EngineFeature
Multimodal Extractor v2 with Gemini Embedding 2
New multimodal extractor generates 3072-dimensional embeddings using Gemini Embedding 2, enabling richer cross-modal search across text, images, and video.
May 2026
PlatformFeature
Plugin Marketplace
Publish, discover, and install custom extractors across organizations. Includes typed SDK, manifest validation, CLI lint/test, auto-versioning, rollback, and BYO container support.
APIFeature
Source Adapters — Iconik, Webhook, Email, Supabase
Connect external sources to Mixpeek with zero code. Iconik adapter for video DAMs, push-based webhook ingestion for any source, inbound email connector, and Supabase Storage. Org-level connections configure once and reuse across buckets.
APIFeature
Feature Search endpoint
New POST /v1/features/search endpoint for direct vector similarity search across extracted features, without needing a retriever.
EngineFeature
Deploy-resilient batch jobs with auto-resubmit
Batch jobs now survive cluster deployments. If a worker dies mid-run, the job is automatically resubmitted from where it left off. Combined with zombie batch TTL that auto-cancels stale drafts and fails stuck jobs.
APIFeature
Queue ETA on batch jobs
Batch jobs now show estimated time to completion based on historical job durations. Visible in the API response and Studio batch details.
April 2026
Vector StoreFeature
MVS High Availability with automatic failover
Primary-replica architecture with WAL-based replication, automatic failover with split-brain prevention, per-namespace recovery gates, atomic snapshots, and multi-namespace pods with lazy-load and LRU eviction.
Vector StorePerformance
Serverless cold start under 1 second at 833K vectors
MVS serverless namespaces now boot in under 1 second even at 833K points, down from 30-60 seconds. Achieved by skipping monolithic shard.bin in serverless mode and lazy-loading partitions.
Vector StoreFeature
MVS HPA autoscaling + gRPC load balancing
Vector store replicas now autoscale based on query load using Kubernetes HPA. Client-side gRPC load balancing distributes queries across all healthy replicas.
Vector StoreFeature
HNSW centroid routing + RaBitQ quantization
New retrieval strategy combines HNSW centroid routing with RaBitQ quantization for faster approximate nearest neighbor search with lower memory footprint.
APIFeature
Annotations API for human-in-the-loop feedback
New Annotations API lets you attach human feedback to documents and features. Supports bulk create, update, and delete — useful for fine-tuning retrieval relevance and training custom models.
APIFeature
Lineage-aware $expand with ancestors and descendants
Query results can now expand to include parent and child documents using $expand keywords. Filter aliases make complex lineage queries simple. New /ancestors and /descendants endpoints.
APIPerformance
Retriever-level and rerank stage caching
Retrievers now cache results at both the retriever and rerank stage level. Repeated queries return in single-digit milliseconds instead of running the full pipeline.
APIFeature
BYO inference on retriever stages
Bring your own inference to retriever pipelines with feature_uri on rerank and LLM stages, plus a new classify stage. Run your own models inline during retrieval.
APIFeature
Video moment-level retrieval with REDUCE stage
New moment_group REDUCE stage groups search results by video moment, returning the most relevant segment of each video instead of individual frames.