GenWiki

An Incremental Personal Knowledge Base using LLMs for Obsidian

GenWiki is an Obsidian plugin inspired by Andrej Karpathy's "LLM Wiki" concept. It aims to solve the limitations of traditional RAG (Retrieval-Augmented Generation) systems—where search is ephemeral and knowledge doesn't compound—by using LLMs to incrementally build, structure, and audit a persistent personal knowledge base directly within your vault.

Unlike standard search tools, GenWiki parses your web clippings, merges new findings with your existing notes, structures them with bidirectionally-linked wiki files, and continuously audits your knowledge base for conflicts and gaps.

🌟 Core Features

1. Incremental Ingest (📥 Ingest Clippings)

• Scan & Parse: Scans your unprocessed clippings directory for raw web clippings.
• Smart Merge: Employs LLMs to read the new source material, look up your existing knowledge index, and decide whether to modify and merge existing Wiki pages or create new ones.
• Automatic Linking: Dynamically formats Obsidian internal WikiLinks ([[Entity_Name]]) to connect concepts automatically.
• Audit Trail: Archives the original clippings and registers audit logs to wiki/log.md.

2. Conversational Sidebar (💬 Chat Panel)

• Contextual Q&A: Asks questions based on your local compiled wiki. The plugin performs a fast, local index lookup to retrieve the top 5 most relevant wiki pages and feeds them to the LLM.
• Source Citations: Answers are fully referenced with standard wikilinks pointing to the source note (e.g., [[Tool_Obsidian#Features]]).
• Wiki Solidification (Save to Wiki): Explored insights or deep syntheses from your chats can be instantly saved as a permanent Wiki note with a single click. The plugin uses LLM to de-conversationalize the chat, creates aliases/summaries, and updates the local metadata index.

3. Knowledge Base Linter (🔍 Lint Wiki)

• Auditing: Performs health checks across your active Wiki nodes in batches.
• Issues Detected:
  • Contradictions: Highlights conflicts where different pages state conflicting facts.
  • Orphans: Detects isolated files that lack any inbound WikiLinks.
  • Stale Pages: Spots blank or underdeveloped placeholders.
• Audit Report: Outputs a thorough report containing precise improvement recommendations.

🛠️ Architecture & Data Flow

GenWiki enforces a clean three-tier structure inside your wiki/ directory:

wiki/
├── _skills/           # Modular Agent Prompts (Ingest.md, Query.md, Lint.md, SavePage.md)
├── _agents/           # System SOP Specifications (CLAUDE.md)
├── _database/         # Flat-file Local Database Index (index.json)
├── index.md           # Visual, card-style personal knowledge map
└── log.md             # Action audit logs

Pure Local Text Database

To bypass iOS sandbox constraints and support frictionless cross-device synchronization (e.g., via iCloud, Git, or Obsidian Sync), all indexing metadata (aliases, summaries, backlinks, and claims) is stored in a lightweight flat JSON database (_database/index.json). No binary database or server infrastructure is required.

🚀 Installation

Manual Installation

1. Download the latest release files (main.js, manifest.json, styles.css) from the Releases page.
2. Inside your Obsidian vault, navigate to .obsidian/plugins/ (create the directory if it doesn't exist).
3. Create a subfolder named genwiki and paste the three files inside it.
4. Open Obsidian, go to Settings $\rightarrow$ Community Plugins, and toggle GenWiki on.

⚙️ Configuration

Open Obsidian Settings and navigate to GenWiki under Community Plugins.

1. LLM Provider: Choose your preferred model provider:
   • Google Gemini (Default: gemini-3.5-flash)
   • Anthropic Claude (Default: claude-sonnet-4-6)
   • OpenAI (Default: gpt-5.4-mini)
   • DeepSeek (Default: deepseek-v4-flash, supports OpenAI/Anthropic-compatible endpoints)
   • Moonshot Kimi (Default: kimi-k2.6)
   • OpenRouter (Default: openrouter/free, handles unified model endpoints)
2. Model Selection:
   • Select a recommended model from the dropdown.
   • If you wish to use a custom or newly-released model, choose Custom, and enter the exact Model ID in the text field that appears.
3. Paths: Customize the source directory for raw clippings (default: Clippings) and the destination Wiki folder (default: wiki).

📖 How to Use

1. Setup: Make sure your API key is configured. On the first startup, GenWiki automatically sets up the folder structure and default Skill templates in your vault.
2. Start Chatting: Click the CPU icon on the left ribbon to pop open the GenWiki Chat Panel on the right side.
3. Ingest Clippings: Drop raw Markdown clippings into your Clippings/ folder, open the Chat Panel, and click 📥 Process Clippings at the top. The plugin will automatically parse, link, and index them.
4. Solidify Chat: Ask questions in the Chat Panel. If the assistant provides a detailed answer you want to keep, click 💾 Save as Wiki Page beneath the answer to compile it into your permanent knowledge wiki.

🧑‍💻 Customizing Prompts (Skills)

GenWiki is completely prompt-driven. You can inspect or rewrite the system prompts and instruction flows directly inside your Obsidian editor:

• wiki/_skills/Ingest.md — Incremental clipping extraction, linking, and contradiction checks.
• wiki/_skills/Query.md — Context retrieval guidelines and link citations.
• wiki/_skills/Lint.md — Knowledge audit heuristics.
• wiki/_skills/SavePage.md — De-conversationalizing text format rules.

📄 License

This project is licensed under the CC BY-NC-SA 4.0 License (Creative Commons Attribution-NonCommercial-ShareAlike 4.0 International). It is free to share and adapt for non-commercial purposes.