Qdrant

Summary

Added TurboQuant 4‑bit vector datatype, per‑query IDF for sparse search, and prefix‑match filter support, boosting vector storage efficiency and search relevance.

Summary

Fix query errors when using shard keys during resharding

Summary

Improved shard WAL recovery logging, cleared ID tracker cache after segment build, and avoided unnecessary payload index rebuilds.

Summary

Added io_uring support for quantized multi‑vector scoring and validated vector dimensions before WAL writes for async upserts.

Summary

Added TurboQuant quantization, named‑vector API, low‑memory mode with strict memory limits, and dynamic CPU pool for faster search; introduced snapshot‑based shard transfers, idempotent resharding, and removed RocksDB support improving s...

Summary

Performance and reliability upgrades: deferred point updates, non‑blocking Gridstore flushes, faster filtered search, improved WAL error handling, and reduced lock contention during shard transfers.

Summary

Added several new APIs (relevance feedback, detailed optimization progress, aggregated telemetry, unlimited update queue, weighted RRF, update mode, secondary API keys, dedicated metrics port, shard key listing) and configurable options ...

Summary

Added timeout handling for searches, point retrieval, and telemetry shard stats; introduced snapshot download duration and speed logging.

Summary

Improve request timeout handling for telemetry and metrics, and add a User-Agent header to HTTP requests sent by Qdrant server.

Summary

Batch query speed up to 3× by reading each point once; Gridstore migration improves storage performance and consistency.

Summary

Introduced numerous new features such as inline HNSW storage, ACORN search, ASCII folding, conditional updates, custom collection metadata, and a profiler for slow operations.

Summary

Added API validation, timeout parameter for peer removal, and strict mode configuration for payload indices; removed vector count field from collection info and stopped anonymizing peer IDs in telemetry

Summary

Reduced Docker image size by up to 40%, added SBOM and cosign signing for supply‑chain security.

Summary

Optimized dot‑product calculations on AVX CPUs for better performance

Summary

Add BM25 inference, adjustable log buffer size, and shard distribution matrix view in the Web UI

Summary

gRPC HealthCheck method now works without authentication, matching REST behavior

Summary

Added phrase matching, stop‑words, multilingual tokenizer (default) and snowball stemmer to full‑text index, plus asymmetric and 2 bit/1.5 bit binary quantization and Maximum Marginal Relevance.

Summary

Multiple performance enhancements across the system, including faster WAL shard transfers, GPU indexing speed-ups, improved payload index load times, optimized GridStore I/O, better optimizer scheduling, and faster HNSW construction.

Summary

Added server‑side score boosting with user‑defined formula and new sum‑scores recommendation strategy for relevance feedback.

Summary

Query API now reads vectors/payloads at shard level and defers reads, reducing internal traffic and boosting search speed across many segments

Summary

Added several performance improvements: split CPU budget into CPU/IO, enhanced payload indices for IsEmpty/IsNull, compressed ID tracker mappings, shard-level undersampling, approximate point counts for faster shard transfers, and GPU HN...

Summary

Added strict‑mode max points limit per collection

Summary

Enable consensus compaction by default and improve Raft loop, accelerating peer join and recovery

Summary

Added GPU support for devices lacking half‑float precision, falling back to full‑float operations.

Summary

Optimized segment merging for better performance and lower memory usage.

Summary

Added GPU support for HNSW indexing, runtime resharding in the cloud offering, strict mode for collections, and a "has vector" filter condition.

Summary

Added multiple features: 64‑bit dimension indices for sparse vectors, Issues API support with limited API keys, JSON‑format logging, and a web UI bundled in the Debian package.

Summary

Implemented streaming of point retrieval, searches, and facet computation across shards and segments, and exposed shard keys in telemetry, improving performance and observability.

Summary

Added a unique identifier for collections to simplify consensus snapshots and sync consensus at the end of stream record transfers

Summary

Improved peer address handling: unchanged replacements now logged as warnings

Summary

Added multiple performance and memory optimizations (matrix API, HNSW search, RocksDB options reuse, JEMalloc, telemetry, etc.)

Summary

Prevent panic when geo radius is set to zero

Summary

Added Facets API for counting unique payload values and Distance Matrix API for many‑to‑many vector distance calculations.

Summary

Fix set payload with nested key not updating nested structures correctly

Summary

Parallelized point deduplication and optimized buffer reads, keyword‑index lookups and snapshot creation, greatly speeding up restarts and reducing disk usage

Summary

Fix matching payload filters on string indices

Summary

Fix potential panic during payload index building.

Summary

Added copy‑on‑write data modification, non‑blocking payload index building, telemetry peer listings, and MatchAny/Except conditions for UUID indexes.

Summary

Added GroupBy, UUID payload index, on-disk support for scalar indexes and random sampling to the Query API; introduced score‑fusion, data locality and prefetch performance improvements; made scroll and retrieve operations cancellable wit...

Summary

Fix multiple query API bugs, including double-applied offsets and incorrect score threshold handling for fusion requests

Summary

Universal query API adds hybrid search, multi‑stage queries, multivector and float16/uint8 support for dense and sparse vectors, plus IDF handling for sparse vectors.

Summary

Prevent embedding the Qdrant web UI in an iframe (security hardening)

Summary

Fix potential panic and in‑place update issues for sparse vectors during recovery

Summary

Added Pyroscope integration for on‑demand continuous profiling.

Summary

Fix potential segment loss when snapshotting with ongoing updates

Summary

Gracefully handle out‑of‑disk errors on insertions and speed up consensus convergence with batched updates.

Summary

Improved proxy segment performance for high-change workloads and accelerated loading of large memory‑mapped segments

Summary

Improved performance for numeric range filters and added antimeridian handling in geo bounding box filter

Summary

Implement shard diff transfer to speed up node recovery and improve reliability of shard transfers

Summary

Fixed persisted crash when using spaces in quoted JSON paths

Summary

Fix arbitrary path traversal vulnerability in the full snapshot REST API (security fix).

Summary

Fix propagation of max optimization threads on collection parameter updates

Summary

Fixed bug allowing payload paths that start with numbers

Summary

Added new APIs: order Scroll results, datetime payload index, collection existence check, minimum‑match payload filters, and nested payload updates.

Summary

Synchronize all nodes after creating a collection or shard key to prevent inconsistencies across the cluster.

Summary

Improved sparse vector plain-search speed.

Summary

Fix /readyz endpoint error in Kubernetes when insufficient nodes are available, preventing readiness failures.

Summary

Fix /readyz endpoint error when some nodes remain on version 1.16

Summary

Added support for sparse vectors (SPLADE/BM25), discovery search, user‑defined sharding, shard snapshot transfer, Manhattan distance metric, and read‑only API keys (security).

Summary

Fixed quantization rescoring issue when handling filtered requests

Summary

Introduced a recommendation engine, extended recommendation API to accept raw vectors, added polygon geo‑filtering, shard update parallelism tuning, and distributed search to improve cluster performance.

Summary

Added unprivileged Docker image for secure environments

Summary

Introduced major features: batch update endpoint, binary quantization, shard snapshot API, indexed_only search flag, gRPC reflection, Kubernetes health probes, stack trace API and optimizer telemetry.

Summary

- Skip outdated Raft heartbeats and use seeded randomness for ID tracking, improving consensus networking and search consistency

Summary

Added binary index for boolean payloads, runtime collection parameter updates, multilingual tokenizer, and UI features like snapshot management and payload editing

Summary

Added configurable temporary directory for snapshot recovery, logged the HTTP server port on startup, and improved scroll performance internals.

Summary

Added graceful shutdown via SIGTERM, enhanced search API validation, and used configured temporary directory for snapshot uploads.

Summary

Added a self‑hosted Web UI dashboard with collection viewer and interactive query editor, plus lookup grouping requests and experimental async I/O via io_uring.

Summary

Fixes the incorrect signal handler introduced by the previous bugfix.

Summary

Fix error handling and loading order for configuration files.

Summary

Added optional vectors, independent vector management APIs, dynamic mmap vector storage, product quantization, nested object filtering, and Group‑By API.

Summary

Fixed quantization to operate only on immutable segments

Summary

Fixed validation logic for the HNSW `m` parameter.

Summary

Added per‑vector HNSW and quantization configuration, TLS support for gRPC/REST and inter‑node communication, and multipart snapshot upload/recovery.

Summary

Added Scalar Quantization (up to 4x memory reduction, 2x speed boost) and new Match‑Any filtering, listener mode, and /metrics API in OpenMetric format.

Summary

Fixed unwanted extra connections issue.

Summary

Fix automatic thread number selection to prevent blocking the consensus execution thread in low‑CPU environments.

Summary

Fixed a critical UUID WAL recovery issue.

Summary

Added new collection alias, snapshot deletion, and collection initialization APIs; introduced optional read consistency and write ordering guarantees.

Summary

Improved mmap configuration, boosting search speed up to 6× on low‑RAM systems

Summary

Added cross‑collection vector lookup in recommendation API

Summary

Added a new payload API that overwrites all fields and supports filtered set/delete operations.

Summary

Added mmap support for HNSW graph storage and enabled gRPC compression, reducing RAM usage and network overhead.

Summary

Added API for recovering snapshots, with support for distributed mode

Summary

Fix replication factor and write consistency errors, and handle zero‑length vectors with cosine distance without normalization

Summary

Handle collections transition from single node to cluster

Summary

Added replication with sharding for high‑availability distributed deployments and a new admin API to disable writes; introduced an exact‑search flag and payload‑index reporting in the info API.

Summary

Fix propagation of payload index during optimization

Summary

Fix bug affecting Geo location payload validation

Summary

Fixed Raft consensus WAL conflict resolution and added split‑brain recovery capability.

Summary

Propagate proper error handling for partial update failures across shards.

Summary

Ensures backward compatibility across qdrant and client versions

Summary

Add multiple named vectors per point, batch search/recommendation APIs, and minimal full‑text filtering support

Summary

Fixed compatibility issue that caused a panic when optimizer max optimization threads was set to 0

Summary

Added dynamic cluster scaling, shard relocation, and peer removal APIs

Summary

Resolved backward compatibility issues with earlier 0.8.x releases

Summary

Added full storage snapshot that captures all collections and aliases in a single call

Summary

Added snapshot capability for safe backup and recovery of collections, with documentation

Summary

Added pagination support for search results.

Summary

Fixed a potential deadlock issue when reading collections.

Summary

Add new web‑based UI accessible at https://ui.qdrant.tech/ for direct interaction with local Qdrant instances.

Summary

Added experimental distributed deployment with Raft‑based cluster topology, sharding, distributed search/updates, and a new cluster status API

Summary

Adds support for arbitrary JSON payloads and makes the uploaded payload structure match retrieved results, deprecating the old schema.

Summary

Added full gRPC support for all endpoints, including docs and payload/vector return capability.

Summary

Improved error reporting for the collection info API

Summary

Added multi‑threaded optimizer and support for UUID point IDs; introduced individual endpoints for collection and point updates, payload/vector in search results, and filter‑based deletion/removal of payload

Summary

Fixed bug affecting support for multiple architectures

Summary

Fixed alias creation issue, proper return codes for non‑existent collections, storage corruption bug, and added a dedicated update thread with per‑record versioning and failure locks.

Summary

Fixed point retrieval from copy‑on‑write proxy segment.

Summary

Upgraded Actix to 4.0‑beta, enabling full upgrade to Tokio 1.0+.

Summary

Added API endpoint for scrolling over points with filters.

Summary

Added payload schema to the collection info API.

Summary

Fix HNSW index and search speed issues

Summary

Filtrable HNSW optimizer now auto‑builds HNSW indexes after a size threshold, considering indexed payloads

Summary

- Resolved filtering issue for indexed fields

Summary

Added ability to mark payload fields (number, keyword) as indexed; geo field indexing is a work in progress.