Qdrant

Qdrant is an open-source vector similarity search engine written in Rust, first released in 2021 by Berlin-based Qdrant Solutions GmbH. The core engine (github.com/qdrant/qdrant) ships under Apache License 2.0 with no source-available or field-of-use carve-out — the GitHub API confirms the license as plain apache-2.0 as of today. It has 34,689 GitHub stars and 2,684 forks (checked 2026-09-20). Deployment options are self-hosted (Docker, binaries, Kubernetes via Helm), Qdrant Cloud (managed), Hybrid Cloud (managed control plane, your infrastructure), and Private Cloud.

Latest release is v1.19.1, published 2026-09-04 (github.com/qdrant/qdrant/releases). The repository was pushed to as recently as 2026-09-19, and pull requests continue to merge weekly. This is a live project, not one coasting on stars.

Qdrant's core differentiator is combining vector similarity with metadata filtering inside the same HNSW index (via its ACORN-style filterable HNSW, documented at qdrant.tech/articles/filtered-vector-search-acorn), rather than filtering before or after the ANN search as a bolt-on step — useful for RAG and agent workloads where most queries carry a filter (tenant ID, date range, permissions). Independent benchmarking from Tiger Data (formerly Timescale), comparing Qdrant 1.13.4 against Postgres+pgvector+pgvectorscale on 50M 768-dim embeddings, found Qdrant has meaningfully better tail latency: 39% lower p95 and 48% lower p99 query latency at 99% recall, and faster index builds (3.3 hours vs. 11.1 hours for the same dataset) (tigerdata.com/blog/pgvector-vs-qdrant, published 2025-04-29). It's built in Rust with no garbage-collector pauses, which is the architectural reason cited for that latency consistency.

The same Tiger Data benchmark found Postgres with pgvector/pgvectorscale delivers 11.4x higher query throughput than Qdrant at 99% recall (471.57 QPS vs. 41.47 QPS) and 4.4x higher at 90% recall, on identical hardware and the same 50M-vector dataset. Tiger Data attributes this to read contention under concurrent load, which it calls a sign of Qdrant's relative immaturity compared to Postgres's decades of concurrency tuning. If your bottleneck is queries-per-second rather than tail latency, this is the number to weigh — and it comes from a vendor with its own dog in the fight (Tiger Data sells the Postgres alternative), so treat the throughput gap as their finding, corroborated by their own published methodology and raw numbers, not as neutral.

Qdrant Cloud publishes a real free tier: a permanent single-node cluster with 0.5 vCPU / 1 GB RAM / 4 GB disk, no credit card required (qdrant.tech/pricing, checked 2026-09-20). Above that, Standard Tier is usage-based, billed hourly by vCPU, RAM, disk, and backup storage consumed, with published per-resource rates surfaced through Qdrant's cloud pricing calculator (cloud.qdrant.io/calculator) rather than a flat monthly card — a sample calculation for a 100M-vector, 64-dim, 5-node cluster in AWS eu-central-1 returns $0.5632/hour (~$412/month). Premium Tier (SSO, private VPC links, 99.9% uptime SLA) requires a minimum spend and a sales conversation; Hybrid Cloud and Private Cloud are quote-only. Self-hosting the OSS engine costs nothing beyond your own infrastructure.

Qdrant Solutions GmbH has raised a total of roughly $88M: a $28M Series A in January 2024 led by Spark Capital (qdrant.tech/blog/series-a-funding-round), and a $50M Series B announced 2026-03-12, led by AVP with participation from Bosch Ventures, Unusual Ventures, Spark Capital, and 42CAP (qdrant.tech/blog/series-b-announcement; businesswire.com/news/home/20260312313902). The company says it has surpassed 250 million downloads across its packages and names Canva, Bazaarvoice, HubSpot, Roche, Bosch, and OpenTable as production users in the Series B announcement — these are vendor claims, not independently confirmed customer counts. Employee-count estimates from third-party data providers (PitchBook, Tracxn) put headcount around 145-163 as of 2026, which is directionally consistent across sources but not a figure Qdrant itself publishes.

Qdrant lists SOC 2 Type 2 and HIPAA certifications with reports available through a public Drata-hosted Trust Center (app.drata.com/trust/9cbbb75b-0c38-11ee-865f-029d78a187d9), and offers a GDPR Data Processing Agreement (qdrant.tech/security). Separately: GitHub's security advisory database lists one high-severity (CVSS 8.5) vulnerability, CVE-2026-25628, disclosed 2026-02-05 — an authenticated, low-privilege user could abuse the /logger endpoint to overwrite server config and escalate to remote code execution on self-hosted instances. Qdrant's own advisory states the flaw does not affect Qdrant Cloud, and it was patched in v1.15.6; current release is v1.19.1, well past the fix. Anyone running an older self-hosted version should upgrade.

G2 lists Qdrant at 4.5 out of 5 stars from 12 reviews as of this check (g2.com/sellers/qdrant) — a small sample, worth noting rather than treating as a settled verdict. InfoWorld's 2024 hands-on review (infoworld.com/article/3477585) called it a highly flexible option for vector search after direct testing, independent of Qdrant's own marketing.