LLM Wiki / AI Second Brain / AI Knowledge Base

LLM Wiki — a self-maintaining personal knowledge management (PKM) system designed specifically for Large Language Models.

Instead of treating an AI like a search engine that loses context between queries, this system acts as an autonomous librarian. By dropping raw files into a designated folder, an LLM agent (such as GitHub Copilot or a custom ingestion pipeline) can analyze content, extract concepts, and compile a structured, interlinked markdown wiki optimized for Retrieval-Augmented Generation (RAG).

---

🧠 Core Philosophy

This project shifts away from passive note-taking systems into an active, agent-driven AI operating environment.

• LLM-Native Formatting: Every markdown file follows strict token-based chunking limits (maximum 512 tokens per section) for reliable context-window alignment.
• Anti-Hallucination Measures: Mandatory YAML frontmatter includes explicit confidence scoring (low, medium, high, verified) to improve retrieval reliability.
• Separation of Extraction and Synthesis: Raw source material remains isolated from derived knowledge and interpretation.
• Structured Linking: Wikilinks and hierarchical relationships create a navigable semantic graph optimized for RAG pipelines.
• Schema-Driven Organization: All generated content follows a strict ontology defined in Schema.md.

---

📂 Repository Structure

The workspace is divided into immutable raw inputs and the generated knowledge base.

.
├── raw/                    # Immutable source files (PDFs, TXT, Markdown, etc.)
├── wiki/                   # Generated interlinked knowledge base
│   ├── .obsidian/          # Obsidian vault configuration
│   ├── concepts/           # Canonical concepts, frameworks, and mental models
│   ├── dashboard/          # Dashboards, analytics, or operational views
│   ├── entities/           # People, organizations, products, or places
│   ├── examples/           # Implementations, case studies, and applied examples
│   ├── notes/              # Freeform thoughts and synthesis notes
│   ├── sources/            # Processed source summaries and metadata
│   ├── index.md            # Global knowledge map and entry point
│   └── log.md              # Append-only activity log
└── Schema.md               # System-wide schema and ingestion rules

---

📜 Schema.md: The Brain of the System

Schema.md is the authoritative specification for the repository.

It defines:

• YAML frontmatter requirements
• Allowed node types
• File naming conventions
• Chunking constraints
• Metadata normalization rules
• Wikilink behavior
• Retrieval-ranking logic
• Agent permissions and restrictions

Any automated agent interacting with this repository should read Schema.md before processing or generating content.

---

Workflow

Ingest Raw Data

Place unprocessed files into the raw/ directory:

• PDFs
• Markdown files
• Articles
• Research papers
• Meeting transcripts
• Notes
• Text exports
• Obsidian unprocessed folder

Process /raw with LLM

1. Pick one of the prompts inside Prompts/ folder;
2. Run an LLM agent of your choise or ingestion pipeline using this prompt;
3. Wait untill agent finish processing;
4. Open your wiki folder with obsidian to study result.

Expected Results

The ingestion system will:

• Extract and summarize content
• Generate structured markdown nodes
• Route files into the correct /wiki/ subdirectory
• Enforce YAML frontmatter validation
• Apply token-aware chunking rules
• Create wikilinks between related concepts
• Update wiki/index.md
• Append actions to wiki/log.md

Query Your LLM wiki

After you have successfuly processed you raw folder into LLM wiki you can use local LLM tooling or RAG systems to query your knowledge base.

---

🔗 Linking System

The wiki uses Obsidian-compatible wikilinks.

Examples:

[[Retrieval Augmented Generation]]
[[Vector Databases|Vector DBs]]
[[LLM Memory Systems#Long-Term Memory]]

Features include:

• Bidirectional linking
• Heading-level references
• Graph traversal
• Context-aware retrieval
• Semantic relationship mapping

---

🛠 Editor Compatibility

The /wiki/ directory is fully compatible with Obsidian.

Opening the wiki/ folder as an Obsidian vault enables:

• Graph visualization
• Native wikilink navigation
• Full-text search
• Tag filtering
• Markdown editing
• Plugin ecosystem support

---

⚙️ Design Goals

The system is designed for:

• Long-term AI memory
• Reliable RAG ingestion
• Structured semantic retrieval
• Low hallucination rates
• Human-readable markdown
• Agent interoperability
• Scalable knowledge graphs

---

📄 License

This project is licensed under the GNU General Public License v3.0 (GPL-3.0).

You may copy, distribute, and modify the software provided that changes remain documented and derivative works using GPL-licensed code are also distributed under the GPL.

See the LICENSE file for details.