdpdyhzhpamafnrdkyq.supabase.co\u002Fstorage\u002Fv1\u002Fobject\u002Fpublic\u002Fcovers\u002Finline-1784017984841-630y.png","2026-07-14T08:32:37.646155+00:00",{"id":67,"slug":68,"title":69,"cover_image":70,"image_url":70,"created_at":71,"category":13},"cc1cd111-b5db-4c71-97ec-11f7d66777c5","gpt-5-6-microsoft-365-copilot-preferred-model-zh","GPT 5.6 進入 Copilot 預設核心","https:\u002F\u002Fxxdpdyhzhpamafnrdkyq.supabase.co\u002Fstorage\u002Fv1\u002Fobject\u002Fpublic\u002Fcovers\u002Finline-1784014362896-it86.png","2026-07-14T07:32:20.311917+00:00",{"id":73,"slug":74,"title":75,"cover_image":76,"image_url":76,"created_at":77,"category":13},"45eef4b4-fff9-4bbc-9860-a3820395f5c9","webx-2026-speaker-lineup-conference-brief-zh","WebX 2026 把聲量拆成會議簡報","https:\u002F\u002Fxxdpdyhzhpamafnrdkyq.supabase.co\u002Fstorage\u002Fv1\u002Fobject\u002Fpublic\u002Fcovers\u002Finline-1783928000041-ukar.png","2026-07-13T07:32:54.333855+00:00",[79,84,89,94,99,104,109,114,119,124],{"id":80,"slug":81,"title":82,"created_at":83},"ee073da7-28b3-4752-a319-5a501459fb87","ai-in-2026-what-actually-matters-now-zh","2026 AI 真正重要的事","2026-03-26T07:09:12.008134+00:00",{"id":85,"slug":86,"title":87,"created_at":88},"83bd1795-8548-44c9-9a7e-de50a0923f71","trump-ai-framework-power-speech-state-preemption-zh","川普 AI 框架瞄準電力、言論與州權","2026-03-26T07:12:18.695466+00:00",{"id":90,"slug":91,"title":92,"created_at":93},"ea6be18b-c903-4e54-97b7-5f7447a612e0","nvidia-gtc-2026-big-ai-announcements-zh","NVIDIA GTC 2026 重點拆解","2026-03-26T07:14:26.62638+00:00",{"id":95,"slug":96,"title":97,"created_at":98},"4bcec76f-4c36-4daa-909f-54cd702f7c93","claude-users-spreading-out-and-getting-better-zh","Claude 用戶更分散，也更會用","2026-03-26T07:22:52.325888+00:00",{"id":100,"slug":101,"title":102,"created_at":103},"bd903b15-2473-4178-9789-b7557816e535","openclaw-raises-hard-question-for-ai-models-zh","OpenClaw 逼問 AI 模型價值","2026-03-26T07:24:54.707486+00:00",{"id":105,"slug":106,"title":107,"created_at":108},"eeac6b9e-ad9d-4831-8eec-8bba3f9bca6a","gap-google-gemini-checkout-fashion-search-zh","Gap 把結帳搬進 Gemini","2026-03-26T07:28:23.937768+00:00",{"id":110,"slug":111,"title":112,"created_at":113},"0740e53f-605d-4d57-8601-c10beb126f3c","google-pushes-gemini-transition-to-march-2026-zh","Google 把 Gemini 轉換延到 2026 年 3…","2026-03-26T07:30:12.825269+00:00",{"id":115,"slug":116,"title":117,"created_at":118},"e660d801-2421-4529-8fa9-86b82b066990","metas-llama-4-benchmark-scandal-gets-worse-zh","Meta Llama 4 分數風波又擴大","2026-03-26T07:34:21.156421+00:00",{"id":120,"slug":121,"title":122,"created_at":123},"183f9e7c-e143-40bb-a6d5-67ba84a3a8bc","accenture-mistral-ai-sovereign-enterprise-deal-zh","Accenture 攜手 Mistral AI 賣主權 AI","2026-03-26T07:38:14.818906+00:00",{"id":125,"slug":126,"title":127,"created_at":128},"191d9b1b-768a-478c-978c-dd7431a38149","mistral-ai-faces-its-hardest-year-yet-zh","Mistral AI 迎來最硬的一年","2026-03-26T07:40:23.716374+00:00"]