Discover and explore top open-source AI tools and projects—updated daily.
ussumantLLM knowledge compilation for agents and developers
Top 95.8% on SourcePulse
This project addresses the inefficiency and high cost of LLM context windows when processing numerous scattered knowledge files or codebases. It compiles these sources into a topic-based wiki, drastically reducing token usage (~90%) and providing AI agents with a synthesized understanding of any project, implementing Andrej Karpathy's LLM Knowledge Base pattern.
How It Works
The core approach leverages an LLM to compile raw markdown files or entire code repositories into a structured, topic-based wiki. This synthesized knowledge base replaces the need for LLMs to re-read hundreds of individual files per session. The system prioritizes efficiency and knowledge consolidation, allowing agents to query concise articles instead of extensive raw data, thereby reducing context costs and improving response times.
Quick Start & Requirements
claude plugin install llm-wiki-compiler) or a local Codex plugin./wiki-init to set up, /wiki-compile to generate the wiki (5-10 min first run), and /wiki-visualize for a knowledge graph. Codex users invoke similar workflows via prompts./wiki-init guide setup.Highlighted Details
/wiki-visualize aids navigation and understanding of topic relationships./fetch-bookmarks, with scheduled syncing options.[coverage: high/medium/low] guide LLM fallback to raw files, balancing speed and accuracy.[[wikilinks]].Maintenance & Community
The provided README does not detail specific contributors, sponsorships, partnerships, or community channels.
Licensing & Compatibility
Limitations & Caveats
Requires Node.js 20+ and Chrome for certain features (e.g., X bookmark syncing). While reducing LLM context costs, initial compilation and updates incur LLM token expenses. /wiki-compile remains a manual user-initiated process.
2 weeks ago
Inactive