Memory, Information Library, and Context Retrieval

The Story Bible stores author-maintained structured canon. The Knowledge Graph connects entities, relations, and evidence. Memory stores durable facts, states, conflicts, and candidate memories. The Information Library is the shared search entry for manuscript, outlines, story data, graph, memory, and references.

Retrieval may combine semantic search, text matching, graph relations, and task rules; vectors are only one mechanism. New Agent-discovered facts first become candidates. Confirm them before they become official, and resolve conflicts by comparing evidence, time, and source rather than model confidence.

Keep sources when editing memory and inspect dependencies before deletion. Old versions support audit and recovery but do not necessarily participate in current retrieval.

Chapter notebook

The chapter notebook gathers the current outline, relevant canon, graph relations, memory, and manuscript evidence for one task. Review the sources before generation. If evidence is missing, add a retrieval clue, then decide whether to continue, keep the candidate, or discard it. The notebook is a task evidence surface: it does not send every local source at once or promote an unconfirmed candidate to official canon.