Definition
What is an LLM Wiki?
An LLM Wiki is a personal knowledge base where the AI does more than retrieve document chunks at question time. It incrementally builds and maintains a durable wiki layer between you and your raw sources.
When a new source arrives, the AI can extract key claims, update entity and concept pages, add cross-references, flag contradictions, and keep summaries current. The knowledge compounds as a maintained artifact instead of disappearing into chat history.
Architecture
The three layers: sources, wiki, schema
The raw sources are the source of truth: articles, notes, transcripts, images, files, and data that should remain stable and traceable. The AI reads them, but should not rewrite them casually.
The wiki is the compiled layer: markdown pages for summaries, people, projects, concepts, comparisons, open questions, and synthesis. The schema is the operating manual that tells the AI how to maintain the wiki, including naming conventions, page formats, review rules, and workflows.
- Raw sources preserve provenance and let you audit where an idea came from.
- Wiki pages make knowledge navigable for both you and future agents.
- Schema files such as AGENTS.md or CLAUDE.md turn a generic chatbot into a disciplined wiki maintainer.
Operations
Ingest, query, lint, then file the useful answers back
An LLM Wiki has recurring operations. Ingest turns a new source into updates across the wiki. Query reads the existing wiki first, answers with citations, and can save valuable analysis as a new page. Lint checks the health of the knowledge base.
Index and log files make the system easier to operate. An index is a content map of pages and summaries. A log is the chronological record of ingests, questions, cleanups, and decisions, so the agent can understand how the wiki evolved.
- Ingest: process one or more sources and update every affected page.
- Query: synthesize from maintained pages instead of re-reading everything from scratch.
- Lint: find stale claims, missing links, orphan pages, contradictions, and gaps worth researching.
Walnut angle
How Walnut adapts the pattern for a local-first second brain
Walnut is not just a folder of notes and not just a RAG upload box. It is designed around a personal index that can be maintained over time: sources, entities, concepts, relationships, decisions, and reviewable AI suggestions.
The desktop app is where fragments become a callable map. Mobile capture can stay lightweight, while desktop review lets AI suggest structure and lets you keep the final judgment. Data sovereignty remains the basis for future agent sovereignty.
Comparison
RAG or notes app vs LLM Wiki
Retrieves raw chunks when you ask
Maintains a persistent wiki that compounds over time
Documents remain scattered or manually filed
Sources feed entity, concept, and synthesis pages
Subtle synthesis is re-derived repeatedly
Contradictions, links, and summaries are kept current
Chat answers disappear after the session
Useful answers can become new pages in the index
Memory is opaque or hosted by default
Local-first memory keeps ownership and review visible
Use cases
Practical ways to use Walnut
Deep research
Accumulate papers, articles, and reports into a thesis that improves with each source.
Personal operating system
Turn journals, goals, health notes, and reflections into a private map of patterns over time.
Book or course companion
Build pages for characters, concepts, chapters, themes, and unresolved questions while reading.
Product and coding memory
Keep decisions, architecture notes, issues, and tradeoffs connected to source context.
Team or business knowledge
Use transcripts, docs, and customer notes to maintain a living wiki that people can review.
FAQ
Questions people actually ask
Is an LLM Wiki the same as RAG?
No. RAG is a retrieval pattern that finds relevant chunks at query time. An LLM Wiki is a maintained knowledge layer that can be updated during ingest, query, and review, so the synthesis compounds.
Does the LLM write the whole wiki by itself?
The pattern lets the LLM do the maintenance work: summarizing, cross-linking, updating pages, and checking consistency. In Walnut, those AI changes should remain reviewable because the user keeps judgment and ownership.
Why does schema matter?
A schema document gives the agent conventions and workflows. Without it, the model behaves like a general chatbot. With it, the model can act like a consistent wiki maintainer across sessions.
Does Walnut require hosted AI?
No. Walnut is BYOK-friendly. Current paid plans cover software access and cloud features, not hosted AI compute unless a plan clearly says otherwise.