TechPrufer
Sign inGet started

Blogs›Qdrant›On-Device Vector Search

On-Device Vector Search

On-Device Vector Search

4
posts
2025–2026

Qdrant Edge enables on-device vector search for real-time anomaly detection in video streams. This system leverages a split architecture with an immutable HNSW shard for baseline normal activity and a mutable shard for live writes, allowing for sub-millisecond kNN lookups with no network dependency and full offline resilience. Video embeddings are provided by Twelve Labs Marengo 3.0, and NVIDIA Metropolis VSS orchestrates GPU-accelerated ingestion on Vultr Cloud GPUs. The system escalates anomalous clips to the cloud for higher-fidelity analysis, reducing cloud processing volume and costs while maintaining high anomaly detection rates. This approach extends beyond surveillance to manufacturing safety, retail analytics, and traffic monitoring.

2026

Memory at the Edge: On-Device Vector Search with Qdrant Edge - Qdrant

6/16/2026

Introduces Qdrant Edge, an embedded, in-process library for on-device vector search. Details its architecture as a library rather than a service, its small install footprint, and its integration via a simple Python API. Demonstrates use cases like robotic memory, edge anomaly triage, and private device memory. Highlights the performance benefits of local retrieval over cloud round-trips and outlines a 'local first, cloud when needed, sync between' pattern for edge deployments.

Video Anomaly Detection From Edge to Cloud With Qdrant - Qdrant

3/15/2026

This post details the implementation of a video anomaly detection system using Qdrant Edge. It describes how Qdrant Edge's two-shard architecture (immutable HNSW for baseline, mutable for live writes) enables on-device, offline vector search for anomaly detection. The system reframes anomaly detection as a nearest-neighbor search problem, indexing embeddings of normal activity and identifying anomalies based on distance from the normal baseline. It highlights the integration with Twelve Labs for video embeddings and NVIDIA Metropolis VSS on Vultr GPUs for cloud-tier processing, and explains the edge-to-cloud escalation strategy for cost and performance optimization.

2025

How Tavus used Qdrant Edge to create conversational AI - Qdrant

9/12/2025

Tavus implemented Qdrant Edge for their conversational AI, colocating per-conversation edge vector stores with conversational workers. This removed network latency, reducing retrieval time to 20-25ms and enabling end-to-end utterance-to-utterance timing near 500-600ms. The architecture allows for retrieval on every utterance and supports grounding every turn with private knowledge, improving conversational quality and simplifying onboarding for customers.

Qdrant Edge: Vector Search for Embedded AI - Qdrant

7/29/2025

Introduces Qdrant Edge, a new embedded, in-process library version of Qdrant for on-device vector search. Key features include in-process execution (no background threads), minimal footprint, and multitenancy support. It targets use cases like robotics, mobile devices, POS systems, and IoT agents where low-latency, offline, and privacy-preserving vector search is critical. The post emphasizes Qdrant's architectural strengths (HNSW, hybrid search, multivector) being adapted for edge environments.