A search application may need to find related text or images even when they do not share exact words. Qdrant stores vectors, the numerical representations produced by a model, and searches for nearby vectors while supporting associated data filters. It provides a dedicated service for this similarity-search step. Your application still needs to produce the vectors and decide whether the results are useful. Self-hosting also makes storage, authentication and recovery your responsibility, with different automation across its deployment offerings.
Chart ownership
The Qdrant project publishes the qdrant/qdrant chart through its Helm repository. Upstream explicitly limits this chart to community support. It also distinguishes the chart from its managed cloud and enterprise operator offerings: automated disaster recovery, resharding and protection during downscaling are not equivalent across those options.
Before adoption
Choose persistent storage, authentication and network exposure deliberately. Size the deployment using representative vectors, payload filters and query concurrency. Pin the chart and application image, and rehearse node replacement, snapshot restoration and scaling down before storing irreplaceable data. A successful Helm upgrade does not demonstrate that every collection remains available or recoverable.
Sources & further reading
Spotted something that needs another look?
Help improve this page →