smartmemory-core 0.7.3__tar.gz → 0.7.4__tar.gz

{smartmemory_core-0.7.3 → smartmemory_core-0.7.4}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: smartmemory-core
-Version: 0.7.3
+Version: 0.7.4
 Summary: Multi-layered AI memory system with graph databases, vector stores, and intelligent processing pipelines
 Author: SmartMemory Team
 License-Expression: AGPL-3.0-or-later

smartmemory_core-0.7.4/VERSION

@@ -0,0 +1 @@
+0.7.4

smartmemory_core-0.7.4/smartmemory/search/recall_profile.py

@@ -0,0 +1,56 @@
+"""Recall profile re-ranking (CORE-MULTI-AGENT-1 Phase 3).
+
+Applies per-agent memory type weight multipliers to search results,
+enabling different agents to get different rankings from the same
+knowledge base. Default profile (empty/None) preserves current behavior.
+
+Recall profiles are dicts with optional keys:
+    memory_type_weights: dict[str, float]  — multipliers per memory type (default 1.0)
+
+channel_weights support is deferred until CORE-SEARCH-2 Phase 3 delivers
+named channel attribution. When available, channel_weights will multiply
+per-channel scores before RRF fusion (profile weights * channel weights).
+
+Example profile for a "skeptical" agent that favors decisions and procedures:
+    {"memory_type_weights": {"decision": 2.0, "procedural": 1.5, "episodic": 0.5}}
+"""
+
+from __future__ import annotations
+
+import logging
+from typing import Any, Optional
+
+logger = logging.getLogger(__name__)
+
+
+def apply_recall_profile(
+    results: list[Any],
+    profile: Optional[dict],
+) -> list[Any]:
+    """Re-rank search results by applying recall profile weights.
+
+    Args:
+        results: Search results (MemoryItem instances with .memory_type and optional score).
+        profile: Recall profile dict with optional memory_type_weights.
+                 None or empty dict = no-op (preserves original ranking).
+
+    Returns:
+        Re-ranked results list. Original list is not mutated.
+    """
+    if not profile or not results:
+        return results
+
+    weights = profile.get("memory_type_weights")
+    if not weights:
+        return results
+
+    def score_key(item: Any) -> float:
+        memory_type = getattr(item, "memory_type", None) or ""
+        multiplier = weights.get(memory_type, 1.0)
+        # Use existing score if available, otherwise use position-based score
+        base_score = getattr(item, "score", None)
+        if base_score is None:
+            base_score = 1.0
+        return base_score * multiplier
+
+    return sorted(results, key=score_key, reverse=True)

{smartmemory_core-0.7.3 → smartmemory_core-0.7.4}/smartmemory_core.egg-info/SOURCES.txt

@@ -418,6 +418,7 @@ smartmemory/retrieval/__init__.py
 smartmemory/retrieval/ssg_traversal.py
 smartmemory/search/__init__.py
 smartmemory/search/query_decomposer.py
+smartmemory/search/recall_profile.py
 smartmemory/search/rrf_merge.py
 smartmemory/similarity/__init__.py
 smartmemory/similarity/enhanced_metrics.py

smartmemory_core-0.7.3/VERSION

@@ -1 +0,0 @@
-0.7.3