What is Feature URI

Feature URI - A universal address for any feature emitted by a Mixpeek extractor, enabling query-time model compatibility across the warehouse.

URI Format

mixpeek://{extractor_name}@{version}/{output_name}
extractor_name: one of the shipping feature extractors, listed by GET /v1/discovery/extractors (text_extractor, image_extractor, multimodal_extractor, universal_extractor, face_identity_extractor, audio_fingerprint_extractor, document_graph_extractor, gemini_multifile_extractor, web_scraper, scrolling_text_extractor, transcoding_extractor, passthrough_extractor)
version: the extractor version (e.g., v1, v2), which guarantees embedding compatibility
output_name: the inference service behind that extractor's vector index, which the same endpoint returns under required_vector_indexes (multilingual_e5_large_instruct_v1, google_siglip_base_v1, insightface__arcface)

Examples

mixpeek://text_extractor@v1/multilingual_e5_large_instruct_v1 - 1024D E5 text embedding
mixpeek://image_extractor@v1/google_siglip_base_v1 - 768D SigLIP image embedding
mixpeek://multimodal_extractor@v1/vertex_multimodal_embedding - 1408D unified video, image and text embedding
mixpeek://multimodal_extractor@v2/gemini-embedding-2 - 3072D Gemini Embedding 2
mixpeek://face_identity_extractor@v1/insightface__arcface - 512D ArcFace identity vector
mixpeek://audio_fingerprint_extractor@v1/laion__clap_htsat_tiny - 512D CLAP audio fingerprint

Where Feature URIs Are Used

Collection output schemas: define which features a collection produces
Retriever stages: the feature_address field in feature_search stages specifies which embedding to query
Taxonomies: reference which feature to classify against
Clustering jobs: specify the embedding space for vector grouping
Caching: inference cache uses URIs to shortcut repeated embedding requests

Why Feature URIs Matter

Model compatibility: prevents querying a CLIP embedding with an E5 query vector
Version safety: upgrading an extractor version creates new URIs, so old and new embeddings coexist
Lineage tracking: every feature in the warehouse can be traced back to its source extractor and version
Cross-collection queries: retrievers can reference features from multiple collections in the same pipeline

Core Concepts - Feature URIs: /docs/overview/concepts
Feature Extractors: /docs/processing/feature-extractors
Retriever Stages - Feature Search: /docs/retrieval/stages/feature-search
Warehouse Architecture: /docs/overview/warehouse-architecture

Put it to work: search your own files

Managed Mixpeek

Put multimodal search to work

Connect a bucket and Mixpeek runs the whole multimodal search pipeline for you: extraction, indexing, and search over your own objects. No models to wire up, nothing to host.

Start with Managed

MVS · bring your own