Search memory

Four engines run in parallel — vector similarity, graph traversal, episodic (temporal) recall and the answer cache — and their results are reranked before synthesis. You do not choose engines; the query class determines the weighting.

Cached answers are invalidated automatically when new content lands in the same scope, so a repeat query after an ingest returns fresh results rather than a stale hit.

Cost: charged as search.query.

curl -X POST "https://example.com/api/v1/memory/search" \  -H "Content-Type: application/json" \  -d '{    "query": "string"  }'

{  "answer": "string",  "citations": [    {      "index": 0,      "title": "string",      "source": "string",      "document_id": "string",      "snippet": "string",      "url": "string"    }  ],  "query": "string",  "chunk_count": 0,  "graph_fact_count": 0,  "episode_count": 0,  "suggested_followups": [    "string"  ],  "chunks_by_source": {}}