[enhancement] Support topics and entities as filter in Summary Views

[PrasanKumar93]

Summary

Summary Views currently support only four keys for filters and group_by: user_id, namespace, session_id, and memory_type. This proposal requests adding topics and entities as supported keys, enabling domain (topic) specific summary views that are far more useful for real-world applications.

Motivation

In a wealth advisor demo application, we have long-term memories tagged with topics like investment, retirement, cash flow. Currently, the only way to create a focused summary is to summarize all memories for a user and depend on LLM prompt property control the output. There's no way to pre-filter at the data level.

With topics and entities support, we could create summary views like:

{
  "name": "Investment Strategy Summary",
  "source": "long_term",
  "group_by": ["user_id"],
  //"prompt":"",
  "filters": { "topics": { "any": ["investment", "portfolio", "retirement"] } }
}

{
  "name": "Per-Topic User Summary",
  "source": "long_term",
  "group_by": ["user_id", "topics"]
}

{
  "name": "Entity-Focused Summary",
  "source": "long_term",
  "group_by": ["user_id"],
  "filters": { "entities": { "any": ["Vanguard"] } }
}

These would allow applications to generate targeted summaries like "What does the system know about this user's retirement planning?" or "Summarize all interactions involving Vanguard funds."

[PrasanKumar93]

Current Behavior : The _build_long_term_filters_for_view function in agent_memory_server/summary_views.py only maps four keys. Any other key (like topics or entities) is silently ignored.

Infrastructure Already Exists

• Filter classes exist -- Topics(TagFilter) and Entities(TagFilter) are defined in agent_memory_server/filters.py with full support for eq, ne, any, all, and startswith operators.

• Search accepts them -- search_long_term_memories() in long_term_memory.py already accepts topics: Topics | None and entities: Entities | None parameters.

• Memories store them -- Every long-term memory record has topics: list[str] and entities: list[str] fields populated during extraction.

The missing piece is the mapping in _apply_filter inside summary_views.py.