LLM Wiki Compiler
Turn messy Notion notes into a clean, interconnected LLM Wiki. The agent reads each new note, distills reusable Concepts, auto cross-links them, and keeps your knowledge base tidy. Inspired by Andrej Karpathy's LLM Wiki idea. Markdown knowledge, no RAG or vector DB.- Crea páginas o bases de datos
- Monitorea cambios o activadores
- Reúne información de varias fuentes
Acerca de este agente
From notes to a living wiki
Most of us are great at capturing notes and terrible at processing them. This agent fixes that. Inspired by Andrej Karpathy's idea of an LLM-maintained personal wiki, it turns your raw Notion notes into a curated, interlinked knowledge base — entirely inside Notion, with no vector database or RAG pipeline.
How it works
The agent watches your Notes database. Whenever a note is created or marked unprocessed, it:
- Reads the full note and extracts reusable ideas
- Decides whether to create a new Concept or enrich an existing one (conservative by design — few Concepts, but good ones)
- Writes the distilled knowledge into the Concept's page body (definition, key points, practical use, distinctions, checklists)
- Cross-links related Concepts both as database relations and as a "See also" section of real page mentions — so you can navigate Zettelkasten-style and still filter and query
- Tags an Area when obvious, then marks the note as processed
Principles it follows
- Notes = raw material (never rewritten)
- Concepts = distilled knowledge (maintained over time)
- Areas/Resources = ontology (read-only)
- Skips purely logistical, personal, or empty notes instead of polluting the wiki
What you need
A Notion setup with Notes and Concepts databases (plus optional Areas/Resources). The agent runs automatically via triggers, so your wiki compiles itself in the background.
Why it's different
No embeddings, no external infrastructure — just clean, human-readable Markdown knowledge that both you and your AI can use.
Features
- Auto-triggers on new and updated notes
- Distills notes into reusable Concepts
- Automatic concept cross-linking (relations + body mentions)
- Conservative create-vs-update logic
- No RAG / no vector database
First-run setup (paste-in prompt)
Instead of shipping companion databases, include this prompt in the listing. After installing the agent, the user pastes it into their own Notion AI and the structure is built natively in their workspace.
Create the database structure for the LLM Wiki Compiler agent. Build these three databases and link the relations as described, keeping the names exactly as written.
1. Notes - your raw capture inbox.
- Name (title)
- Processed (checkbox)
- Summary (text)
- URL (url)
- Content type (select: Text, Video, Carousel, Lead magnet)
- Concepts (relation to Concepts)
- Area (relation to Areas/Resources)
2. Concepts - your distilled wiki entries.
- Name (title)
- Summary (text)
- Type (select: Methodology, Insight, Pattern, Tool, Framework, Principle, Person)
- Depth (select: Stub, Emerging, Mature)
- Last Compiled (date)
- Source Notes (relation to Notes - this is the reverse side of the Notes -> Concepts relation)
- Related Concepts (self-relation to Concepts)
- Area (relation to Areas/Resources)
3. Areas/Resources - read-only ontology the agent uses to tag notes.
- Name (title)
- Area/Resource (select: Area, Resource)
- Concepts (relation to Concepts)
- Notes (relation to Notes)
After the databases exist
The installer still needs to do two things in the agent settings (these can't be set by a prompt):
- Grant the agent edit access to Notes and Concepts, and view access to Areas/Resources.
- Point the agent's triggers at their new Notes database (run on note created / on Processed = false).