Product Updates

Active Tasks Listing & Enhanced Taxonomy Classifications

We’ve added the ability to list active processing tasks and enhanced our taxonomy classification system with new assignment modes.

Active Tasks

You can now monitor all active processing tasks across your collections, providing better visibility into ongoing operations.

Enhanced Taxonomy Classifications

The taxonomy classification system now supports two assignment modes:

• Threshold Mode: Assigns categories based on a confidence threshold
• Nearest Mode: Assigns to the nearest matching category regardless of confidence

{
  "collections": ["string"],
  "filters": {},
  "assignment": {
    "enabled": false,
    "mode": "threshold",  // or "nearest"
    "append": false,
    "confidence_threshold": 0.8
  },
  "limit": 1000
}

Key Updates

• Monitor active tasks with status, progress, and estimated completion time
• Choose between threshold-based or nearest-neighbor taxonomy assignments
• Control whether new classifications append to or replace existing ones
• Set confidence thresholds for more precise category matching

For detailed implementation guidelines, see our Taxonomy Classification Guide.

Breaking Change: Added “/v1” Prefix to All API Endpoints

We’ve standardized our API routing by adding a “/v1” prefix to all API endpoints. This change improves API versioning and follows REST API best practices.

Required Changes

• Update all API endpoint calls to include the “/v1” prefix
• Before: https://api.mixpeek.com/assets
• After: https://api.mixpeek.com/v1/assets

Affected Endpoints

All API endpoints are affected. Common examples include:

• /features/search → /v1/features/search
• /ingest/videos/url → /v1/ingest/videos/url
• /collections → /v1/collections

Migration Guide

1. Audit your API calls
2. Update endpoint URLs to include the “/v1” prefix
3. Test your integration with the new endpoints

Feature: Averaged Multi-Input Embeddings

Added support for averaging multiple embeddings when the same embedding model is used across different inputs. This allows for creating more representative embeddings by combining multiple perspectives or descriptions of the same content.

Examples

# Before: Later inputs would overwrite earlier ones
settings = [
    EmbeddingRequest(
        type="text",
        value="a dog running",
        embedding_model="clip"
    ),
    EmbeddingRequest(
        type="text",
        value="canine in motion",
        embedding_model="clip"
    )
]
# Result:
# {
#     "vectors": {
#         "clip": [0.2, 0.3, 0.5]  # Only keeps last embedding
#     },
#     "original_values": {
#         "clip": "canine in motion"  # Only keeps last value
#     }
# }

# After: Embeddings are averaged and all inputs are preserved
settings = [
    EmbeddingRequest(
        type="text",
        value="a dog running",
        embedding_model="clip"
    ),
    EmbeddingRequest(
        type="text",
        value="canine in motion",
        embedding_model="clip"
    )
]
# Result:
# {
#     "vectors": {
#         "clip": [0.25, 0.35, 0.45]  # Average of both embeddings
#     },
#     "original_values": {
#         "clip": [
#             {
#                 "type": "text",
#                 "value": "a dog running",
#                 "embedding_model": "clip"
#             },
#             {
#                 "type": "text",
#                 "value": "canine in motion",
#                 "embedding_model": "clip"
#             }
#         ]
#     }
# }

Use Cases

• Combine multiple text descriptions to create a more comprehensive embedding
• Mix image and text embeddings in the same vector space
• Preserve all original inputs for reference and debugging

Technical Details

• Embeddings from the same model are averaged element-wise
• All original EmbeddingRequest objects are stored in original_values
• Works with both text and image inputs
• Compatible with all supported embedding providers (Vertex, SageMaker)

Enhancement: Read-Only API Keys

You can now create API keys with granular permissions: read-only, write-only, or read-write access. This enables building secure client-side applications by restricting access to sensitive operations.

Permission Levels

• Read-only: Allows access to GET endpoints only (e.g., /api/v1/search, /api/v1/items/{id})
• Write-only: Enables POST, PUT, DELETE operations (e.g., /api/v1/items, /api/v1/batch)
• Read-write: Full access to all endpoints

For a complete list of endpoints and their required permissions, see our API Reference.

How to Configure

1. Navigate to Settings > API Keys in the studio portal
2. Create a new key or edit an existing one
3. Select the desired permission level
4. Save changes

Breaking Change: vector_index renamed to embedding_models

We’ve renamed the vector_index configuration parameter to embedding_models to better reflect its purpose and align with industry terminology. This change affects all API endpoints and configuration files that previously used vector_index.

What You Need To Do

• Update all references to vector_index in your configuration files to use embedding_models
• Review and update any API calls that included the vector_index parameter
• Check your integration tests and documentation for any mentions of the old parameter name

For a complete list of supported embedding models and migration guidance, please refer to our available models documentation.

Taxonomies and Clustering

Introducing powerful taxonomies to organize and classify your content using multimodal AI. Create hierarchical categories that understand both text and images, enabling intelligent content organization at scale.

Key Features

• Flexible Hierarchy - Build nested categories with independent node matching and confidence scoring
• Multimodal Support - Define taxonomy nodes using text, images, or both
• Test-Driven Development - Validate taxonomies on sample data before production deployment
• Automated Classification - Automatically categorize new content during ingestion
• Manual Controls - Override or supplement AI classifications with manual assignments

Check out our taxonomy documentation to get started organizing your content library.