NewYour coding agent can read the release notes before it upgrades.Set up the MCP server →
PyPI
PyPI, #1030 most downloaded on PyPI
Client library for the Qdrant vector search engine
Latest stable
1.20.0
Last release 1 months ago
22 Oct 2026
Ships fairly regularly
a new release about every 6 weeks
Most releases are documented
notes for 40 of the last 60 stable releases
1 version withdrawn
withdrawn after publishing
6 years old
127 releases · first in 2021
Drag across the ruler to compare, or tap a release to open it.
BM25 is now supported through TextQuery: set scoring on a text index and rank points by how well that field matches the query text. Local mode support
BM25 is now supported through TextQuery: set scoring on a text index and rank points by how well that field matches the query text. Local mode supports it only for text indexes with the default tokenization. Formulas get the max, min, and acosh expressions.
The new MatchSubstring condition finds keyword values that contain a given string, using a keyword index. Payload indexes can be nested, indexing every object of an array as its own entry: with nested: offers, a text index on offers[].description matches and BM25-scores each offer separately.
With pydantic v2, parsing REST responses with vectors is about 2x quicker, and about 3x with the orjson extra (pip install 'qdrant-client[orjson]'). On protobuf 6.30 and later, gRPC writes with payloads are up to about 3x quicker.
PointStruct, PointVectors, and Batch are much quicker to build from numpy vectors, and parallel uploads of numpy arrays over REST barely add to the main process's memory.
spawn instead of forkserver, which crashed on Python 3.15. Pass method="forkserver" to upload_points or upload_collection to keep the old start method (#1551, #1568)./, #, %, or ? address the right collection, and names that are exactly . or .. raise ValueError (#1291).ValueError when ids or payload is shorter than the vectors, when parallel is negative, or when auth_token_provider is async. parallel=0 uses all CPU cores (#1578, #1531, #1520).TextQuery) with scoring on text indexes, MatchSubstring, max, min, and acosh formula expressions, nested payload indexes, id_tracker collection config, turbo8 and turbo16 datatypes, and the code tokenizer; migrate keeps the payload and id_tracker collection params by @joeinorjson extra for faster decoding by @joeinPointStruct, PointVectors, and Batch by @joeinwith_ to get_optimizations to request queued, completed, and idle_segments, and request completed when completed_limit is set by @joeinword in create_payload_index, and convert WriteOrdering in create_payload_index and delete_payload_index over gRPC by @feiiiiii5, @astroknot-sheep, @joeinNone instead of {} for absent collection metadata over gRPC by @feiiiiii5, @joeinrequest_sync from ApiClient and AsyncApiClient by @joeinembed_models_batch, and pass local_inference_batch_size to query embedding in local inference by @moneebullah25, @RAMZI0TO99, @joeinBatch columns of different lengths in local mode and over gRPC, and ids or payload shorter than vectors in upload_collection by @fhgffy, @joeinpoints lists in batch payload and vector updates over gRPC instead of falling back to filter by @joeinmigrate, validate its batch_size before touching the destination, and drop its own retry loop, since upload_points already retries by @nicoloangileri, @shayanxcyber, @09Catho, @joeinspawn instead of forkserver for parallel uploads and local inference on macOS by @joeinmax_retries in uploads as retries after the first attempt, so the default 3 makes up to 4 attempts, use all cores for parallel=0, and reject negative values of both by @feiiiiii5, @joeinauth_token_provider and grpc_compression to gRPC uploads, and reject async token providers in uploads by @breken-ai, @nandanadileep, @joeinUUID, Decimal, sets, bytes, and so on) over gRPC like REST does by @IncognitoQuack, @joein. and .. by @vaibzde, @vaibhavaol, @joeinmeta.json atomically by @adityaanikam, @joeinTextQuery (BM25 with the default tokenization only), MatchSubstring, and the new formula expressions; validate vector names and sizes like the server, reject unknown vector names in update_collection, and raise on payload index operations for a missing collection by @joeinlin_decay midpoints 0.0 and 1.0, reject invalid slice conditions, and ignore extra fields on geo points in formula queries by @aniketkrs, @digit50, @kabishou11, @joeinuint32 range, nested dense vectors, and payloads that can't be converted to JSON before writing by @Taranum01, @ktz03, @digit50, @joeingroup_by values independent of the hash seed by @pacocartones, @joeinlookup_from in prefetches, and honor search_params (IDF corpus) by @galleonli, @nandanadileep, @j45856021-dev, @joeinwith_payload keys as JSON paths, including quoted keys, and treat an empty with_payload list as selecting no fields by @Harsh23Kashyap, @feiiiiii5, @joeinget_collection results, and keep unset fields on partial sparse vector config updates by @lllleolin-max, @wcsnbilil, @joeinscore_threshold, prefetch, and fusion queries by @kabishou11, @joeinnull when merging payload and collection metadata by @Yuilona, @joein[] and [i] one level in payload paths and group_by, like the server by @IncognitoQuack, @joeinupdate_mode in batch_update_points and on deleted points, and reject empty selectors and empty point lists in update operations by @truecallerabreham, @xvimelese-ctrl, @Harsh23Kashyap, @joeincreate_vector_name and delete_vector_name with the server, and reject an empty sparse vector name in create_collection by @Marcuswang0824, @joeinFalse when deleting an absent collection by @Harsh23Kashyap, @joeinstorage.dbm storage (qdrant-client < 1.1.7) by @joeinThanks to everyone who contributed to the current release! @feiiiiii5 @astroknot-sheep @moneebullah25 @RAMZI0TO99 @fhgffy @nicoloangileri @shayanxcyber @09Catho @breken-ai @nandanadileep @IncognitoQuack @vaibzde @vaibhavaol @adityaanikam @aniketkrs @digit50 @kabishou11 @Taranum01 @ktz03 @pacocartones @Harsh23Kashyap @galleonli @j45856021-dev @lllleolin-max @wcsnbilil @Yuilona @truecallerabreham @xvimelese-ctrl @Marcuswang0824 @winklemad @tjwjdrok @joein
Your coding agent can read these notes before it upgrades. Set up the MCP server →
Thanks to everyone who contributed to the current release! @HaokaiDing @pentaoa @betacatsling @bijay-odyssey @2sumtech @Nimra3261 @feiiiiii5 @nazsats
MatchAny in local mode, matching the server by @joeinstr and non-list Sequence with_payload/with_vectors selectors in local mode by @betacatsling, @joeinPowExpression by @bijay-odyssey, @joeinstart_from with a date object by @joeinorder_by by @joein, @Nimra3261order_by to match server semantics by @feiiiiii5, @joeinGroupId semantics by @feiiiiii5, @joeinUpdateResult.operation_id and CollectionInfo.points_count as None over gRPC by @2sumtech, @joeindelete_payload by @winklemad, @joeinis_integer usage in the local mode formula API by @joeinIsNull in local mode by @generall, @joeinscore_threshold strictly to match server semantics by @linhongyu510, @joeinTypeError when sorting mixed-type point ids in the distance matrix by @winklemad, @joeinmin_should min_count values that the server rejects by @nazsats, @joeinUser-Agent override warning a stacklevel so it points at the caller by @Anai-Guocosine_similarity by @aravgarg28prefix=False round-trip behavior over gRPC by @2sumtechStartFrom instead of returning None by @Anai-Guoshould filters by @nazsats, @joeinFieldCondition.is_empty/is_null in local mode by @sainikhiljuluri, @joeinThanks to everyone who contributed to the current release! @HaokaiDing @pentaoa @betacatsling @bijay-odyssey @2sumtech @Nimra3261 @feiiiiii5 @nazsats @winklemad @generall @SiluPanda @linhongyu510 @aiedwardyi @Anai-Guo @aravgarg28 @Aryan-Pardeshi @shashvat-singham @adityat210 @nightcityblade @sainikhiljuluri @joein
Thanks to everyone who contributed to the current release! @nazsats @winklemad @ErenAta16 @MushiSenpai @charles-dyfis-net @aloktripathi1 @Goodnight77
payload storage configuration in create_collection (supersedes on_disk_payload) and more by @joeinsearch_params is ignored in local mode instead of silently dropping it by @MushiSenpai, @joeinvalues_count bounds must be satisfied by a single value in local mode by @nazsatslimit > 0 rules in local mode by @joeinconvert_points_update_operation by @ErenAta16set_payload on shared nested objects by @joein, @ErenAta16check_match() raises TypeError when MatchText applied to a non-string field by @aloktripathi1, @joeinadd and query methods have been removed, use FastEmbed with upsert/query_points instead by @joeinmax_resident_memory_percent has been deprecated by @joeinsearch, search_batch, search_groups, recommend, recommend_batch, recommend_groups and on_disk_payload have been deprecated in favour of query_points, query_batch_points, query_points_groups and the new payload config by @joeinParallelWorkerPool class by @Goodnight77Thanks to everyone who contributed to the current release! @nazsats @winklemad @ErenAta16 @MushiSenpai @charles-dyfis-net @aloktripathi1 @Goodnight77 @hassanzafarr @estebany-qd @joein
Thanks to everyone who contributed to the current release! @jnMetaCode @Anush008 @joein
#1174 - python-client docs has been deprecated in favour of qdrant.tech documentation by @joein
#1169 - do not modify date filters in local mode by @jnMetaCode
#1168 - run server version check in a thread to avoid blocking async client by @joein
#1157 - fix type hint error in grpc_uploader with older versions of protobuf by @joein
Thanks to everyone who contributed to the current release! @jnMetaCode @Anush008 @joein
Thanks to everyone who contributed to the current release! @jnMetaCode @Anush008 @joein
Thanks to everyone who contributed to the current release! @jnMetaCode @Anush008 @joein
Thanks to everyone who contributed to the current release! @cbcoutinho @generall @coszio @joein
Thanks to everyone who contributed to the current release! @cbcoutinho @generall @coszio @joein
Thanks to everyone who contributed to the current release! @holyMolyTolli @joein
lookup_from correctly in query_points_groups by @joeinThanks to everyone who contributed to the current release! @holyMolyTolli @joein
# Change Log ## Features 🪐 * #1116 - implement grpc version of cluster_collection_update by @joein * #1123 - expose the remaining cluster methods: rem
cluster_collection_update by @joeinremove_peer, collection_cluster_info, recover_current_peer, cluster_status by @joeincommon.proto to qdrant_common.proto to avoid conflicts with other libraries by @joeinThanks to everyone who contributed to the release!
TextAny filter, metadata to collections, add parametrized rrf query, ACORN (qdrant changelog) by @joeinadd, query, query_batch methods deprecated in favour of upsert and query_points by @joeinsearch, recommend, discovery, upload_records, search_batch, recommend_batch, discovery_batch, rest, init_from in create_collection, all lock-related methods by @joeinThanks to everyone who contributed to the release!
@ivandasch @JojiiOfficial @joein @generall @tbung @coszio
Thanks to everyone who contributed to the release! @coszio @joein
Thanks to everyone who contributed to the release! @coszio @joein
Thanks to everyone who contributed to the release! @kurtqq @coszio @generall @joein
Thanks to everyone who contributed to the release! @kurtqq @coszio @generall @joein
# Change Log ## Features 🎸 * #994 - add convenience methods like list_text_models, list_image_models, etc to fastembed mixin by @joein * #1024 - do no
list_text_models, list_image_models, etc to fastembed mixin by @joeinThanks to everyone who contributed to this release! @coszio @joein
Thanks to everyone who contributed to this release! @coszio @joein
add and query methods are now deprecated in favour of local inference usage
sum_score by @joeinadd and query methods are now deprecated in favour of local inference usageThanks to everyone who contributed to the current release! @edjinks @abionics @tellet-q @I8dNLo @hh-space-invader @agourlay @coszio @generall @joein
Nothing published for this version
Nothing published for this version
Nothing published for this version
Nothing published for this version
Thanks to everyone who contributed to the current release! @tellet-q @pavelm10 @hh-space-invader @I8dNLo @rovermicrover @generall @joein
Thanks to everyone who contributed to the current release! @tellet-q @pavelm10 @hh-space-invader @I8dNLo @rovermicrover @generall @joein
# Change Log ## Features - #870 updated fastembed to 0.5.0 - #870 removed support for python3.8
Thanks to everyone, who've contributed to this release! @joein @hh-space-invader
Thanks to everyone, who've contributed to this release! @joein @hh-space-invader
Thanks to everyone, who've contributed to this release!
Thanks to everyone, who've contributed to this release!
@agourlay @coszio @hh-space-invader @generall @joein
Thanks to everyone, who've contributed to this release! @VanaMartin @icfly2 @joein
Thanks to everyone, who've contributed to this release! @VanaMartin @icfly2 @joein
Thanks to everyone, who've contributed to this release! @paul2302 @hh-space-invader @joein
with_vectors in local mode when retrieving by default vector name by @joeinThanks to everyone, who've contributed to this release! @paul2302 @hh-space-invader @joein
Thanks to everyone, who've contributed to this release! @hh-space-invader @joein
fastembed to 0.3.6modifier.None conversion by @joeinrest property in QdrantRemote in favour of http by @hh-space-invaderQdrantClient in favour of AsyncQdrantClient by @hh-space-invaderThanks to everyone, who've contributed to this release! @hh-space-invader @joein
Thanks to everyone, who've contributed to this release! @agourlay @coszio @Anush008 @generall @hh-space-invader @joein
add and query methods by @joeinmigrate by @hh-space-invaderquery_points by @joeinThanks to everyone, who've contributed to this release! @agourlay @coszio @Anush008 @generall @hh-space-invader @joein
https://github.com/qdrant/qdrant-client/pull/674 - fix gRPC sparse vector params conversion
query_points for flat structuresSupport search, recommendations, discovery, ordering by payload in a single request
Universal query API - [Docs]
Support for multivectors - [Docs]
Support for float16 datatype for dense vectors - [Docs]
Support for float16 and uint8 datatype for sparse vectors index - [Docs]
Support for Inverse Document Frequency (IDF) in sparse vectros - [Docs]
Thanks to everyone, who've contributed to this release! @agourlay @coszio @tellet-q @generall @joein
Thanks to everyone who contributed to the current release: @generall @coszio @joein
RecommendStrategy with grpc by @coszioThanks to everyone who contributed to the current release: @generall @coszio @joein
Thanks to everyone who contributed to the current release: @agourlay @Anush008
Thanks to everyone who contributed to the current release: @agourlay @Anush008
Thanks everyone who contributed to the release @coszio @yasyf @Apmats @skvark @generall @joein @agourlay
Thanks everyone who contributed to the release @coszio @yasyf @Apmats @skvark @generall @joein @agourlay
remove redundant pytest import which prevented qdrant-client import without pytest being installed by @joein
pytest import which prevented qdrant-client import without pytest being installed by @joeinupdate fastembed integration, incorporate hybrid search into add method #553 by @generall @joein
add method #553 by @generall @joeinupload_collection and upload_points #531 by @almostimplementedThanks everyone who contributed to the release @almostimplemented @generall @joein
remove deprecated field conversions in grpc by @coszio
order_by functionality to scroll #491 by @coszioThanks everyone who contributed to the release @coszio @geetu040 @FranckZibi @NirantK @Anush008 @joein
fix search_batch with sparse vectors in grpc #484 by @agourlay
search_batch with sparse vectors in grpc #484 by @agourlayclose method signature between QdrantClient and QdrantRemote #477 by @joeinfix parallel and bugs with implicit ids in upload_collection, upload_records, upload_points #460
Bugfix release 💊
upload_collection, upload_records, upload_points #460grpc_options with https in sync mode, and with http+api-key in async mode #467datetime and uuid #462upload_records has been deprecated in favour of upload_points #447
upload_collection and upload_points #396urllib3 v2 #418fastembed models #412fastembed configuration #416httpx data with content #426local mode #397local mode #398 #421local mode #425wait parameter to upload_records in migrate to avoid false-positive asserts #428migrate from collections with custom shard keys #447timeout param types in QdrantClient and its methodsupload_records has been deprecated in favour of upload_points #447Thanks to the devs who contributed to this release
@generall @shivas1516 @praveen-palanisamy @joein
Add Discovery API (docs) - https://github.com/qdrant/qdrant-client/pull/368 by @coszio
wait parameter to snapshots API by @joeinNothing published for this version
Nothing published for this version
Nothing published for this version
Nothing published for this version
Nothing published for this version
Nothing published for this version
Nothing published for this version
Missing fix form previous release: https://github.com/qdrant/qdrant-client/pull/338 - add missing parameters in recommend batch API
Proper support for Async client - https://github.com/qdrant/qdrant-client/pull/319
Proper support for Async client - https://github.com/qdrant/qdrant-client/pull/319
Improvements of fastembed integration:
add API, useful for lazy reading of large datasetAsync Qdrant client usage example:
from qdrant_client import AsyncQdrantClient, models
import numpy as np
import asyncio
async def main():
client = AsyncQdrantClient(url="http://localhost:6333")
res = await client.search(
collection_name="my_collection",
query_vector=np.random.rand(10).tolist(), # type: ignore
limit=10,
)
print(res)
asyncio.run(main())
Support for Qdrant 1.6.x features
wait param to the upload records methodinit_from parameterNothing published for this version
Nothing published for this version
Nothing published for this version
Nothing published for this version
https://github.com/qdrant/qdrant-client/pull/280, https://github.com/qdrant/qdrant-client/commit/6f8c5172d77a1775474d6ae145e7552543803806 - Compatibil
Use fastembed library to easily encode & index documents into qdrant
pip install fastembed qdrant-client
from qdrant_client import QdrantClient
# Initialize the client
client = QdrantClient(":memory:") # or QdrantClient(path="path/to/db")
# Prepare your documents, metadata, and IDs
docs = ["Qdrant has Langchain integrations", "Qdrant also has Llama Index integrations"]
metadata = [
{"source": "Langchain-docs"},
{"source": "Linkedin-docs"},
]
ids = [42, 2]
# Use the new add method
client.add(
collection_name="demo_collection",
documents=docs,
metadata=metadata,
ids=ids
)
search_result = client.query(
collection_name="demo_collection",
query_text="This is a query document"
)
print(search_result)
More in Notebook
Nothing published for this version
Forbid extra fields in pydantic models #222
Dict[str, np.ndarray] in uploading collection in local mode #183Nothing published for this version
Nothing published for this version
Support for Qdrant v1.2.0 features:
Nothing published for this version
Nothing published for this version
One column per quarter.