[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"article-llm-wiki-karpathy-knowledge-base-app-en":3,"tags-llm-wiki-karpathy-knowledge-base-app-en":27,"related-lang-llm-wiki-karpathy-knowledge-base-app-en":28,"related-posts-llm-wiki-karpathy-knowledge-base-app-en":32,"series-tools-b7414266-729b-4471-8a0f-6e1f7ea8f04b":69},{"id":4,"title":5,"content":6,"summary":7,"source":8,"source_url":9,"author":9,"image_url":10,"keywords":11,"language":17,"translated_content":9,"views":18,"is_premium":19,"created_at":20,"updated_at":20,"cover_image":10,"published_at":20,"rewrite_status":21,"rewrite_error":9,"rewritten_from_id":9,"slug":22,"category":23,"related_article_id":24,"status":25,"google_indexed_at":26,"x_posted_at":9,"tweet_text":9,"title_rewritten_at":9,"title_original":9,"key_takeaways":9,"topic_cluster_id":9,"embedding":9,"is_canonical_seed":19},"b7414266-729b-4471-8a0f-6e1f7ea8f04b","From Karpathy Gist to 4,400-Star App: LLM Wiki in Three Weeks","\u003Cp>In early April, Andrej Karpathy published a deceptively simple idea on GitHub Gist: let an AI incrementally compile a collection of documents into a self-maintaining knowledge base. Rather than traditional RAG—which reconstructs context anew on every query—this architecture treats knowledge as a compiled artifact that evolves continuously. The Gist gathered 5,000+ stars in days.\u003C\u002Fp>\n\n\u003Cp>Three weeks later, the real story began. Developer Nash Su transformed that conceptual sketch into \u003Ca href=\"https:\u002F\u002Fgithub.com\u002Fnashsu\u002Fllm_wiki\" target=\"_blank\" rel=\"noopener\">LLM Wiki\u003C\u002Fa>, a complete cross-platform desktop application now at 4,400 stars. v0.3.13 landed at month-end with installers for macOS, Linux, Windows, and a Chrome clipper extension. This is what agent-era iteration looks like: ideas circulate as architecture blueprints; competent builders materialize them into shipping products in weeks.\u003C\u002Fp>\n\n\u003Ch2>Three-Layer Architecture in Practice\u003C\u002Fh2>\n\n\u003Cp>LLM Wiki implements Karpathy's three-layer structure faithfully. Immutable raw documents live in a raw\u002F directory. LLM-generated knowledge pages—summaries, entity references, concept explorations—populate the wiki\u002F directory. A schema file (modeled after CLAUDE.md) defines structure, conventions, and workflows. Three core operations keep the system coherent: ingest (absorb new sources, extract entities, link to existing pages), query (synthesize answers with citations, optionally file discoveries back into the wiki), and lint (detect contradictions, flag orphaned pages, maintain health).\u003C\u002Fp>\n\u003Cfigure class=\"my-6\">\u003Cimg src=\"https:\u002F\u002Fxxdpdyhzhpamafnrdkyq.supabase.co\u002Fstorage\u002Fv1\u002Fobject\u002Fpublic\u002Fcovers\u002Finline-1777388008679-5nr6.png\" alt=\"From Karpathy Gist to 4,400-Star App: LLM Wiki in Three Weeks\" class=\"rounded-xl w-full\" loading=\"lazy\" \u002F>\u003C\u002Ffigure>\n\n\n\u003Cp>The implementation deepens the concept. A two-stage ingestion pipeline first analyzes documents—LLM extracts entities, concepts, contradictions, and locates connections to existing wiki pages—then generates (creating summary pages, entity pages, concept pages, updating indices, flagging items for human review). A single source can cascade-update 10–15 wiki pages. SHA256 incremental caching skips unchanged files; a persistent queue survives crashes and retries failures up to three times.\u003C\u002Fp>\n\n\u003Ch2>Knowledge Graphs That Reveal Insight\u003C\u002Fh2>\n\n\u003Cp>The standout innovation is the knowledge graph design. Instead of treating all links equally, the system weights edges across four dimensions: direct wikilinks (×3.0), source overlap (×4.0—two pages citing the same document), Adamic-Adar shared neighbors (×1.5—rarer shared connections count higher), and type affinity (×1.0). Sigma.js and ForceAtlas2 render the graph visually. Louvain community detection auto-discovers thematic clusters.\u003C\u002Fp>\n\n\u003Cp>The graph then surfaces two kinds of insight. Surprise links—unexpected edges crossing clusters or crossing types—are flagged as \"starting points for cognitive breakthroughs.\" Knowledge gaps—isolated pages, sparsely-connected communities, bridge nodes linking 3+ clusters—each get a one-click \"deep research\" button that triggers Tavily-powered web search, synthesizes results into a research page, and re-enters it into the ingest pipeline. A confirmation modal prevents accidental triggers.\u003C\u002Fp>\n\n\u003Cp>Retrieval spans multiple stages: tokenization (CJK bigrams for Chinese), optional vector ANN via LanceDB, two-hop graph traversal from search seeds. Context windows are tunable (4K–1M tokens), allocated as 60% wiki, 20% chat history, 5% index, 15% system prompt. Reported recall improves from 58.2% to 71.4% when vector search is enabled.\u003C\u002Fp>\n\n\u003Cp>A subtle detail: each project includes a purpose.md intent file. The LLM reads it on every ingest and query, ensuring the wiki develops direction rather than aimless accumulation. Deletion is intelligent too—removing a source auto-cleans its summary page, strips source citations from shared pages (without deleting the shared entities themselves), purges index entries, and fixes broken links.\u003C\u002Fp>\n\n\u003Ch2>Knowledge as Compilation, Not Retrieval\u003C\u002Fh2>\n\n\u003Cp>The real lesson isn't any single feature—it's the iteration speed and philosophical shift. Karpathy shared an architecture sketch on Gist, a new form of open-source knowledge circulation. A competent developer materialized it into a shipping product in three weeks. RAG hasn't failed; it's been demoted from default to fallback. The frontier is treating knowledge as a compiled artifact, not a query target. mem0 and MemOS solve this for agents; LLM Wiki solves it for human users managing their own domains.\u003C\u002Fp>\n\u003Cfigure class=\"my-6\">\u003Cimg src=\"https:\u002F\u002Fxxdpdyhzhpamafnrdkyq.supabase.co\u002Fstorage\u002Fv1\u002Fobject\u002Fpublic\u002Fcovers\u002Finline-1777388011705-d5ne.png\" alt=\"From Karpathy Gist to 4,400-Star App: LLM Wiki in Three Weeks\" class=\"rounded-xl w-full\" loading=\"lazy\" \u002F>\u003C\u002Ffigure>\n\n\n\u003Cp>The tool is fully Obsidian-compatible: the generated wiki directory is a standard Obsidian Vault with .obsidian\u002F configuration. LLM Wiki edits; Obsidian views. Neither tool tries to do everything. This is the agent-era shift: from monolithic applications to interoperable, purpose-built tools that compose elegantly. Knowledge isn't locked into any single system. It's portable, hackable, and evolves under both AI and human direction.\u003C\u002Fp>","Andrej Karpathy's LLM Wiki concept—shared as a Gist in April—inspired a fully-featured cross-platform desktop app with 4,400 GitHub stars within three weeks. The speed exemplifies the agent-era open-source cycle: architecture sketches become production tools overnight. Knowledge management shifts from query-driven retrieval to compiled artifacts, reshaping how both humans and agents organize understanding.","oracore-original",null,"https:\u002F\u002Fxxdpdyhzhpamafnrdkyq.supabase.co\u002Fstorage\u002Fv1\u002Fobject\u002Fpublic\u002Fcovers\u002Finline-1777388008679-5nr6.png",[12,13,14,15,16],"LLM Wiki","Karpathy","Knowledge Graph","Obsidian","RAG Alternative","en",0,false,"2026-04-28T14:53:00.060063+00:00","done","llm-wiki-karpathy-knowledge-base-app-en","tools","c204da96-eaaf-4e34-b1c4-7dc1a28d6085","published","2026-04-29T09:00:09.722+00:00",[],{"id":24,"slug":29,"title":30,"language":31},"llm-wiki-karpathy-knowledge-base-app-zh","Karpathy 的 LLM Wiki 三週變桌面應用","zh",[33,39,45,51,57,63],{"id":34,"slug":35,"title":36,"cover_image":37,"image_url":37,"created_at":38,"category":23},"a6c1d84d-0d9c-4a5a-9ca0-960fbfc1412e","why-gemini-api-pricing-is-cheaper-than-it-looks-en","Why Gemini API pricing is cheaper than it looks","https:\u002F\u002Fxxdpdyhzhpamafnrdkyq.supabase.co\u002Fstorage\u002Fv1\u002Fobject\u002Fpublic\u002Fcovers\u002Finline-1778869846824-s2r1.png","2026-05-15T18:30:26.595941+00:00",{"id":40,"slug":41,"title":42,"cover_image":43,"image_url":43,"created_at":44,"category":23},"8b02abfa-eb16-4853-8b15-63d302c7b587","why-vidhub-huiyuan-hutong-bushi-quan-shebei-tongyong-en","Why VidHub 会员互通不是“买一次全设备通用”","https:\u002F\u002Fxxdpdyhzhpamafnrdkyq.supabase.co\u002Fstorage\u002Fv1\u002Fobject\u002Fpublic\u002Fcovers\u002Finline-1778789439875-uceq.png","2026-05-14T20:10:26.046635+00:00",{"id":46,"slug":47,"title":48,"cover_image":49,"image_url":49,"created_at":50,"category":23},"abe54a57-7461-4659-b2a0-99918dfd2a33","why-buns-zig-to-rust-experiment-is-right-en","Why Bun’s Zig-to-Rust experiment is the right move","https:\u002F\u002Fxxdpdyhzhpamafnrdkyq.supabase.co\u002Fstorage\u002Fv1\u002Fobject\u002Fpublic\u002Fcovers\u002Finline-1778767895201-5745.png","2026-05-14T14:10:29.298057+00:00",{"id":52,"slug":53,"title":54,"cover_image":55,"image_url":55,"created_at":56,"category":23},"f0015918-251b-43d7-95af-032d2139f3f6","why-openai-api-pricing-is-product-strategy-en","Why OpenAI API pricing is a product strategy, not a footnote","https:\u002F\u002Fxxdpdyhzhpamafnrdkyq.supabase.co\u002Fstorage\u002Fv1\u002Fobject\u002Fpublic\u002Fcovers\u002Finline-1778749841805-uyhg.png","2026-05-14T09:10:27.921211+00:00",{"id":58,"slug":59,"title":60,"cover_image":61,"image_url":61,"created_at":62,"category":23},"7096dab0-6d27-42d9-b951-7545a5dddf33","why-claude-code-prompt-design-beats-ide-copilots-en","Why Claude Code’s prompt design beats IDE copilots","https:\u002F\u002Fxxdpdyhzhpamafnrdkyq.supabase.co\u002Fstorage\u002Fv1\u002Fobject\u002Fpublic\u002Fcovers\u002Finline-1778742651754-3kxk.png","2026-05-14T07:10:30.953808+00:00",{"id":64,"slug":65,"title":66,"cover_image":67,"image_url":67,"created_at":68,"category":23},"1f1bff1e-0ebc-4fa7-a078-64dc4b552548","why-databricks-model-serving-is-right-default-en","Why Databricks Model Serving is the right default for production infe…","https:\u002F\u002Fxxdpdyhzhpamafnrdkyq.supabase.co\u002Fstorage\u002Fv1\u002Fobject\u002Fpublic\u002Fcovers\u002Finline-1778692290314-gopj.png","2026-05-13T17:10:32.167576+00:00",[70,75,80,85,90,95,100,105,110,115],{"id":71,"slug":72,"title":73,"created_at":74},"8008f1a9-7a00-4bad-88c9-3eedc9c6b4b1","surepath-ai-mcp-policy-controls-en","SurePath AI's New MCP Policy Controls Enhance AI Security","2026-03-26T01:26:52.222015+00:00",{"id":76,"slug":77,"title":78,"created_at":79},"27e39a8f-b65d-4f7b-a875-859e2b210156","mcp-standard-ai-tools-2026-en","MCP Standard in 2026: Integrating AI Tools","2026-03-26T01:27:43.127519+00:00",{"id":81,"slug":82,"title":83,"created_at":84},"165f9a19-c92d-46ba-b3f0-7125f662921d","rag-2026-transforming-enterprise-ai-en","How RAG in 2026 is Transforming Enterprise AI","2026-03-26T01:28:11.485236+00:00",{"id":86,"slug":87,"title":88,"created_at":89},"6a2a8e6e-b956-49d8-be12-cc47bdc132b2","mastering-ai-prompts-2026-guide-en","Mastering AI Prompts: A 2026 Guide for Developers","2026-03-26T01:29:07.835148+00:00",{"id":91,"slug":92,"title":93,"created_at":94},"d6653030-ee6d-4043-898d-d2de0388545b","evolving-world-prompt-engineering-en","The Evolving World of Prompt Engineering","2026-03-26T01:29:42.061205+00:00",{"id":96,"slug":97,"title":98,"created_at":99},"3ab2c67e-4664-4c67-a013-687a2f605814","garry-tan-open-sources-claude-code-toolkit-en","Garry Tan Open-Sources a Claude Code Toolkit","2026-03-26T08:26:20.245934+00:00",{"id":101,"slug":102,"title":103,"created_at":104},"66a7cbf8-7e76-41d4-9bbf-eaca9761bf69","github-ai-projects-to-watch-in-2026-en","20 GitHub AI Projects to Watch in 2026","2026-03-26T08:28:09.752027+00:00",{"id":106,"slug":107,"title":108,"created_at":109},"231306b3-1594-45b2-af81-bb80e41182f2","claude-code-vs-cursor-2026-en","Claude Code vs Cursor in 2026","2026-03-26T13:27:14.177468+00:00",{"id":111,"slug":112,"title":113,"created_at":114},"9f332fda-eace-448a-a292-2283951eee71","practical-github-guide-learning-ml-2026-en","A Practical GitHub Guide to Learning ML in 2026","2026-03-27T01:16:50.125678+00:00",{"id":116,"slug":117,"title":118,"created_at":119},"1b1f637d-0f4d-42bd-974b-07b53829144d","aiml-2026-student-ai-ml-lab-repo-review-en","AIML-2026 Is a Bare-Bones Student Lab Repo","2026-03-27T01:21:51.661231+00:00"]