{"_id":"@edisonssszzz/llmwiki","_rev":"2-dbf62685b43b1628792a7af29e074dfd","name":"@edisonssszzz/llmwiki","dist-tags":{"latest":"0.1.0"},"versions":{"0.1.0":{"name":"@edisonssszzz/llmwiki","version":"0.1.0","keywords":["llm","wiki","knowledge-base","rag","karpathy","markdown","personal-knowledge","memex","second-brain","cli"],"author":{"name":"llmwiki contributors"},"license":"MIT","_id":"@edisonssszzz/llmwiki@0.1.0","maintainers":[{"name":"edisonssszzz","email":"shengzhao123123@gmail.com"}],"homepage":"https://github.com/Edisonzszs/llmwiki#readme","bugs":{"url":"https://github.com/Edisonzszs/llmwiki/issues"},"bin":{"llmwiki":"dist/cli.js"},"dist":{"shasum":"1ace143066c033de06c4fbd35db9ec0691d50605","tarball":"https://registry.npmjs.org/@edisonssszzz/llmwiki/-/llmwiki-0.1.0.tgz","fileCount":7,"integrity":"sha512-cjPbtdc7Wk83CWC0j9BJizg4sUixcGuZQB8hBZqzEhKzt2lTjg2NksdOSf5XKEO08jbx1i8DdQs/D017HoH18w==","signatures":[{"sig":"MEMCIDHfVIzW2ULDmL01JVTwPd+WcS+6h7fWt3pM7GeAs5OvAh8CSKixtV0T97ylP4fNi86UqcwInamk6jN4OiQB6sf/","keyid":"SHA256:DhQ8wR5APBvFHLF/+Tc+AYvPOdTpcIDqOhxsBHRwC7U"}],"unpackedSize":21599},"type":"module","engines":{"node":">=20"},"gitHead":"f306246b15a2623baeb6ad59165925f50fbdb67e","scripts":{"build":"tsc -b","typecheck":"tsc --noEmit","prepublishOnly":"tsc -b"},"_npmUser":{"name":"edisonssszzz","email":"shengzhao123123@gmail.com"},"repository":{"url":"git+https://github.com/Edisonzszs/llmwiki.git","type":"git","directory":"packages/cli"},"_npmVersion":"10.9.8","description":"A portable, headless LLM Wiki — compile knowledge once, maintain it continuously. CLI.","directories":{},"_nodeVersion":"22.22.3","dependencies":{"llmwiki-core":"^0.1.0"},"_hasShrinkwrap":false,"_npmOperationalInternal":{"tmp":"tmp/llmwiki_0.1.0_1783699639108_0.7072037464631071","host":"s3://npm-registry-packages-npm-production"},"deprecated":"renamed to llm-wiki-agent — use: npm i -g llm-wiki-agent"}},"time":{"created":"2026-07-10T16:07:18.946Z","modified":"2026-07-10T16:24:17.466Z","0.1.0":"2026-07-10T16:07:19.264Z"},"bugs":{"url":"https://github.com/Edisonzszs/llmwiki/issues"},"author":{"name":"llmwiki contributors"},"license":"MIT","homepage":"https://github.com/Edisonzszs/llmwiki#readme","keywords":["llm","wiki","knowledge-base","rag","karpathy","markdown","personal-knowledge","memex","second-brain","cli"],"repository":{"url":"git+https://github.com/Edisonzszs/llmwiki.git","type":"git","directory":"packages/cli"},"description":"A portable, headless LLM Wiki — compile knowledge once, maintain it continuously. CLI.","maintainers":[{"name":"edisonssszzz","email":"shengzhao123123@gmail.com"}],"readme":"# llmwiki\n\n> Compile knowledge once, maintain it continuously.\n\nA portable, headless **LLM Wiki** CLI. An LLM incrementally builds and maintains\na persistent, interlinked markdown knowledge base from your sources — so\nknowledge **compounds** instead of being re-derived on every query (RAG).\n\nThis is [Andrej Karpathy's LLM Wiki](https://gist.github.com/karpathy/442a6bf555914893e9891c11519de94f)\npattern, as a reusable engine. Bring your own key (any OpenAI-compatible endpoint\n— OpenAI, DeepSeek, Ollama, OpenRouter — or Anthropic).\n\n## Install\n\n```bash\nnpm install -g llmwiki\n```\n\n## Quickstart\n\n```bash\nllmwiki init                              # scaffold a knowledge base\nllmwiki ingest ./paper.md                 # two-step ingest → compiled, cross-linked pages\nllmwiki ask \"what are the main claims?\"   # retrieval + synthesis, cited [[like-this]]\nllmwiki lint --fix                        # deterministic hygiene (relocate/normalize/repair)\nllmwiki maintain                          # fill knowledge gaps, refresh stale pages\nllmwiki health                            # is the wiki compounding? (coverage/citations/freshness/…)\n```\n\nWith a model provider:\n\n```bash\nllmwiki --provider openai --base-url https://api.deepseek.com/v1 \\\n        --model deepseek-chat --api-key $KEY ingest ./paper.md\n```\n\nThe deterministic commands (`init`, `list`, `read`, `search`, `graph`, `lint`,\n`health`, `insights`, `impact`, `index rebuild`) work with **no model at all**.\nUse `--mock` to exercise the full ingest→ask loop with a scripted LLM (no key).\n\n## What's special\n\n- **Owned autonomy** — the engine plans + runs maintenance itself (no external\n  scheduler); `--auto` gates on accumulated signal.\n- **Quality evals** — `health` makes \"is the wiki compounding?\" measurable.\n- **Hybrid retrieval by default** — BM25 + optional vectors (RRF) + graph\n  expansion; CJK content is first-class.\n\n## Related packages\n\n- [`llmwiki-core`](https://www.npmjs.com/package/llmwiki-core) — the headless engine.\n- [`llmwiki-mcp`](https://www.npmjs.com/package/llmwiki-mcp) — MCP server (drive from Claude Code / Codex).\n- [`llmwiki-skill`](https://www.npmjs.com/package/llmwiki-skill) — portable Skill bundle.\n\nRepo + architecture: see the [project README](https://github.com/Edisonzszs/llmwiki).\n\nMIT.\n","readmeFilename":"README.md"}