Related Memories API

Find semantically related memories using vector cosine similarity, with source-based fallback.

Endpoint

GET /api/v1/memories/related?id=<memoryId>&limit=6

Parameters

Param	Type	Default	Description
id	UUID	required	Source memory ID
limit	number	6	Max related memories (1-20)

Response

{
  "memoryId": "uuid",
  "related": [
    {
      "id": "uuid",
      "content": "...",
      "sourceType": "chatgpt",
      "sourceTitle": "...",
      "timestamp": "2026-03-25T...",
      "similarity": 85,
      "method": "vector"
    }
  ],
  "method": "vector",
  "total": 5
}

Strategy

1. Vector similarity — Uses pgvector cosine distance if the source memory has an embedding. Filters results below 0.3 similarity threshold.
2. Source-based fallback — If vector results are insufficient, falls back to same-source and same-title matching ordered by temporal proximity.

Usage

The Memory Drawer component (MemoryDrawer) automatically calls this endpoint when a memory is opened, showing related memories with similarity scores.

// Open a memory in the drawer from anywhere
import { openMemoryDrawer } from '@/components/MemoryDrawer';

openMemoryDrawer({ id, content, source, sourceTitle, timestamp });