LLM Wiki

An LLM Wiki is a persistent, agent-maintained markdown knowledge base where an LLM incrementally compiles sources into structured, cross-linked pages. The key distinction from ordinary retrieval is that knowledge is integrated once and then maintained, rather than rediscovered from raw chunks on every query.

The architecture has three layers:

• Raw sources: immutable source documents under raw/.
• Compiled wiki: summaries, entity pages, concept pages, comparisons, and filed queries maintained by the agent.
• Schema: rules such as this wiki's SCHEMA.md, which define page formats, naming, links, indexing, logging, and update policy.

The wiki should compound. Each source can update many pages, strengthen or challenge existing synthesis, and create links that future queries can reuse.

Why it matters

For Jay, the pattern turns recurring research into a durable asset: company research, fintech/payments analysis, role prep, product strategy memos, and essay ideas can accumulate in one navigable system instead of staying scattered across chats and source files.

Operating principles

• Raw sources stay immutable.
• The agent owns bookkeeping and maintenance.
• The human curates sources and directs judgement.
• Queries should read from compiled pages first.
• Valuable answers should be filed back into the wiki.