Memory Extraction

Automatically extract and store memories from text content, files, and URLs.

Overview

The Memory API provides two approaches for extracting memories:

1. Text Extraction - Extract memories from text content (conversations, documents, notes)
2. File Ingestion - Upload and process files directly (PDFs, images, audio, etc.)

Text Extraction

Extract memories from text content using the extraction endpoint.

Basic Usage

curl -X POST https://memoryapi.tensorheart.com/v1/query/extract \
  -H "Authorization: Bearer $API_KEY" \
  -H "Content-Type: application/json" \
  -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"

Request Parameters

Parameter	Type	Required	Description
content	string	Yes	The text content to extract memories from (max 500,000 characters)
content_type	string	No	Type of content: conversation, document, or notes (default: conversation)
metadata	object	No	Custom metadata to attach to all extracted memories
space_id	string	No	Memory space to store extracted memories in

Content Types

Type	Best For
conversation	Chat logs, transcripts, dialogue
document	Articles, reports, long-form text
notes	Meeting notes, summaries, bullet points

Adding Metadata

Tag extracted memories with source information for better organization:

{
  "content": "Meeting with client discussed Q4 goals...",
  "content_type": "notes",
  "metadata": {
    "source": "client_meeting",
    "date": "2024-01-15",
    "session_id": "abc123"
  }
}

---

File Ingestion

Upload files directly for processing into memories. The API automatically extracts text content and creates searchable memories.

Supported File Formats

Format	Extensions	Description
PDF	.pdf	Text extraction with optional OCR for scanned pages
Images	.png, .jpg, .jpeg, .gif, .webp	Text and content extraction from images
Audio	.mp3, .wav, .m4a, .ogg	Speech-to-text transcription
Plain Text	.txt, .md, .csv	Direct text processing

Upload a File

curl -X POST https://memoryapi.tensorheart.com/v1/ingest/file \
  -H "Authorization: Bearer $API_KEY" \
  -F "file=@document.pdf" \
  -F "chunk_size=500" \
  -F "overlap=50" \
  -F "ocr=true"

File Upload Parameters

Parameter	Type	Default	Description
file	file	Required	The file to upload and process
chunk_size	integer	500	Target tokens per chunk
overlap	integer	50	Overlap tokens between chunks
ocr	boolean	true	Enable OCR for scanned PDF pages
space_id	string	null	Memory space to store memories in

Response

{
  "success": true,
  "data": {
    "document_id": "doc_abc123def456",
    "status": "completed",
    "memories_created": 12,
    "chunks_created": 12,
    "tokens_extracted": 5840
  }
}

---

URL Ingestion

Ingest content directly from a URL. Supports web pages and PDF links.

Ingest from URL

curl -X POST https://memoryapi.tensorheart.com/v1/ingest/url \
  -H "Authorization: Bearer $API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "url": "https://example.com/article",
    "chunk_size": 500,
    "overlap": 50
  }'

URL Support

Type	Description
Web pages	Extracts main article/content from HTML pages
PDF URLs	Downloads and processes as PDF

---

Managing Documents

Get Document Status

curl https://memoryapi.tensorheart.com/v1/ingest/{document_id} \
  -H "Authorization: Bearer $API_KEY"

List Documents

curl "https://memoryapi.tensorheart.com/v1/ingest?limit=50&offset=0" \
  -H "Authorization: Bearer $API_KEY"

Get Document Chunks

View individual chunks from a processed document:

curl https://memoryapi.tensorheart.com/v1/ingest/{document_id}/chunks \
  -H "Authorization: Bearer $API_KEY"

---

Best Practices

1. Choose the right method - Use text extraction for structured text, file ingestion for documents
2. Clean input - Remove irrelevant content before text extraction
3. Use metadata - Tag memories with source information for better retrieval
4. Tune chunk size - Larger chunks preserve more context, smaller chunks enable finer retrieval
5. Enable OCR - For scanned PDFs, keep OCR enabled to extract text from images
6. Batch wisely - Extract from complete conversations or documents, not fragments
7. Use spaces - Organize memories into spaces for different projects or contexts
8. Review results - Periodically audit extracted memories for quality