Memory Extraction

Automatically extract memories from conversations and documents.

Overview

The extraction endpoint creates and stores memories from the content you provide:

Content → Extracted memories → Stored memories

Basic Extraction

curl -X POST https://api.memory.tensorheart.com/v1/query/extract \
  -H "Authorization: Bearer $API_KEY" \
  -d '{
    "content": "User: Hi, I am Sarah and I work at Netflix as a product manager.",
    "content_type": "conversation"
  }'

Extracted memories:

• "User's name is Sarah"
• "User works at Netflix"
• "User is a product manager"

Content Types

Type	Best For
conversation	Chat logs, transcripts
document	Articles, reports
notes	Meeting notes, summaries

Adding Metadata

Tag extracted memories with source information:

{
  "content": "...",
  "metadata": {
    "source": "onboarding_call",
    "date": "2024-01-15",
    "session_id": "abc123"
  }
}

Best Practices

1. Clean input - Remove irrelevant content before extraction
2. Use metadata - Track where memories came from
3. Batch wisely - Extract from complete conversations, not fragments
4. Review results - Periodically audit extracted memories