What an LLM Wiki is, why it's different from Obsidian or Notion, and how to build one so your AI assistant always has the context it needs to give you useful answers.
Most people who use AI heavily notice the same failure pattern: they get a great answer in one session, close the tab, and have to re-explain the same context next time. The AI is stateless between sessions. Every conversation starts cold. The output quality is directly proportional to how well you brief it, and briefing from scratch every time is exhausting.
The fix that's emerging in technically curious circles is the personal LLM Wiki: a structured, AI-readable knowledge base about you, your work, and your domain that an AI can load as context at the start of a session. Not a notes app. Not a second brain. Something specifically designed to make an LLM immediately useful without a briefing.
This post explains what an LLM Wiki is, why it's different from the knowledge management tools you already use, and how to build one, including how to populate it from the knowledge you've already accumulated in notes, conversations, and documents.
What Is an LLM Wiki?
A traditional wiki is a collection of hyperlinked pages designed for humans to navigate. You browse a hierarchy, follow links, search for terms. The structure serves the human reader's way of moving through information.
An LLM Wiki is organized around a different reader: the AI. It's a structured document (or set of documents) written to give an LLM the context it needs to reason about your situation, answer questions about your domain, and make decisions consistent with your values and constraints. The human rarely reads it directly. You maintain it and the AI uses it.
Traditional notes / wiki:
- Organized for human navigation (hierarchy, links)
- Written in personal shorthand or stream-of-consciousness
- Comprehensive: captures everything
- Retrieval: you search for it
- Value: a record of what you thought
LLM Wiki:
- Organized for AI consumption (structured sections)
- Written explicitly, no implied context
- Curated: contains what the AI needs to know
- Retrieval: AI reads it automatically
- Value: makes every AI session immediately useful
The practical implication: your Obsidian vault or Notion workspace is probably not already an LLM Wiki, even if it's comprehensive and well-organized. Notes written for yourself are full of implied context that an AI can't infer: shorthand, references to things you understand but never wrote down, emotional subtext, and gaps that would be obvious to you but aren't on the page.
An LLM Wiki is explicit by design. It assumes the reader has never met you.
How It Differs From a Second Brain
The "second brain" concept (popularized by Tiago Forte and implemented in tools like Obsidian, Notion, and Roam) is about externalizing your thinking so you can retrieve and recombine it later. The goal is your future self: building a system your future self can search through and find useful.
An LLM Wiki has a different goal: making the AI immediately useful in the current session. It's not about archiving your thinking. It's about curating the context an AI needs to reason well on your behalf.
Second brain: comprehensive capture. You want to capture everything: ideas, quotes, meeting notes, random connections. More is better. The system is a long-term archive you query over years.
LLM Wiki: curated context. You want the AI to know the things that make the most difference to its output quality: your background, your current priorities, your domain knowledge, your constraints and preferences. Less is more. A focused, well-maintained wiki outperforms an enormous unstructured dump of notes.
The best setup for heavy AI users: a second brain (notes system) for comprehensive capture, and an LLM Wiki for curated AI context. They serve different purposes and work best together.
What Goes In an LLM Wiki
The contents depend on how you use AI. For most people, the high-value sections cluster into five categories:
1. Who you are. Background, current role, domain expertise, career arc. Not a CV. The specific things that change how an AI should frame its answers to you. A cardiologist and a graphic designer both asking "explain oxidative stress" want different depth and vocabulary. The wiki tells the AI which one you are.
2. Current context and priorities. What you're working on right now, what matters this quarter, open decisions, active projects. This is the section that needs the most frequent updating (monthly or when major context changes) and the section with the highest immediate impact on output quality.
3. Domain knowledge and opinions. Your considered views on topics in your field. Not Wikipedia-level facts, but your specific perspective. "I think the consensus on X is wrong because Y." "My framework for evaluating Z is..." This makes AI responses align with your actual thinking rather than the generic consensus.
4. Preferences and constraints. How you like to communicate, what format you want responses in, what you consider a good answer vs. a great one. Practical constraints: tools you use, things you can't change, resources you have or don't have. Eliminates suggestions that are technically correct but don't apply to your situation.
5. Reference materials for specific tasks. Definitions that matter in your field, templates you use repeatedly, standards you work to, terminology that has specific meaning in your context. The things you'd have to explain if a new colleague started tomorrow.
Building It: The Guided Interview Approach
Rather than staring at a blank document and wondering what to include, the best approach is a guided interview that extracts the right information in the right structure:
"Help me build my personal LLM Wiki from scratch. Interview me section by section: background and expertise, current work and priorities, domain knowledge and opinions, preferences and constraints, and key reference material. Ask follow-up questions to make my answers specific. Push back on vague answers. After each section, show me the wiki text you're building so I can confirm it's accurate."
The "push back on vague answers" instruction matters. The most common mistake in building an LLM Wiki is writing in the same shorthand you use in personal notes. "I work on AI stuff" is useless context. "I'm a machine learning engineer at a Series B startup building recommendation systems for e-commerce, with five years of production ML experience" is the context that changes output quality.
Budget 45 to 60 minutes for the initial build. The output is a Markdown file (your LLM Wiki) that you put in every project folder alongside your skill files.
What using it looks like afterward:
Before wiki: "I'm a [role] at a [company] and I'm trying to [explain context for 5 minutes]. What's your recommendation?"
After wiki: "What's your recommendation?" Claude has already read the wiki. It knows who you are, what you're working on, and your constraints. The answer is immediately calibrated.
Populating From Knowledge You Already Have
Most people with years of experience have the raw material for a great LLM Wiki scattered across notes apps, documents, and past conversations. The gap is synthesis: extracting the structured, explicit knowledge from the unstructured accumulation.
From your notes. If you have an Obsidian vault, Notion workspace, or folder of markdown notes, AI can read through them and extract the content most relevant to your wiki: your recurring opinions on domain topics, the frameworks you use repeatedly, the decisions and their reasoning, the terminology that has specific meaning in your context.
"Read through my notes from the last 12 months. Extract: my recurring opinions and frameworks on [domain], terminology I use with specific meaning, patterns in how I think about [topic], and decisions I made with reasoning that might inform future decisions. Format the output as sections I can paste into my LLM Wiki."
From your AI conversations. Heavy AI users have a goldmine of knowledge buried in their conversation history: problems worked through, decisions made, frameworks developed through back-and-forth. AI can extract structured intelligence from conversation exports: the decisions and their reasoning, recurring themes across sessions, open questions, and the patterns in how you think and work.
"Here are my exported AI conversations from the last 6 months. Extract the content that belongs in my LLM Wiki: what do I keep coming back to, what frameworks do I seem to use repeatedly, what opinions have I developed and stated explicitly, what constraints and preferences come up across sessions? Format as wiki sections."
Run both approaches once when you first build your wiki. They bootstrap a knowledge base from material you've already produced without requiring you to reconstruct it from memory.
What a Well-Built LLM Wiki Looks Like
Here's the structure, annotated so you understand why each section is there:
# Personal LLM Wiki — [Your Name]
Last updated: [Date]
## Identity & Background
[Role, domain, years of experience, relevant credentials.
Written explicitly: assumes the reader knows nothing about you.]
## Current Context
### Active Projects
[What you're working on right now with enough detail to matter]
### Open Decisions
[Decisions you're actively wrestling with]
### This Quarter's Priorities
[What success looks like in the next 90 days]
## Domain Knowledge & Opinions
### [Topic Area 1]
[Your actual views, not the consensus. What you think and why.]
### [Topic Area 2]
[Frameworks you use, approaches you favor, things you've found don't work]
## Preferences & Constraints
### Communication Style
[How you want responses: length, format, tone, what to avoid]
### Tools & Environment
[What you use, what you don't, what you can't change]
### Decision-Making Style
[How you like to evaluate options, what you over/underweight]
## Reference Material
### Terminology
[Terms that mean specific things in your context]
### Templates & Formats
[Structures you use repeatedly]
### Standards & Criteria
[What a good X looks like in your world]
The wiki lives as a Markdown file. It goes in the same folder as your project or task, alongside your CLAUDE.md skill file. Claude reads both: the wiki provides who you are, the skill file provides what Claude should do. Together they produce output calibrated to both the task and the person.
Maintaining Your LLM Wiki
The most common failure mode: building a great wiki and never updating it. Within three months it's stale. Your priorities have shifted, you've developed new views, the projects in your current context section are done. Stale context is worse than no context in some ways, because it produces confidently wrong outputs.
Monthly. Update the Current Context section. Replace completed projects with new ones, update open decisions, refresh this quarter's priorities. Takes 10 to 15 minutes.
Quarterly. Review the Domain Knowledge section. Have any of your views evolved? Are there new frameworks you've developed? Add what's new, remove what's no longer accurate.
Ad hoc. When you have an insight, develop a new framework, or have a conversation that produces something worth keeping, add it to the relevant section.
Annually. Full review. Rewrite sections that have grown stale, consolidate overlapping entries, remove things that no longer apply.
What Changes When You Have One
The difference isn't subtle. With a well-maintained LLM Wiki, questions that previously required a paragraph of setup become one-liners. Recommendations stop being generic and start being specific to your situation. Suggestions that don't apply to you (wrong tool, wrong scale, wrong context) stop appearing because the wiki has already communicated your actual constraints.
The deeper shift: AI stops being a tool you use for isolated tasks and starts functioning more like a well-briefed collaborator. One that knows your background, understands your priorities, and doesn't need to be re-introduced every session. The briefing overhead disappears. The context compounds.
That's what distinguishes heavy AI users who get dramatically better results from those who plateau at "it's useful sometimes." The difference is almost never model quality. It's almost always context quality.
Getting Started
The LLM Wiki is probably the highest-leverage single thing you can do to improve AI output quality. Not by changing the model, but by changing the quality of context it works from. Build it once, maintain it in 15 minutes a month, and the compounding benefits show up in every session afterward.
I publish free playbooks for building and maintaining your LLM Wiki at claudecodehq.com: the guided wiki builder, a second brain synthesis tool for extracting wiki content from your existing notes, and a conversation history mapper for pulling structured knowledge from past AI sessions. Each one is a single file you drop into a folder and start using immediately.
Originally published on claudecodehq.com
Top comments (0)