Markdown first
Exportable Obsidian pages, wikilinks, and frontmatter instead of another locked AI workspace.
Build a source-backed LLM Wiki.
Turn articles, papers, transcripts, or research memos into Obsidian-ready Markdown you can improve with AI.
Source manuscript
raw/001LLM Wiki turns raw material into a persistent Markdown knowledge base. The user keeps source notes, concept pages, links, and open questions.
Unsupported claims stay in Open Questions until the source gives enough evidence.
article.md
paper-notes.md
customer-call.md
Compiled wiki specimen
3 pages--- type: concept confidence: medium source_references: - article.md --- # Personal AI Wiki ## Source-Backed Notes - Keep raw sources in raw/ - Compile [[Concept Pages]] - Link [[NotebookLM]] and [[RAG]] ## Open Questions - Which claims need citations?
Source governed
Source notes, confidence, and open questions keep unsupported claims out of settled wiki pages.
Graph ready
Every ingest pass strengthens concepts, entities, indexes, and the next answer path.
The problem is not taking notes.
The hard part is turning messy source material into a maintained knowledge layer that still makes sense three months later.
AI chats disappear
A useful answer gets buried in conversation history, then the next session starts from zero.
Notes pile up
Highlights, PDFs, transcripts, and meeting notes keep growing, but they do not become reusable knowledge.
Search lacks structure
Keyword search can find a file. It cannot tell you which claims are supported, stale, duplicated, or unresolved.
How LLM Wiki works
A useful LLM Wiki is a repeatable compiler loop: source in, Markdown pages out, quality rules around the result.
01
Ingest
Paste one source and extract source-backed notes, concepts, entities, related pages, and open questions.
02
Link
Turn repeated ideas into [[wikilinks]] so every new source strengthens the existing wiki graph.
03
Lint
Check unsupported claims, weak page titles, missing source references, and duplicate concepts before trust drifts.
Next: turn Wikis into answer Lens.
A finished Wiki should not sit unused in a folder. Context Lens turns curated LLM Wikis into the background layer for direct, traceable answers.
Explore Context Lens01
Ask anything
The user writes the real question without choosing a topic first.
02
Route Lens
The system picks the smallest useful set of curated Wiki Context Packs.
03
Show evidence
The answer exposes the wiki pages, caveats, and next move instead of hiding its context.
One source becomes a small wiki update.
The first win is concrete: paste one source and get a source note, concept pages, links, and open questions instead of a disposable summary.
Generate a sample page1Input
One article, paper excerpt, course note, or market research memo.
2Source note
Title, source type, summary, source-backed notes, and confidence.
3Concept pages
3-5 reusable Markdown pages with frontmatter and wikilinks.
4Open questions
Claims that need more source support before they become settled notes.
5Index links
A small reading path that makes the next query or ingest pass easier.
Why not just use ChatGPT, NotebookLM, or RAG?
Use them when they fit. LLM Wiki 101 focuses on the portable Markdown layer those tools often leave to you.
ChatGPT
Best for: Fast exploration and drafting
Gap: The answer is transient unless you manually turn it into durable notes.
NotebookLM
Best for: Studying an uploaded source set
Gap: Great workspace, but the finished knowledge layer is not an Obsidian vault you fully own.
RAG
Best for: Question answering over a corpus
Gap: Retrieval serves each question; it does not automatically maintain concept pages and indexes.
Obsidian plugin
Best for: Automating inside an existing vault
Gap: A plugin helps after you know your schema, page types, prompts, and quality rules.
Built for people who keep source material.
The pattern works whenever the work benefits from citations, backlinks, reusable concepts, and repeatable maintenance.
Research reading
Book and course learning
Founder market research
Writing and content reuse
AI product notes
Personal operating manuals
The Starter Kit is the finished vault skeleton.
Start with folders, schemas, prompts, templates, examples, and lint checklists instead of inventing the workflow from a blank Obsidian vault.
See Starter Kit contentsvault/LLM Wiki Starter Kit
raw/ inbox/ wiki/ Index.md Concepts/ Entities/ Sources/ schema/ schema.md quality-rules.md prompts/ ingest-source.md query-wiki.md lint-wiki.md logs/ ingest-log.md lint-log.md
Start with one source, not a platform migration.
Paste a non-sensitive excerpt, generate the first page, then decide if you want the full vault structure.
FAQ
Is this a RAG product?+
Not in the first MVP. LLM Wiki 101 focuses on durable Markdown outputs: pages, schemas, prompts, and quality checks that you can keep in Obsidian.
Do I need Obsidian?+
No, but Obsidian is the best first target because Markdown files, frontmatter, and wikilinks are native there.
Is pasted text stored?+
The MVP generator runs with local deterministic rules in the browser/API boundary and does not persist your pasted source text.
Can I use Claude, ChatGPT, or Gemini?+
Yes. The workflow is model-agnostic. The Starter Kit is designed around prompts and schemas that work with any capable LLM.