consolidate_memories
Merge similar memories using AI to resolve conflicts and create consolidated notes. Supports dry-run preview or direct application modes for memory cluster management.
Instructions
Consolidate similar memories using LLM-driven merging (NOT YET IMPLEMENTED).
This tool will use an LLM to intelligently merge similar memories,
resolve conflicts, and create consolidated notes. Currently returns
a placeholder message.
Args:
cluster_id: Cluster ID to consolidate.
mode: Operation mode - "dry_run" or "apply".
Returns:
Consolidation results (when implemented).Input Schema
TableJSON Schema
| Name | Required | Description | Default |
|---|---|---|---|
| cluster_id | Yes | ||
| mode | No | dry_run |
Implementation Reference
- The main handler function for the 'consolidate_memories' tool, decorated with @mcp.tool() which handles both implementation and registration. Includes input validation, auto-detection clustering, preview, apply, and link modes for memory consolidation.
@mcp.tool() def consolidate_memories( cluster_id: str | None = None, mode: str = "preview", auto_detect: bool = False, cohesion_threshold: float = 0.75, ) -> dict[str, Any]: """ Consolidate similar memories using algorithmic merging or linking. This tool handles clusters in three ways: 1. MERGE (mode="apply"): Combine memories into one (high cohesion ≥0.75) 2. LINK (mode="link"): Create 'related' relations without merging (medium cohesion 0.40-0.75) 3. PREVIEW (mode="preview"): Show what would happen without making changes Merging intelligently: - Combines content (preserving unique information) - Merges tags and entities (union) - Calculates appropriate strength based on cohesion - Preserves earliest created_at and latest last_used timestamps Linking creates bidirectional 'related' relations to form knowledge graph connections. Modes: - "preview": Generate merge preview without making changes - "apply": Execute the consolidation/merge (requires cluster_id or auto_detect) - "link": Create relations between cluster members without merging Args: cluster_id: Specific cluster ID to act on (valid UUID, required unless auto_detect=True). mode: Operation mode - "preview", "apply", or "link". auto_detect: If True, automatically find high-cohesion clusters. cohesion_threshold: Minimum cohesion for auto-detection (0.0-1.0, default: 0.75). Returns: Consolidation/linking preview or execution results. Raises: ValueError: If cluster_id is invalid or cohesion_threshold is out of range. """ # Input validation if cluster_id is not None: cluster_id = validate_uuid(cluster_id, "cluster_id") cohesion_threshold = validate_score(cohesion_threshold, "cohesion_threshold") if mode not in ("preview", "apply", "link"): raise ValueError(f"mode must be 'preview', 'apply', or 'link', got: {mode}") # Auto-detect mode: find clusters worth consolidating if auto_detect: memories = db.list_memories(status=MemoryStatus.ACTIVE) # Create cluster config cluster_config = ClusterConfig( strategy="similarity", threshold=cohesion_threshold, max_cluster_size=12, min_cluster_size=2, use_embeddings=True, ) clusters = cluster_memories_simple(memories, cluster_config) # Filter to high-cohesion clusters worth consolidating candidates = [c for c in clusters if c.cohesion >= cohesion_threshold] if mode == "preview": # Show top candidates previews = [] for cluster in candidates[:5]: # Top 5 candidates preview = generate_consolidation_preview(cluster) previews.append(preview) return { "success": True, "mode": "auto_detect_preview", "candidates_found": len(candidates), "showing": len(previews), "previews": previews, "message": f"Found {len(candidates)} clusters ready for consolidation", } elif mode == "apply": # Apply consolidation to all candidates results = [] for cluster in candidates: result = execute_consolidation(cluster, db, centroid_embedding=cluster.centroid) results.append(result) total_saved = sum(r.get("space_saved", 0) for r in results) return { "success": True, "mode": "auto_detect_apply", "consolidated_clusters": len(results), "total_memories_saved": total_saved, "results": results, "message": f"Consolidated {len(results)} clusters, saved {total_saved} memory slots", } else: # mode == "link" # Link all candidates without merging results = [] for cluster in candidates: result = link_cluster_memories(cluster, db) results.append(result) total_relations = sum(r.get("relations_created", 0) for r in results) return { "success": True, "mode": "auto_detect_link", "linked_clusters": len(results), "total_relations_created": total_relations, "results": results, "message": f"Linked {len(results)} clusters with {total_relations} relations", } # Specific cluster mode if not cluster_id: return { "success": False, "error": "cluster_id is required when auto_detect is False", "hint": "Use auto_detect=True to find clusters automatically", } # Find the cluster (need to re-cluster to get the cluster object) memories = db.list_memories(status=MemoryStatus.ACTIVE) cluster_config = ClusterConfig( strategy="similarity", threshold=0.75, max_cluster_size=12, min_cluster_size=2, use_embeddings=True, ) clusters = cluster_memories_simple(memories, cluster_config) target_cluster = next((c for c in clusters if c.id == cluster_id), None) if not target_cluster: return { "success": False, "error": f"Cluster {cluster_id} not found", "hint": "Cluster IDs change on each run. Use auto_detect or get fresh cluster IDs from cluster_memories tool", } if mode == "preview": preview = generate_consolidation_preview(target_cluster) return { "success": True, "mode": "preview", **preview, } elif mode == "apply": result = execute_consolidation( target_cluster, db, centroid_embedding=target_cluster.centroid ) return { "success": True, "mode": "apply", **result, } elif mode == "link": result = link_cluster_memories(target_cluster, db) return { "success": True, "mode": "link", **result, } else: return { "success": False, "error": f"Unknown mode: {mode}", "valid_modes": ["preview", "apply", "link"], }