Problem

Memory deletion endpoints only updated PostgreSQL state but never deleted from Qdrant vector store. This caused:

• Deleted memories still appearing in search results
• Vector store growing indefinitely
• Database and vector store getting out of sync

What was broken:

1. REST API (DELETE /api/v1/memories/): Only called update_memory_state() to mark as deleted in DB
2. MCP API (delete_all_memories): Called memory_client.delete(memory_id) but passed UUID instead of string

How It Works Elsewhere

The add_memories flow (both REST and MCP) properly syncs:

1. Add to vector store via memory_client.add()
2. Store metadata in PostgreSQL

Deletion should mirror this - remove from both stores.

The Fix

MCP API (delete_all_memories in mcp_server.py):

• Convert UUID to string: memory_client.delete(str(memory_id))
• Qdrant expects string IDs, not UUID objects

REST API (delete_memories in memories.py):

• Added vector store deletion before DB update:
  1. Get memory client
  2. Delete from Qdrant via memory_client.delete(str(memory_id))
  3. Mark as deleted in PostgreSQL

Both endpoints now properly sync deletions across storage layers.

Type of change

• Bug fix (non-breaking change which fixes an issue)

How Has This Been Tested?

• Test Script - verified memories deleted from both Qdrant and PostgreSQL
• Tested delete_all via MCP and bulk delete via REST API
• Confirmed deleted memories no longer appear in search results

Checklist:

• My code follows the style guidelines of this project
• I have performed a self-review of my own code
• My changes generate no new warnings
• Follows existing pattern from add_memories workflow