Search Documentation
Search across all documentation pages
search_memories
Search durable memories semantically with optional filters and recency fallback. Use the entity parameter for exact entity-based search instead of semantic search. Supports temporal decay, diversity tuning, and compact mode for token-efficient browsing.
Parameters#
| Parameter | Type | Required | Description |
|---|---|---|---|
| query | string | required | Natural language query (ignored when entity is provided) |
| topK | integer (1-50) | optional | Max results (default 10) |
| threshold | number (0-1) | optional | Similarity threshold 0-1 (default 0.4) |
| vaultId | string (UUID) | optional | Optional vault/project filter |
| memoryType | string | optional | Optional memory type filter: fact, skill, preference, constraint, task, episodic, correction |
| includeArchived | boolean | optional | Include archived memories |
| entity | string | optional | Search by entity name using JSONB containment instead of semantic search |
| compact | boolean | optional | Return truncated content previews (200 chars) instead of full content. Use read_memories for full content on specific IDs. |
| decayHalfLife | integer (1-365) | optional | Temporal decay half-life in days (default 30). Older memories score lower unless importance >= 4. |
| diversity | number (0-1) | optional | MMR diversity balance 0-1 (default 0.7). Higher = more relevance, lower = more diversity. |
Returns#
An array of memory objects matching the query, each with ID, content (or preview if compact), memory type, importance, confidence, entities, and similarity score.
Example#
json
{
"tool": "search_memories",
"arguments": {
"query": "database architecture decisions",
"memoryType": "constraint",
"topK": 5,
"compact": true
}
}