A new tool called LLM Wiki converts any LLM agent into a research engine with wiki capabilities. The system operates as an append-only, Markdown-native knowledge base that runs parallel multi-agent research workflows. It collects provenance-rich catalogs of artifacts and examples while preserving agent-session context.

LLM Wiki stores redacted context and feedback candidates under .sessions, rehydrating compact context for future interactions. The tool ingests URLs and files, archives topic wikis, and tracks durable inventory state across large datasets with manifest-based indexing. It compiles raw sources into synthesized articles with cross-references and audits whether outputs remain trustworthy.

Users can query the compiled knowledge base to generate various artifacts including reports, slides, study guides, and implementation plans. The project draws inspiration from Andrej Karpathy's LLM wiki concept, extending it into a practical toolkit for developers working with language models.

LLM Wiki addresses the challenge of maintaining persistent, verifiable knowledge across AI research sessions. By combining multi-agent workflows with provenance tracking, it creates a structured approach to knowledge accumulation that standard chatbots lack. The tool is particularly useful for technical teams building documentation or research systems.