[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"project-1377":3},{"id":4,"name":5,"fullName":6,"owner":7,"repo":5,"description":8,"homepage":9,"htmlUrl":9,"language":10,"languages":9,"totalLinesOfCode":9,"stars":11,"forks":12,"watchers":13,"openIssues":14,"contributorsCount":15,"subscribersCount":15,"size":15,"stars1d":16,"stars7d":17,"stars30d":18,"stars90d":15,"forks30d":15,"starsTrendScore":19,"compositeScore":20,"rankGlobal":9,"rankLanguage":9,"license":9,"archived":21,"fork":21,"defaultBranch":22,"hasWiki":23,"hasPages":21,"topics":24,"createdAt":9,"pushedAt":9,"updatedAt":25,"readmeContent":26,"aiSummary":27,"trendingCount":15,"starSnapshotCount":15,"syncStatus":28,"lastSyncTime":29,"discoverSource":30},1377,"llm-wikid","shannhk\u002Fllm-wikid","shannhk","Karpathy-style LLM knowledge base for Obsidian. Clone, run Claude Code, start building your second brain.",null,"Shell",278,26,233,1,0,4,7,31,12,4.29,false,"main",true,[],"2026-06-12 02:00:27","\u003Cp align=\"center\">\n  \u003Cimg src=\"assets\u002Fcover.png\" alt=\"LLM Wikid\" width=\"700\">\n\u003C\u002Fp>\n\n# LLM Wikid\n\nAn AI-maintained knowledge base that lives in Obsidian. Based on [Karpathy's LLM Wiki pattern](https:\u002F\u002Fx.com\u002Fkarpathy\u002Fstatus\u002F1890540708772143562).\n\nYou dump raw sources into a folder. An AI agent reads them, compiles structured wiki pages with cross-references, runs bias checks, and maintains a master index. Every question you ask gets filed back in. The wiki compounds the more you use it.\n\nWorks with any agent that can read markdown and run shell commands: Claude Code, OpenClaw, Hermes, Codex, or your own setup.\n\n## Quick Start\n\n```bash\ngit clone https:\u002F\u002Fgithub.com\u002Fshannhk\u002Fllm-wikid.git my-wiki\ncd my-wiki\n\n# open as an Obsidian vault\n# File > Open Vault > select the my-wiki folder\n\n# start your agent\nclaude --dangerously-skip-permissions   # Claude Code\n# or open with OpenClaw, Hermes, Codex, etc.\n```\n\nThe agent reads `CLAUDE.md` and knows everything. That file is the schema that controls the entire system.\n\n## How It Works\n\nThis is **not RAG**. RAG re-derives answers every time by chunking documents and running vector search. This system compiles knowledge once into structured wiki pages, keeps them current, and pre-builds cross-references. At ~100 articles \u002F ~400K words, Karpathy found this outperforms RAG for Q&A.\n\n```\n                    ┌──────────────────────────────────┐\n                    │           CLAUDE.md              │\n                    │   the schema that controls       │\n                    │   everything the agent does      │\n                    └──────────┬───────────────────────┘\n                               │\n    ┌──────────┐      ┌────────▼────────┐      ┌──────────────┐\n    │  raw\u002F    │      │    INGEST       │      │   wiki\u002F      │\n    │          │─────>│                 │─────>│              │\n    │ clips    │      │ 0. sort clips   │      │ concepts\u002F    │\n    │ ideas    │      │ 1. resolve URLs │      │ entities\u002F    │\n    │ tweets   │      │ 2. extract media│      │ sources\u002F     │\n    │ articles │      │ 3. classify     │      │ index.md     │\n    │ papers   │      │ 4. compile pages│      │ log.md       │\n    │          │      │ 5. cross-link   │      │              │\n    └──────────┘      │ 6. bias check   │      └──────┬───────┘\n                      │ 7. update index │             │\n                      └─────────────────┘     ┌───────▼───────┐\n                                              │   Obsidian    │\n    ┌──────────┐      ┌─────────────────┐     │  graph view   │\n    │ question │─────>│    QUERY        │     │  + Dataview   │\n    │          │      │                 │     └───────────────┘\n    │          │      │ scan TLDRs      │\n    │          │\u003C─────│ read relevant   │\n    │  answer  │      │ synthesize      │\n    │  (filed  │      │ file back in    │───> wiki grows\n    │   back)  │      └─────────────────┘\n    └──────────┘\n```\n\nThe compound loop: every answer filed back means the next query has more to work with. Your curiosity makes the system smarter.\n\n## The Ingest Pipeline\n\nWhen you run `\u002Fwiki-ingest`, the agent processes everything in `raw\u002F` through these phases:\n\n**Phase 0 - Sort clippings.** Anything in `raw\u002Fclippings\u002F` gets moved to the right subfolder based on its URL. Tweet goes to `raw\u002Fbookmarks\u002F`, article to `raw\u002Farticles\u002F`, PDF to `raw\u002Fpapers\u002F`. The clippings folder is just an inbox.\n\n**Phase 1 - Resolve.** The agent detects what each file contains and fetches the full content:\n- YouTube URL: extracts transcript via `yt-dlp`\n- X\u002FTwitter URL: fetches tweet text, thread, metrics via X API\n- Web URL \u002F Reddit: scrapes full page via `scrapling`\n- PDF: reads directly\n- Plain text: already resolved\n\nThe resolved content replaces the file in-place. The original URL stays in frontmatter for provenance.\n\n**Phase 1.5 - Media extraction.** Images get downloaded to `raw\u002Fassets\u002Fimages\u002F` and analyzed. Videos get their transcripts pulled. A tweet that says \"here's my stack\" with an image of the stack is useless without the image.\n\n**Phase 2 - Classify and compile.** Each source gets classified by type (transcript, paper, report, article, tweet, notes) and extracted differently. A transcript gets speaker attribution and action items. A paper gets method and findings. An article gets core thesis and supporting arguments. Generic extraction misses what makes each format valuable.\n\nThe agent creates wiki pages, adds `[[wikilinks]]` between them, runs bias checks (counter-arguments and data gaps on every page), and updates the master index.\n\n**Phase 3 - Re-index.** If you have `qmd` installed, the search index gets updated with new content.\n\n## Commands\n\n| Command | What it does |\n|---------|-------------|\n| `\u002Fwiki-ingest` | Process new raw sources into wiki pages |\n| `\u002Fwiki-query [question]` | Research a question, get a cited answer, file it back |\n| `\u002Fwiki-explore [topic]` | Actively research and expand a topic using web search and `\u002Flast30days` |\n| `\u002Fwiki-lint` | Health check: contradictions, orphans, broken links, stale content |\n| `\u002Fsave [slug]` | File the current conversation as a wiki note (concept\u002Fsynthesis\u002Foutput\u002Fsop) |\n| `\u002Fautoresearch \u003Ctopic>` | Autonomous multi-round loop: search, fetch, extract, cross-reference, gap-fill, file. Budget configured in `wiki\u002Fmeta\u002Fprogram.md` |\n\n### Hot cache\n\n`wiki\u002Fhot.md` holds a rolling summary of recent sessions. It's auto-injected at SessionStart (via hook) and regenerated at Stop (via `scripts\u002Fupdate-hot-cache.sh` calling `claude -p` one-shot in the background). New sessions start with recent context. No recap needed.\n\nThe hooks are defined in `.claude\u002Fsettings.json` and use `$CLAUDE_PROJECT_DIR` so they work in any clone without edits.\n\n### Dashboard\n\n`wiki\u002Fmeta\u002Fdashboard.base` is an Obsidian Bases dashboard (requires Obsidian v1.9.10+). Four views: Recent, Low confidence, Unexplored, Stale (> 90 days). Open it in Obsidian to browse the vault visually.\n\n## Folder Structure\n\nAfter your first ingest, this is what it looks like:\n\n```\nraw\u002F                                          # your messy inbox\n  clippings\u002F                                  # Web Clipper landing zone (auto-sorted on ingest)\n  ideas\u002F\n    idea-brand-voice-skill.md                 # half-formed thought you typed out\n    idea-ai-marketing-repos.md                # project idea\n  bookmarks\u002F\n    karpathy-2026-llm-wiki.md                 # clipped tweet, resolved to full content\n    gregisenberg-ai-saas-framework.md         # another clip\n  articles\u002F\n    my-2026-three-moats-article.md            # your own published content\n  papers\u002F                                     # research PDFs\n  assets\u002Fimages\u002F\n    flowchart-agent-architecture.png          # downloaded from a tweet\n    ascii-marketing-levels.png                # screenshot you added\n  x-archive\u002F                                  # X\u002FTwitter account export (87 tweets, etc.)\n\nwiki\u002F                                         # compiled knowledge (AI-maintained)\n  index.md                                    # master catalog with TLDRs\n  log.md                                      # append-only changelog\n  dashboard.md                                # Dataview dashboard\n  concepts\u002F\n    ai-marketing.md                           # appeared in 8+ sources\n    distribution-moat.md                      # cross-cutting concept\n    context-engineering.md                    # stub (1 source, will grow)\n  entities\u002F\n    claude-superpowers.md                     # tool page\n    greg-isenberg.md                          # person page\n  sources\u002F\n    karpathy-2026-llm-wiki.md                 # summary of one raw source\n    your-2026-ai-agent-workflows.md          # themed group of your tweets\n  syntheses\u002F                                  # cross-cutting analysis (grows over time)\n  outputs\u002F\n    context-engineering-explained.md          # filed answer to a \u002Fwiki-query\n  sops\u002F\n    visual-content-style.md                   # your documented process\n\ntemplates\u002F                                    # starter templates for each page type\n.claude\u002Fcommands\u002F                             # slash commands\n```\n\n## Quality Controls\n\nThe system has safeguards to prevent it from becoming a pile of confident-sounding AI output:\n\n**Bias checks.** Every concept and source page includes a `## Counter-arguments` section (what pushes back) and a `## Data gaps` section (what we don't know). Without these, the wiki just agrees with every source.\n\n**Validation gate.** The agent sets `explored: false` on every page it creates. Only you can mark something as reviewed by setting it to `true`. You always know what's been human-verified and what hasn't.\n\n**Confidence levels.** Pages are tagged `high`, `medium`, `low`, or `uncertain` based on how well-supported the content is. Multiple corroborating sources = high. Single mention with thin evidence = low. The agent has to be honest.\n\n**Source tracing.** Every claim links back to a specific source page. No vague \"research shows\" without pointing to which research.\n\n**No blind links.** Every `[[wikilink]]` resolves to an actual page. If a concept is mentioned but doesn't have a page yet, the agent creates a stub.\n\n## Curation\n\nThis is an idea bank, not a fact database. Some of what goes in will be wrong, half-formed, or just bad. That's fine. The point is capturing your thinking so it compounds over time.\n\nBut curation matters. Before clipping something into `raw\u002F`, ask: does this feel 80%+ relevant to what I'm working on or thinking about? If it's noise, skip it. The wiki gets stronger when the inputs are high-signal.\n\nRunning `\u002Fwiki-lint` regularly catches the drift: contradictions between pages, stale claims, orphan pages nobody links to, concepts mentioned but never defined. The agent fixes what it can and flags what needs your judgment.\n\n## Git Sync\n\nGitHub is the source of truth. Every operation that changes `wiki\u002F` or `raw\u002F` ends with a commit and push. This means:\n\n- Every change is reversible (`git revert`)\n- Multiple agents can work on the same wiki (pull before writing, push after)\n- You can access the wiki from anywhere that can clone a repo\n- Your entire knowledge history is preserved\n\n```bash\n# every agent session, every ingest, every query:\ngit add . && git commit -m \"wiki update\" && git push\n```\n\n### Claude Dispatch (automated ingest)\n\nSet up a scheduled trigger that runs `\u002Fwiki-ingest` every morning. You clip things during the day, they get processed overnight.\n\n```bash\n# in Claude Code:\n\u002Fschedule\n```\n\nOr configure directly: a cron trigger pointing at your repo, running the ingest prompt daily. The remote agent clones the repo, reads CLAUDE.md, processes new sources, commits, and pushes. You wake up to a richer wiki.\n\nThe Dispatch agent runs remotely, so it won't have local tools like X API keys. It can still process web URLs, plain text, and anything already resolved. Tweet resolution happens when you run ingest locally.\n\n## Search (qmd)\n\nFor fast search across your wiki, install [qmd](https:\u002F\u002Fgithub.com\u002Ftobi\u002Fqmd) by Tobi Lutke. Hybrid BM25\u002Fvector search with LLM re-ranking, 100% on-device.\n\n```bash\nnpm install -g @tobilu\u002Fqmd\nqmd collection add wiki --name my-wiki\nqmd embed\n```\n\nThree search modes:\n```bash\nqmd search \"query\"        # BM25 keyword search (fast)\nqmd vsearch \"query\"       # vector semantic search\nqmd query \"query\"         # hybrid with LLM re-ranking (best)\n```\n\nqmd has a built-in MCP server so your agent can use it as a native tool:\n```json\n{\n  \"mcpServers\": {\n    \"qmd\": { \"command\": \"qmd\", \"args\": [\"mcp\"] }\n  }\n}\n```\n\nThe `\u002Fwiki-explore` command uses qmd automatically when installed.\n\n## Recommended Skills\n\n| Skill | What it adds |\n|-------|-------------|\n| [last30days](https:\u002F\u002Fgithub.com\u002Fmvanhorn\u002Flast30days-skill) | Searches Reddit, HN, X, YouTube, GitHub, Polymarket for recent community signals. Used by `\u002Fwiki-explore`. |\n| [Obsidian Web Clipper](https:\u002F\u002Fobsidian.md\u002Fclipper) | Clip anything from your browser directly to `raw\u002Fclippings\u002F` |\n\n## Obsidian Plugins\n\n| Plugin | Purpose |\n|--------|---------|\n| Dataview | Query frontmatter, build dashboards from wiki data |\n| Obsidian Git | Auto-commit on interval, push to remote |\n| Templater | Auto-populate dates and fields on new notes |\n| Tag Wrangler | Bulk rename and merge tags |\n\n## Scaling\n\n| Wiki Size | Strategy |\n|-----------|----------|\n| 0-300 pages | File-based, index TLDR scanning + qmd search |\n| 300-500 pages | qmd becomes primary search layer |\n| 500+ pages | Consider PostgreSQL\u002FSupabase |\n\n## Optional Tools\n\nInstalled as needed during ingestion:\n\n| Tool | What it does | Install |\n|------|-------------|---------|\n| `yt-dlp` | YouTube transcripts | `brew install yt-dlp` |\n| `scrapling` | Web scraping | `pipx install scrapling` |\n| `summarize` | PDF extraction | `brew install steipete\u002Ftap\u002Fsummarize` |\n| X API | Tweet\u002Fthread fetching | OAuth keys in `.env` (optional) |\n\n## Credits\n\n- [Andrej Karpathy's LLM Wiki pattern](https:\u002F\u002Fx.com\u002Fkarpathy\u002Fstatus\u002F1890540708772143562) - the original idea\n- [hooeem's LLM Knowledge Base course](https:\u002F\u002Fwww.youtube.com\u002Fwatch?v=IVpOyKCNZYw) - practical walkthrough\n- [qmd](https:\u002F\u002Fgithub.com\u002Ftobi\u002Fqmd) by Tobi Lutke - local markdown search engine\n- [last30days](https:\u002F\u002Fgithub.com\u002Fmvanhorn\u002Flast30days-skill) by mvanhorn - multi-platform signal search\n- Built by [@shannholmberg](https:\u002F\u002Fx.com\u002Fshannholmberg)\n","LLM Wikid 是一个基于 Obsidian 的 AI 维护的知识库。其核心功能是通过将原始资料放入指定文件夹，AI 代理会读取这些资料并编译成结构化的维基页面，同时进行交叉引用、偏见检查和维护主索引。项目支持任何能够读取 Markdown 并运行 Shell 命令的 AI 代理，如 Claude Code、OpenClaw 或自定义设置。适用于需要构建个人知识管理系统或第二大脑的场景，尤其适合处理大量文本数据，并希望以结构化方式存储和检索信息的用户。随着使用频率增加，该系统在问答方面的表现优于传统的检索增强生成（RAG）方法。",2,"2026-06-11 02:43:22","CREATED_QUERY"]