wang-junjian/wikillm
Data science & MLThis page was translated automatically. Technical details may differ from the original. View the original
WikiLLM: An AI autonomous personal knowledge base based on Andrej Karpathy's philosophy
Overview
A system for building personal knowledge bases using LLMs. WikiLLMs "compile" raw materials into structured, cross-linked, high-quality Chinese Wikis, which can be viewed on Obsidian. 1. : Source documents (articles, papers, codebases, datasets, images) are indexed to raw/directory 2. : LLM incrementally "compiles" raw data into a wiki with markdown files, including abstracts, backlinks, classification concepts, and interlinked articles 3. : Obsidian is used as a frontend to view raw data, compiled wikis, and visualizations 4. : LLMs can answer complex wiki questions by studying relevant data 5. : Results are rendered as markdown files, Marp slides, or matplotlib images, which can be viewed in Obsidian 6. : LLM "health check" detects inconsistencies, fills in missing data, and suggests new article candidates 7. : Additional tools such as naive search engines on wikis - Externalization in LLM Agents: A Unified Overview of Agent Memory/Skills/Protocols/Harness Engineering - Meta-Harness: End-to-End Optimization of Model Harnesses - Anthropic: Architecture Design of Managed Agents: Separation of Mind and Hand - Anthropic: Harness Design for Long Lifecycle Applications - OpenAI: Codex Harness Engineering in an Agent-First World - OpenAI: Original English Harness Engineering Guide - MiniMax M2.7 Model Self-Evolution Release Announcement - RedHat: AI-Assisted Structured Harness Workflow - Mitchell Hashimoto (Founder of HashiCorp) AI Application Implementation - NxCode: Complete Guide to Harness Engineering 2026 - LangChain: Deep Agent Optimization Based on Harness Engineering - Martin Fowler: Harness Engineering Practices for Coding Agent Users - Martin Fowler: Early Thoughts on Harness Engineering 1. Download and install Obsidian 2. Open this repository in Obsidian as a vault 3. Start exploring from wiki/INDEX.md - Markdown rendering, supports GFM format - Obsidian style wiki link parsing ([[Page| Label]]) - Sidebar navigation, organize pages by category - Image resource support - Responsive design
Overview (machine translated)
A system for building personal knowledge bases using LLMs. WikiLLMs “compile” raw materials into structured, cross-linked, high-quality Chinese Wikis, which can be viewed on Obsidian.
This project is built on the concept proposed by Andrej Karpathy. See also: LLM Knowledge Bases
Project Overview
The WikiLLM workflow includes:
- Data ingestion: Source documents (articles, papers, codebases, datasets, images) are indexed to the
raw/directory - Wiki Compilation: The LLM incrementally “compiles” the raw data into a wiki with markdown files, including abstracts, backlinks, categorization concepts, and interlinked articles
- IDE: Obsidian is used as a frontend for viewing raw data, compiled wikis, and visualizations
- Q&A: LLMs can answer complex wiki-specific questions by studying relevant data
- Output: The result is rendered as a markdown file, Marp slide, or matplotlib image, which can be viewed in Obsidian
- Linting: LLM “health checks” identify inconsistencies, fill in missing data, and suggest new article candidates
- Additional Tools: Additional tools such as simple search engines on wikis
Core Principles
- LLM writes and maintains all wiki data; Manual editing is rare
- User exploration and queries are archived back to wiki to enhance it
- System focused on markdown files and Obsidian-compatible formats
- Images are downloaded locally for easy reference by LLMs
Directory Structure
Current content
This wiki currently contains a comprehensive knowledge base about Harness Engineering, compiled from the following sources:
- Externalization in LLM Agents: A unified overview of agent memory/skills/protocols/harness engineering
- Meta-Harness: End-to-end optimization of the model Harness
- Anthropic: Architecture design of managed agents: separation of brain and hand
- Anthropic: Harness design for long lifecycle applications
- OpenAI: Codex Harness engineering in an agent-first world
- OpenAI: Original English Harness Engineering Guide
- MiniMax M2.7 model self-evolution announcement released
- RedHat: AI-assisted structured Harness workflow
- Mitchell Hashimoto (founder of HashiCorp): The journey of AI application implementation
- NxCode: Complete Guide to Harness Engineering 2026
- LangChain: A deep intelligent agent optimized based on Harness engineering
- Martin Fowler: Harness engineering practice for coding agent users
- Martin Fowler: Early Thoughts on Harness Engineering
Get Started Quickly
View in Obsidian
This page was translated automatically. Technical details may differ from the original.
Recommended Tools
Try a different keyword or remove a filter.
Install
npx skillfish add wang-junjian/wikillm