Ask naturally. Get structured answers.
The agent asks for relevant context in natural language. MemState pulls the topic bundles that fit the question, then optionally expands along the graph or through field history. The default path returns the current truth; deeper paths add neighborhood and history when the question calls for it.
Three retrieval ideas
Meaning-first
Start from the topics whose stored content answers the question; internal retrieval narrows the set when the semantic stage runs, without treating the API as a similarity-score service.
Neighborhood context
Walk a bounded set of related topics when the answer spans more than one node.
How it evolved
Surface field histories when the user cares about change over time, not only the latest value.
How to phrase requests
Pass the agent or user question verbatim as the query text. Optional flags tell MemState to include neighbors or field history; you are not hand-assembling graph traversals in application code.