[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"article-llm-wiki-compiler-raw-sources-to-wiki-zh":3,"article-related-llm-wiki-compiler-raw-sources-to-wiki-zh":32,"series-industry-4ea87c5d-e6f5-4b0b-8725-886c8bf90724":83},{"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":24,"views":28,"created_at":29,"published_at":30,"topic_cluster_id":31},"4ea87c5d-e6f5-4b0b-8725-886c8bf90724","llm-wiki-compiler-raw-sources-to-wiki-zh","7 個把原始資料編成可追溯 Wiki 的功能","\u003Cp data-speakable=\"summary\">llm-wiki-compiler 能把筆記、文件和論文編成可追溯的連結式 wiki，方便後續查證、維護與交給代理人使用。\u003C\u002Fp>\n\u003Cp>如果你正在評估一套「從原始資料到\u003Ca href=\"\u002Fnews\u002Ffine-tuning-slms-turns-enterprise-ai-practical-zh\">可用\u003C\u002Fa>知識庫」的工具，這 7 項功能足以幫你判斷：它適不適合拿來做研究整理、團隊文件，或是長期維護的內部百科。\u003C\u002Fp>\n\u003Ctable>\u003Cthead>\u003Ctr>\u003Cth>項目\u003C\u002Fth>\u003Cth>主要效益\u003C\u002Fth>\u003Cth>關鍵細節\u003C\u002Fth>\u003C\u002Ftr>\u003C\u002Fthead>\u003Ctbody>\u003Ctr>\u003Ctd>編譯後 wiki\u003C\u002Ftd>\u003Ctd>結構化輸出\u003C\u002Ftd>\u003Ctd>型別頁面與引用\u003C\u002Ftd>\u003C\u002Ftr>\u003Ctr>\u003Ctd>混合檢索\u003C\u002Ftd>\u003Ctd>更穩定找證據\u003C\u002Ftd>\u003Ctd>Embeddings、BM25、圖擴展\u003C\u002Ftd>\u003C\u002Ftr>\u003Ctr>\u003Ctd>本機檢視器\u003C\u002Ftd>\u003Ctd>方便瀏覽稽核\u003C\u002Ftd>\u003Ctd>搜尋、圖譜、引用標記\u003C\u002Ftd>\u003C\u002Ftr>\u003Ctr>\u003Ctd>評測框架\u003C\u002Ftd>\u003Ctd>品質檢查\u003C\u002Ftd>\u003Ctd>健康分數與回歸差異\u003C\u002Ftd>\u003C\u002Ftr>\u003Ctr>\u003Ctd>MCP 伺服器\u003C\u002Ftd>\u003Ctd>代理人可用\u003C\u002Ftd>\u003Ctd>Claude Desktop、Cursor context pack\u003C\u002Ftd>\u003C\u002Ftr>\u003C\u002Ftbody>\u003C\u002Ftable>\n\u003Ch2>1. 編譯後的 wiki 輸出\u003C\u002Fh2>\n\u003Cp>\u003Ca href=\"https:\u002F\u002Fgithub.com\u002Fatomicstrata\u002Fllm-wiki-compiler\">atomicstrata\u002Fllm-wiki-compiler\u003C\u002Fa> 的核心不是摘要，而是把零散來源編成能長期保存的 wiki。它會把內容整理成 concept、entity、comparison、overview 這類型別頁面，並把段落與主張層級的引用連回原始行號。\u003C\u002Fp>\n\u003Cfigure class=\"my-6\">\u003Cimg src=\"https:\u002F\u002Fxxdpdyhzhpamafnrdkyq.supabase.co\u002Fstorage\u002Fv1\u002Fobject\u002Fpublic\u002Fcovers\u002Finline-1781389970790-qaws.png\" alt=\"7 個把原始資料編成可追溯 Wiki 的功能\" class=\"rounded-xl w-full\" loading=\"lazy\" \u002F>\u003C\u002Ffigure>\n\n\u003Cp>這種輸出適合要留下可讀、可查、可再利用成果的人，而不是只想要一次性答案的人。它受到 Karpathy 的 LLM Wiki 概念啟發，但多了明確的 provenance 與頁面型別，後續交接會更順。\u003C\u002Fp>\n\u003Cul>\n  \u003Cli>輸入：notes、docs、papers、READMEs、ADRs\u003C\u002Fli>\n  \u003Cli>輸出：互相連結的 markdown 頁面\u003C\u002Fli>\n  \u003Cli>引用：claim-level source line ranges\u003C\u002Fli>\n\u003C\u002Ful>\n\u003Ch2>2. 混合檢索流程\u003C\u002Fh2>\n\u003Cp>它不是只靠向量相似度找資料。流程會先用可增量、可追蹤內容雜湊的 embeddings 篩出候選，再用 BM25 重新排序，最後用 wikilink graph 擴展上下文，組成更完整的 evidence pack。\u003C\u002Fp>\n\u003Cp>這個設計對資料量漸增的專案特別有用。當來源不再只是幾個檔案時，單靠語意相似度很容易漏掉關鍵脈絡，混合檢索能把精準度和召回率拉回平衡。\u003C\u002Fp>\n\u003Cul>\n  \u003Cli>semantic chunk embeddings 做 top-K 收斂\u003C\u002Fli>\n  \u003Cli>BM25 reranking 補 lexical precision\u003C\u002Fli>\n  \u003Cli>graph expansion 補連結上下文\u003C\u002Fli>\n\u003C\u002Ful>\n\u003Ch2>3. 本機瀏覽與稽核介面\u003C\u002Fh2>\n\u003Cp>\u003Ca href=\"https:\u002F\u002Fgithub.com\u002Fatomicstrata\u002Fllm-wiki-compiler\">llmwiki view\u003C\u002Fa> 提供只讀的瀏覽器介面，讓你直接看編譯後的 wiki。側欄導覽、搜尋、force-directed graph，還有每頁的 provenance chips，都讓查證來源變得直觀。\u003C\u002Fp>\n\u003Cfigure class=\"my-6\">\u003Cimg src=\"https:\u002F\u002Fxxdpdyhzhpamafnrdkyq.supabase.co\u002Fstorage\u002Fv1\u002Fobject\u002Fpublic\u002Fcovers\u002Finline-1781389969075-w5mu.png\" alt=\"7 個把原始資料編成可追溯 Wiki 的功能\" class=\"rounded-xl w-full\" loading=\"lazy\" \u002F>\u003C\u002Ffigure>\n\n\u003Cp>這一層很重要，因為它把輸出從機器可讀的資料，變成真人也能審閱的知識庫。你不必翻 JSON 或 log，就能像逛內部百科一樣檢查內容。\u003C\u002Fp>\n\u003Cul>\n  \u003Cli>只讀瀏覽器介面\u003C\u002Fli>\n  \u003Cli>側欄導覽與站內搜尋\u003C\u002Fli>\n  \u003Cli>圖譜視圖與引用標記\u003C\u002Fli>\n\u003C\u002Ful>\n\u003Ch2>4. 評測框架與健康檢查\u003C\u002Fh2>\n\u003Cp>專案內建 \u003Ca href=\"https:\u002F\u002Fgithub.com\u002Fatomicstrata\u002Fllm-wiki-compiler\">llmwiki eval\u003C\u002Fa>，會從 0 到 100 評估 wiki health，並回報 citation coverage、precision 與 regression deltas。它也支援 LLM-as-judge，適合放進 CI 做閾值檢查。\u003C\u002Fp>\n\u003Cp>對團隊來說，這是很實際的保護網。你可以在變更擴散前知道新一輪 ingest 是改善了內容，還是把引用弄壞，或是悄悄讓品質下滑。\u003C\u002Fp>\n\u003Ccode>llmwiki eval --threshold 85 --judge --json\u003C\u002Fcode>\n\u003Ch2>5. 新鮮度、回滾與稽核歷史\u003C\u002Fh2>\n\u003Cp>知識庫最怕來源變動後沒人發現。這個工具會追蹤 stale 與 orphaned pages，支援 \u003Ca href=\"https:\u002F\u002Fgithub.com\u002Fatomicstrata\u002Fllm-wiki-compiler\">llmwiki refresh --stale\u003C\u002Fa> 這種定點修補，也會把每次 ingest、compile、query 寫進持久化操作紀錄。\u003C\u002Fp>\n\u003Cp>它還提供 rollback 與 diff 報告，適合需要回復錯誤 ingest，或向同事解釋某頁為何改動的場景。再加上 .llmwiki\u002Flast-lint.json 的快取，viewer 也能直接顯示最近的健康結果。\u003C\u002Fp>\n\u003Cul>\n  \u003Cli>stale claim 檢查與 freshness 報告\u003C\u002Fli>\n  \u003Cli>回滾與 compile diff 報告\u003C\u002Fli>\n  \u003Cli>timestamped log.md 稽核軌跡\u003C\u002Fli>\n\u003C\u002Ful>\n\u003Ch2>6. MCP 伺服器與 in-process SDK\u003C\u002Fh2>\n\u003Cp>\u003Ca href=\"https:\u002F\u002Fgithub.com\u002Fatomicstrata\u002Fllm-wiki-compiler\">llmwiki serve\u003C\u002Fa> 把整個流程暴露成 MCP，讓 Claude Desktop、Cursor、\u003Ca href=\"\u002Ftag\u002Fclaude-code\">Claude Code\u003C\u002Fa> 這類工具可以直接拿到有預算\u003Ca href=\"\u002Fnews\u002Fopen-source-ai-control-over-benchmarks-june-2026-zh\">控制\u003C\u002Fa>、帶引用的 context pack。這讓它不只是 CLI，而是可以進入代理人工作流的記憶層。\u003C\u002Fp>\n\u003Cp>如果你想把它嵌進自己的應用，createWiki({ root }) 也能在同一個程序裡跑 ingest、compile、query、status、freshness、export 和 eval。比起每一步都 shell out，這種方式更適合客製化工具。\u003C\u002Fp>\n\u003Ccode>createWiki({ root }).query(\"what changed?\")\u003C\u002Fcode>\n\u003Ch2>7. 多供應商支援與匯出路徑\u003C\u002Fh2>\n\u003Cp>這套工具能搭配多種模型後端，包括 \u003Ca href=\"\u002Ftag\u002Fanthropic\">Anthropic\u003C\u002Fa>、\u003Ca href=\"\u002Ftag\u002Fclaude\">Claude\u003C\u002Fa> Agent SDK、\u003Ca href=\"\u002Ftag\u002Fopenai\">OpenAI\u003C\u002Fa>-compatible servers、Ollama 與 \u003Ca href=\"\u002Ftag\u002Fgithub-copilot\">GitHub Copilot\u003C\u002Fa>。它也能匯出 typed JSON envelopes，並匯入 \u003Ca href=\"https:\u002F\u002Fgithub.com\u002Fatomicmemory\u002Fllmwiki\">atomicmemory\u002Fllmwiki\u003C\u002Fa>，作為原樣的 Atomic Memory records。\u003C\u002Fp>\n\u003Cp>這種可攜性對混用雲端 API 和本地模型的團隊很重要。你可以保留同一套工作流程，只換 provider 設定，就能在\u003Ca href=\"\u002Fnews\u002Fsolana-latest-updates-real-utility-pressure-zh\">真實\u003C\u002Fa>環境裡測試不同部署方式。\u003C\u002Fp>\n\u003Cul>\n  \u003Cli>Anthropic 與 Claude Agent SDK 支援\u003C\u002Fli>\n  \u003Cli>OpenAI-compatible 本地伺服器與 Ollama\u003C\u002Fli>\n  \u003Cli>JSON export 可接 runtime memory system\u003C\u002Fli>\n\u003C\u002Ful>\n\u003Ch2>哪種適合你\u003C\u002Fh2>\n\u003Cp>如果你要的是「可追溯、可更新、可交給代理人」的 source-to-wiki 流程，這個專案很對路。研究者、技術寫作者、維護者，或任何需要長期保留知識資產的人，都會比只看一次性摘要更有收穫。\u003C\u002Fp>\n\u003Cp>如果你的需求只是幾個檔案的快速搜尋，完整編譯器可能偏重；但只要你想要一個能持續刷新、能評測、能稽核的知識庫，這套設計就很有價值。\u003C\u002Fp>","7 項功能看 llm-wiki-compiler 如何把筆記、文件與論文編成可連結的 wiki，還保留引用、稽核與代理人存取。","github.com","https:\u002F\u002Fgithub.com\u002Fatomicstrata\u002Fllm-wiki-compiler",null,"https:\u002F\u002Fxxdpdyhzhpamafnrdkyq.supabase.co\u002Fstorage\u002Fv1\u002Fobject\u002Fpublic\u002Fcovers\u002Finline-1781389970790-qaws.png","industry","zh","f50df767-8d6e-4ceb-886f-f8cb610a9f19",[17,18,19,20,21,22,23],"llm-wiki-compiler","wiki compiler","knowledge base","citations","MCP","hybrid retrieval","audit trail",[25,26,27],"把原始資料編成帶引用的型別化 wiki，適合長期保存與查證。","混合檢索、健康評測與稽核歷史讓知識庫更適合團隊使用。","MCP 與 SDK 讓它能進入 Claude、Cursor 等代理人工作流。",0,"2026-06-13T22:32:22.507306+00:00","2026-06-13T22:32:22.498+00:00","fe20f6f6-432b-47bf-a410-a5f516d885ed",{"tags":33,"relatedLang":42,"relatedPosts":46},[34,35,37,38,40],{"name":20,"slug":20},{"name":21,"slug":36},"mcp",{"name":17,"slug":17},{"name":19,"slug":39},"knowledge-base",{"name":18,"slug":41},"wiki-compiler",{"id":15,"slug":43,"title":44,"language":45},"llm-wiki-compiler-raw-sources-to-wiki-en","llm-wiki-compiler turns raw sources into a wiki","en",[47,53,59,65,71,77],{"id":48,"slug":49,"title":50,"cover_image":51,"image_url":51,"created_at":52,"category":13},"1479dc4b-5684-4a0c-83f4-ee09595092bf","mastercard-opens-ai-payments-stablecoins-zh","Mastercard 讓 AI 直接付穩定幣","https:\u002F\u002Fxxdpdyhzhpamafnrdkyq.supabase.co\u002Fstorage\u002Fv1\u002Fobject\u002Fpublic\u002Fcovers\u002Finline-1781408867181-ad8z.png","2026-06-14T03:47:21.631777+00:00",{"id":54,"slug":55,"title":56,"cover_image":57,"image_url":57,"created_at":58,"category":13},"d88f40d6-a3d3-4bf5-b149-bf17bcaf009b","5-ji-jie-kan-chuan-zhe-zhi-jia-ren-xing-ji-qi-ren-ying-pian-zh","5 個細節看穿這支假人形機器人影片","https:\u002F\u002Fxxdpdyhzhpamafnrdkyq.supabase.co\u002Fstorage\u002Fv1\u002Fobject\u002Fpublic\u002Fcovers\u002Finline-1781406168128-28g0.png","2026-06-14T03:02:20.268133+00:00",{"id":60,"slug":61,"title":62,"cover_image":63,"image_url":63,"created_at":64,"category":13},"1e72405b-51cf-4651-9957-a87f067fd43b","jensen-huang-lg-ai-cooperation-five-bets-zh","黃仁勳與 LG 的 5 個 AI 合作重點","https:\u002F\u002Fxxdpdyhzhpamafnrdkyq.supabase.co\u002Fstorage\u002Fv1\u002Fobject\u002Fpublic\u002Fcovers\u002Finline-1781399872213-o4jc.png","2026-06-14T01:17:22.011305+00:00",{"id":66,"slug":67,"title":68,"cover_image":69,"image_url":69,"created_at":70,"category":13},"132ee9f1-b43f-4ed0-a6dc-10b512e95c04","nvidia-sk-group-expand-ai-ties-co-development-zh","Nvidia 與 SK Group 擴大 AI 合作","https:\u002F\u002Fxxdpdyhzhpamafnrdkyq.supabase.co\u002Fstorage\u002Fv1\u002Fobject\u002Fpublic\u002Fcovers\u002Finline-1781398973415-mq94.png","2026-06-14T01:02:29.836242+00:00",{"id":72,"slug":73,"title":74,"cover_image":75,"image_url":75,"created_at":76,"category":13},"3dd972ed-9161-445b-aa09-0e3b2b62cd76","python-jit-pause-pep-3-15-zh","Python JIT 暫停後，3.15 先看這 5 件事","https:\u002F\u002Fxxdpdyhzhpamafnrdkyq.supabase.co\u002Fstorage\u002Fv1\u002Fobject\u002Fpublic\u002Fcovers\u002Finline-1781389068504-ftgp.png","2026-06-13T22:17:20.132247+00:00",{"id":78,"slug":79,"title":80,"cover_image":81,"image_url":81,"created_at":82,"category":13},"02876c48-1813-437e-9dc2-5bd980955ec6","ukraines-ai-war-network-faster-combat-zh","烏克蘭 AI 戰網帶來的 5 個戰場變化","https:\u002F\u002Fxxdpdyhzhpamafnrdkyq.supabase.co\u002Fstorage\u002Fv1\u002Fobject\u002Fpublic\u002Fcovers\u002Finline-1781363879297-jm1p.png","2026-06-13T15:17:21.914474+00:00",[84,89,94,99,104,109,114,119,124,129],{"id":85,"slug":86,"title":87,"created_at":88},"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":90,"slug":91,"title":92,"created_at":93},"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":95,"slug":96,"title":97,"created_at":98},"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":100,"slug":101,"title":102,"created_at":103},"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":105,"slug":106,"title":107,"created_at":108},"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":110,"slug":111,"title":112,"created_at":113},"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":115,"slug":116,"title":117,"created_at":118},"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":120,"slug":121,"title":122,"created_at":123},"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":125,"slug":126,"title":127,"created_at":128},"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":130,"slug":131,"title":132,"created_at":133},"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"]