广场

Fine-Grained RAG Permissions with Qdrant and SpiceDB

Fine-Grained RAG Permissions with Qdrant and SpiceDB Time: 45 min Level: Intermediate Output: GitHub A standard RAG pipeline answers questions using information retrieved from your documents. Qdrant stores their embeddings and associated metadata and finds relevant documents to provide as context to the language model. But without an authorization step, any user can get an answer from any document in the collection, including documents they aren’t allowed to read. The OWASP Top 10 for LLM Applications lists this kind of leak as Sensitive Information Disclosure.

QdrantAI 工具英文

Embedding at the Speed of Light

Embedding a large corpus is usually the slowest part of building a vector search pipeline: transformer-based models capture semantics very well, but they are heavy on CPU, so you either wait days for the embeddings or pay for a costly GPU. Static embedding models avoid that cost by skipping the neural network altogether. A static model is a lookup table that associates each token in the model’s vocabulary with an embedding: to embed a text, you tokenize it, look up the vector of each token, and average them. With potion-base-8M , this let us embed 32.9 million passages on a laptop CPU in 35 minutes, where a transformer-based model would have needed about 53 days. The tradeoff is retrieval qu…

QdrantAI 工具英文

Universal I/O in Qdrant

Most software is built on an unspoken assumption: data lives on disk, and is read from and written to it. The assumption usually holds, until a new way (or need) of storing data appears. For Qdrant, that happened with serverless mode , where the data sits in object storage instead of on a disk next to the process. This blog tells the story of Universal I/O, the interface we built so that the same code could work in both worlds, and of the design choices that keep the local path cheap. This read is particularly relevant if you are a systems programmer, and some knowledge of Rust will help you follow the code, but the lessons should carry over to many other kinds of software.

QdrantAI 工具英文

Semantic Caching for RAG

Build a Semantic Cache for a RAG Pipeline Time: 30 min Level: Intermediate What’s a Semantic Cache Traditional caches operate on an exact match basis, while semantic caches search for the meaning of the key rather than an exact match. For example, “What is the capital of Brazil?” and “Can you tell me the capital of Brazil?” are semantically equivalent: a semantic cache can recognize that.

QdrantAI 工具英文

miniCOIL EN-ES: Sparse Neural Retrieval Across the Language Barrier

Sparse neural retrieval finally started getting more and more attention ( SPARSEUP , MILCO , sparse encoders in Sentence Transformers …). Well, amazing, we like attention! Our miniCOIL v1 sparse neural retriever article ended with a promise: to keep working on this sparse neural retriever in-depth, improving the model’s quality, and in-width, “extending it to other dense encoders and to languages beyond English.” Here is the next “in-width” step of the miniCOIL saga: to try and cross the language barrier, that is, make the model suitable for multilingual retrieval. It’s a fun challenge, as sparse retrievers are hard to adapt to this scenario: exact matching in cross-lingual text search usual…

QdrantAI 工具英文

Why Hyperbolic Embeddings Matter for Hierarchical Data

We choose embedding models, dimensions, and indexes. The geometry usually comes with the package. But why use a flat space, and what else could we choose? What Is a Manifold, and Where Do Our Vectors Live? A manifold is the space our embeddings live in. For embeddings, we care about the geometry we give that space. It determines how we measure distance, what the shortest path looks like, and how much room there is as we move outward.

QdrantAI 工具英文

Memory Tiers in Qdrant: What to Use and When

Memory Tiers in Qdrant: What to Use and When A growing vector collection eventually outgrows the RAM it started with: Qdrant handles that by letting you assign dense vectors, the HNSW graph, quantized vectors, payloads, and payload indexes each to whichever memory tier that structure supports, instead of forcing one RAM-versus-disk trade-off onto the whole collection. This article will give you practical guidance over which combination of tiers and quantization to reach for at each stage of a collection’s growth, and the reasons behind the choice.

QdrantAI 工具英文

Low-Latency Search

Tips for Low-Latency Search with Qdrant Create Payload Indexes If your search queries include filters, create payload indexes for the fields you filter on. Payload indexes are the primary way to improve filtered search performance in Qdrant. For best results, create payload indexes before uploading data. Queries that filter on unindexed fields are not only slower; they can also unnecessarily consume cluster resources, negatively impacting the latency of other search queries. Consider blocking queries that filter on unindexed fields . This rejects queries that would degrade performance at the API boundary, surfacing misconfigured indexes as errors rather than latency spikes.

QdrantAI 工具英文

From Amazon S3 Vectors

Migrate from Amazon S3 Vectors to Qdrant What You Need from AWS S3 bucket name — the bucket containing your vector data Index name — the S3 Vectors index to migrate AWS credentials — configured via aws configure or environment variables ( AWS_ACCESS_KEY_ID , AWS_SECRET_ACCESS_KEY ) Set your AWS credentials using the AWS CLI's configure command or environment variables before running the migration container. Run the Migration docker run --net = host --rm -it \ -e AWS_ACCESS_KEY_ID = 'your-access-key' \ -e AWS_SECRET_ACCESS_KEY = 'your-secret-key' \ -e AWS_REGION = 'us-east-1' \ registry.cloud.qdrant.io/library/qdrant-migration s3 \ --s3.bucket 'your-bucket-name' \ --s3.index 'your-index-name'…

QdrantAI 工具英文

Changelog

Changelog 1.15.0 (2026-11-05) Component Version qdrant-kubernetes-api v1.41.0 operator 2.37.2 qdrant-cluster-manager v0.3.25 qdrant-cluster-exporter 1.8.3 Latest validated Qdrant version: 1.20.0 Fixed global quotas being re-enabled after a user disabled or changed them through the Qdrant quotas API. For Qdrant v1.20.0 and later, the operator now renders its default 90% quotas into storage.quotas in the Qdrant configuration instead of calling the quotas API. Quotas set through the API take precedence, so a user’s change stays. Running pods are not restarted. They pick up the defaults when they are next recreated The operator no longer writes status.globalQuotas on QdrantCluster Updated operat…

QdrantAI 工具英文

Blue-Green Cluster Deployment

Blue-Green Cluster Deployment Time: 30 min Level: Intermediate A blue-green deployment runs two clusters in parallel: one live cluster serving production traffic (blue), and one staging cluster (green) where you prepare and validate changes. Once the green cluster is ready, you switch application traffic to it and decommission the old cluster. This is different from a blue-green migration at the collection level , where both collections live inside the same cluster. With collection-level blue-green, a bad upload or a heavy indexing run can still affect the production cluster. A cluster-level strategy removes that risk entirely by keeping the risky work on a separate cluster.

QdrantAI 工具英文

Automate Hybrid Cloud

Automating Hybrid Cloud You can create Hybrid Cloud environments and clusters without the Console. This page shows how with the qcloud CLI , Terraform , and the Cloud API . All of them authenticate with a Cloud Management Key . Hybrid Cloud access must be enabled for your account by the Qdrant sales team. The workflow has three steps: Create the Hybrid Cloud environment. Bootstrap your Kubernetes cluster into it, by running the generated commands against the cluster. They install the Operator and the Agent, as described in Setup . Create Qdrant clusters in the environment. qcloud CLI # 1. Create the environment qcloud hybrid create --name my-hybrid-env --namespace qdrant-hybrid # 2. Generate…

QdrantAI 工具英文

Cluster Metrics Reference

Cluster Metrics Reference Each Qdrant Cloud cluster exposes the metrics below at its /sys_metrics endpoint, in a Prometheus-compatible format. The endpoint is not available when running open source Qdrant. To scrape it, see Scrape Metrics with Prometheus and Grafana . /sys_metrics combines the database metrics from /metrics with infrastructure data from the load balancers, ingresses, and cluster workloads. Query it once per cluster, on the cluster endpoint. It reports on all nodes. For the per-node database metrics, see Qdrant monitoring .

QdrantAI 工具英文

Scrape Metrics with Prometheus and Grafana

Scraping Managed Cloud Metrics with Prometheus and Grafana This page shows how to set up Prometheus and Grafana to monitor Qdrant databases running in Qdrant Managed Cloud. Prerequisites You need a Kubernetes cluster running where you want to deploy your monitoring stack, and a Qdrant database created in Qdrant Managed Cloud. You should also have kubectl and helm configured to interact with your cluster. Cluster metrics versus node metrics Qdrant Managed Cloud exposes two Prometheus-compatible endpoints, and they answer different questions:

QdrantAI 工具英文

Cluster Alerts

Cluster Alerts The account owner receives these alerts by email. Each row gives the condition that triggers the alert, what it means for your cluster, and how to resolve it. For the metrics behind them, see Monitor Clusters . To read firing and resolved alerts in your own automation, run qcloud cluster alert list <CLUSTER_ID> in the Cloud CLI or call the ListClusterAlerts method of the Cloud API . Capacity Alert Triggers when What it means How to resolve Memory Overutilized Memory use is above 80% of the allocation for more than 5 minutes If usage keeps growing past the allocation, the cluster fails from resource pressure and you see disruption Scale vertically for more capacity per node, sc…

QdrantAI 工具英文