Qdrant 1.16 on Ubuntu 22.04 by cloudimg. Open source vector similarity search engine written in Rust, purpose built for AI and ML workloads: semantic search, retrieval augmented generation, recommendation systems, and anomaly detection. REST and gRPC APIs, built in web dashboard, api key authentication. 24/7 expert support.
Qdrant 1.16 Community on Ubuntu 22.04 LTS, packaged and supported by cloudimg as an Azure Marketplace virtual machine image.
Qdrant is an open source vector similarity search engine written in Rust, licensed under Apache 2.0. It stores high dimensional vectors with JSON payloads and performs approximate nearest neighbour search with filtering, purpose built for retrieval augmented generation pipelines, semantic search, recommendation systems, image similarity, and anomaly detection. Qdrant 1.16 includes the improved HNSW index, sparse vector support, and hybrid search combining dense and sparse vectors in a single query. The server is a single statically compiled Rust binary with no Java, no Python, and no other embedded runtime.
Why Qdrant for AI and ML workloads
* Purpose built for retrieval augmented generation. The index is HNSW with filtering, so metadata conditions are evaluated during graph traversal rather than as a post filter, which keeps recall high on selective queries
* Dense plus sparse vectors in one query. Hybrid search combines semantic similarity with BM25 style lexical matching without stitching two systems together
* REST and gRPC on the same server. Curl for debugging, gRPC for production throughput from Python, Go, Rust, Java, and JavaScript client SDKs
* Built in web dashboard. Browse collections, run searches, and inspect cluster state at http colon slash slash the VM IP colon 6333 slash dashboard with no separate service to install
* Works with every embedding provider. OpenAI, Cohere, HuggingFace, and every open source sentence transformer produce vectors that Qdrant indexes identically
Pre configured for production posture on day one
* Qdrant 1.16 installed from the official GitHub release tarball (statically linked musl binary), with future version upgrades pulled from the same source
* HTTP API on port 6333 and gRPC API on port 6334, both bound to 0.0.0.0 so application servers and client SDKs on the same virtual network can reach them
* api_key authentication enforced on every HTTP and gRPC call
* Unique api_key generated on the very first boot of every deployed virtual machine, written to a root only credentials file
* Payloads stored on disk by default so RAM stays reserved for the HNSW graph and vector data
* systemd supervision with a firstboot oneshot that rotates the api_key on customer first boot then disables itself
* Convenience wrappers at /usr/local/sbin for start, stop, and environment setup
What this image is
* A production ready single node Qdrant server ready to ingest vectors and serve similarity searches on day one
* Pre installed from the official upstream release, with systemd units and a YAML config cloudimg owns for configuration management
* Hardened by cloudimg: credentials never baked into the image, SSH keys truncated not deleted to preserve Azure's deployment injection, build time admin user removed to pass Azure Marketplace certification
What this image is not
* Not a replicated Qdrant cluster. The image is single node by design; Qdrant supports clustering which can be configured after deployment if required
* Not TLS encrypted out of the box. See the user guide for the recommended nginx or Caddy reverse proxy pattern
* Not the Qdrant Cloud managed service. This is the Apache 2.0 licensed open source server, self managed on your own Azure subscription
Components
* Qdrant 1.16, pinned at build time, installed from github.com/qdrant/qdrant/releases
* Binary: /usr/local/bin/qdrant (single statically compiled Rust binary)
* Data Directory: /var/lib/qdrant (storage path, snapshots path, HNSW indexes, static dashboard assets)
* Config: /etc/qdrant/config.yaml (YAML configuration: ports, api_key, storage paths, on disk payload)
* Default OS User: your chosen admin user
* Qdrant OS User: qdrant (owns the data directory and runs the server)
* Recommended Size: Standard_D4s_v3 (4 vCPU, 16 GB RAM) because vector search is memory intensive and the HNSW index lives in RAM; scale to D8s_v3 or larger for collections beyond a few million vectors
* VM Generation: Hyper V Gen2 with UEFI boot
* Filesystem: Default Ubuntu gallery LVM layout
Security
* Latest CVE patches applied at build time, with a first boot apt-get update and upgrade for any patches published between build and deployment
* SSH hardened with key based authentication
* No Java, no Python, no embedded interpreter: the runtime CVE surface is the Rust binary plus the base Ubuntu image
* api_key authentication enforced out of the box; the api_key is generated on first boot and is unique per virtual machine
* Recommended to restrict ports 6333 and 6334 to application subnets in the NSG and terminate TLS with a reverse proxy before production
Support
cloudimg provides 24/7/365 expert technical support. Guaranteed response within 24 hours, one hour average for critical issues. Contact support@cloudimg.co.uk.
Visit www.cloudimg.co.uk/guides/qdrant-1-16-on-ubuntu-22-04-azure for the full user guide.
Qdrant is a trademark of Qdrant Solutions GmbH. This image uses the Apache 2.0 licensed open source edition. This image is a repackaged upstream distribution provided by cloudimg. Additional charges apply for build, maintenance, and 24/7 support.