Skip to main content

​
Overview

Custom metadata allows you to attach arbitrary key-value pairs to your memories, enabling powerful filtering capabilities when querying. This is useful for categorizing documents by department, priority level, confidentiality, or any other custom attributes relevant to your application.

​
Adding Metadata

​
Text Memories

When adding text memories via /memories/add, include a metadata object with your custom fields: 
from hyperspell import Hyperspell

client = Hyperspell(api_key="API_KEY", user_id="YOUR_USER_ID")

memory = client.memories.add(
    text="Q1 planning meeting notes discussing product roadmap.",
    resource_id="meeting-2024-q1",
    metadata={
        "department": "engineering",
        "priority": 5,
        "confidential": True,
        "meeting_date": "2024-01-15T09:00:00Z"
    }
)
print(memory.resource_id)

​
Supported Value Types

Metadata values can be:
  • Strings: "department": "engineering"
  • Numbers: "priority": 5 or "score": 0.95
  • Booleans: "confidential": true
  • ISO 8601 Dates: "meeting_date": "2024-01-15T09:00:00Z"

​
File Uploads

When uploading files via /memories/upload, pass metadata as a JSON string in the form data: 
import json
from hyperspell import Hyperspell

client = Hyperspell(api_key="API_KEY", user_id="YOUR_USER_ID")

with open("quarterly_report.pdf", "rb") as file:
    memory = client.memories.upload(
        file=file,
        collection="reports",
        metadata=json.dumps({
            "department": "finance",
            "quarter": "Q1",
            "year": 2024,
            "confidential": True
        })
    )
print(memory.resource_id)
For file uploads, metadata must be passed as a JSON-encoded string since the endpoint uses multipart form data.

​
Updating Metadata

When you add a memory with the same resource_id, the metadata is merged with any existing metadata. New keys are added, and existing keys are overwritten: 
# First call creates the resource
client.memories.add(
    text="Initial notes",
    resource_id="doc-123",
    metadata={
        "department": "engineering",
        "priority": 3
    }
)

# Second call merges metadata
client.memories.add(
    text="Updated notes",
    resource_id="doc-123",
    metadata={
        "priority": 5,
        "reviewed": True
    }
)

# Result: metadata = { "department": "engineering", "priority": 5, "reviewed": True }

​
Querying with Metadata Filters

Use the options.filter parameter when querying to filter results by metadata. Filters use MongoDB-style operators and are combined with AND logic.

​
Basic Filtering

Filter by exact value match: 
response = client.memories.search(
    query="product roadmap",
    sources=["vault"],
    options={
        "filter": {
            "department": "engineering"
        }
    }
)
print(response.documents)

​
Multiple Conditions

Multiple conditions are combined with AND logic: 
response = client.memories.search(
    query="meeting notes",
    sources=["vault"],
    options={
        "filter": {
            "department": "engineering",
            "confidential": False
        }
    }
)

​
Comparison Operators

Use MongoDB-style operators for advanced filtering:
OperatorDescriptionExample
$eqEqual to (implicit){"priority": 5} or {"priority": {"$eq": 5}}
$neNot equal to{"department": {"$ne": "sales"}}
$gtGreater than{"priority": {"$gt": 3}}
$gteGreater than or equal{"priority": {"$gte": 3}}
$ltLess than{"priority": {"$lt": 5}}
$lteLess than or equal{"priority": {"$lte": 5}}
$inValue in list{"department": {"$in": ["engineering", "marketing"]}}

​
Complex Filter Examples

High priority, non-confidential documents: 
response = client.memories.search(
    query="important updates",
    sources=["vault"],
    options={
        "filter": {
            "priority": {"$gte": 4},
            "confidential": False
        }
    }
)
Documents from specific departments: 
response = client.memories.search(
    query="team updates",
    sources=["vault"],
    options={
        "filter": {
            "department": {"$in": ["engineering", "product", "design"]}
        }
    }
)
Exclude certain categories: 
response = client.memories.search(
    query="company news",
    sources=["vault"],
    options={
        "filter": {
            "category": {"$ne": "internal"}
        }
    }
)
Range queries on numeric values: 
response = client.memories.search(
    query="documents",
    sources=["vault"],
    options={
        "filter": {
            "priority": {"$gt": 2, "$lte": 8}
        }
    }
)

​
Combining with Date Filters

Metadata filters can be combined with the after and before date range options: 
response = client.memories.search(
    query="quarterly reports",
    sources=["vault"],
    options={
        "after": "2024-01-01",
        "before": "2024-12-31",
        "filter": {
            "department": "finance",
            "confidential": False
        }
    }
)

​
Metadata in Query Results

When you query documents, the custom metadata is included in each document’s metadata field alongside system fields: 
{
  "documents": [
    {
      "resource_id": "meeting-2024-q1",
      "source": "vault",
      "title": "Q1 Planning Meeting",
      "metadata": {
        "status": "completed",
        "indexed_at": "2024-01-15T10:30:00Z",
        "department": "engineering",
        "priority": 5,
        "confidential": true,
        "meeting_date": "2024-01-15T09:00:00Z"
      },
      "highlights": [...]
    }
  ]
}
You can access this metadata in your code: 
response = client.memories.search(
    query="engineering roadmap",
    sources=["vault"],
    options={
        "filter": {"department": "engineering"}
    }
)

for doc in response.documents:
    print(f"Title: {doc.title}")
    print(f"Department: {doc.metadata.get('department')}")
    print(f"Priority: {doc.metadata.get('priority')}")

​
Best Practices

  1. Use consistent key names across your application to enable reliable filtering
  2. Keep metadata flat - nested objects are not supported for filtering
  3. Use appropriate types - numbers for numeric comparisons, booleans for true/false values
  4. Plan your taxonomy - decide on standard values for categorical fields like department or category
  5. Don’t over-filter - metadata filtering happens after semantic search, so overly restrictive filters may exclude relevant results