Agent-Maintained Knowledge Base

An agent-maintained knowledge base is a repository where the LLM is responsible for the repetitive maintenance work: summarising, filing, cross-linking, updating pages, flagging contradictions, and keeping indexes/logs current.

The human remains responsible for source selection, emphasis, taste, and final judgement. The agent handles the maintenance burden that usually causes personal wikis to decay.

What the agent does

• Reads new sources and extracts durable claims.
• Updates existing entity and concept pages.
• Creates new pages only when a topic deserves to become a durable node.
• Maintains index.md and log.md.
• Flags contradictions rather than silently overwriting them.
• Files substantial query answers back into the wiki.

Why this is different from note-taking

The user does not need to manually write and maintain every note. The wiki becomes closer to a codebase: the schema is the spec, Obsidian or a text editor is the IDE, and the LLM is the maintainer making structured edits.

Risks

• Over-creating pages for minor mentions.
• Letting weak single-source claims harden into facts.
• Failing to update the index/log.
• Creating links without preserving provenance.