What is Mixpeek?

Mixpeek is a multimodal search platform designed for teams with large media libraries. It allows users to describe a scene in plain English and retrieve exact, timestamped moments from video, images, audio, and PDFs stored on object storage. The platform automatically extracts structured features like faces, scenes, transcripts, OCR, and embeddings, enabling cross-modal search and retrieval.

How to use Mixpeek?

To use Mixpeek, you can either bring your own vectors and use the Mixpeek Vector Store (MVS) for dense, sparse, and BM25 search, or connect your raw files (e.g., a bucket) and let Mixpeek handle extraction, embedding, and indexing. After connecting your data, you can build retrievers to compose multi-stage searches, filter, join, and rerank results, and use the API or SDKs to integrate search into your applications.

Core features of Mixpeek

Multimodal search across video, images, audio, and PDFs
Managed indexing with auto-extraction of scenes, faces, OCR, transcripts, and embeddings
Vector Store (MVS) for bring-your-own-vectors with dense, sparse, and BM25 search
Feature extractors for faces, scenes, transcripts, OCR, and fingerprints
Cross-modal joins to combine features like face, transcript, and on-screen text
Clustering and taxonomy discovery to surface unlabeled concepts
Integrations with S3, GCS, R2, Mux, LangChain, MCP, and more