LLM Wiki: Ready-to-use baseline templates and a new pattern implementation for evidential reasoning

The article introduces the LLM Wiki pattern, which shifts how knowledge is stored when working with language models. Instead of relying on chat history, it proposes an incremental wiki concept built by an agent from various sources. The author presents two key repositories: llm-wiki-baseline, featuring six ready-to-use 'second brain' templates for different domains, and llm-wiki-kbt, which focuses on a more rigorous architecture. The latter implementation utilizes typed YAML records, extensible data schemas, and a shared ontology, enabling programmatic contradiction detection and evidential reasoning. A Euclidean geometry wiki with an interactive knowledge graph is provided as a demonstration. This approach significantly improves data structure and the reliability of AI agent responses, making their operations more transparent and verifiable for developers aiming to build intelligent systems with long-term memory.