The Problem

The central agent should not have to read the entire website or query the latest entities every time to understand what the website knows. As the cluster grows, such crawling becomes a major source of redundant work.

We need to investigate the concept of a local website knowledge map: a sufficiently compact representation that allows understanding the composition and structure of knowledge without copying all content into the central system.

Expected Outcome

The map should help answer questions such as: what topics and entities are present on the site, what materials are the sources of this knowledge, what has changed since the previous review, and what explicit connections already exist.

It is necessary to define a reasonable boundary between the map and the content itself. The website must remain the source of truth, while the central loop should be able to work with the description first and only request the full material when it is truly needed.

Related to the central information hub, synchronization, and the existing task on managing the quality of concepts and facts.

Research Questions

It should not be assumed yet that the map must be a graph, a set of embeddings, or a specific schema. We first need to understand the minimally useful composition of the map, the requirements for provenance, node identity, modification time, and relationships, as well as the extent to which it can be built automatically from existing Concepts, Facts, Resources, and other entities.