Change8

Milvus

AI & LLMs

Python SDK for Milvus Vector Database

Latest: v3.0.125 releases6 breaking changes7 common errorsView on GitHub

Release History

View all versions →
v3.0.1Breaking5 fixes27 features
Jul 29, 2026

PyMilvus 3.0.1 introduces new APIs for data salvage, role management, and search aggregation, alongside significant improvements in schema evolution, type handling, and connection recovery. This release also drops support for Python 3.8.

v2.6.17Breaking7 fixes3 features
Jul 17, 2026

This release synchronizes `milvus-proto` with Milvus v2.6.20, introducing support for inclusive `DumpMessages` start options and fixing several bugs related to array and struct fields, schema management, and data insertion.

v2.6.164 fixes4 features
Jun 25, 2026

This release introduces element-level search, data salvage support via `dump_messages`, and enhanced role/user description support in auth APIs. It also includes significant internal improvements to connection handling and logical type model migration.

v2.6.154 features
Jun 1, 2026

This release introduces new methods for checking replication state and refactors type handling within the client, alongside modernizing development workflows using uv.

v2.6.141 fix2 features
May 15, 2026

This release introduces new features like an internal data type metadata registry and array partial update helpers, alongside a bug fix for oversized compaction targets.

v3.0.0
May 7, 2026
v2.6.1310 fixes4 features
May 7, 2026

This release focuses on bug fixes, including parameter validation and header additions for import APIs, alongside new features like cloud import enhancements and replication configuration access. Deprecation warnings for the ORM API have also been introduced.

v3.0.1.dev0
May 7, 2026

Added deprecation warnings for ORM-style public APIs backed by the connections registry. This change targets specific ORM entry points while leaving query and search iterator APIs unaffected for now.

v2.6.126 fixes
Apr 9, 2026

This release primarily focuses on bug fixes, including correcting default ports for HTTPS connections and improving deserialization logic. It also includes several internal refactorings and CI updates.

v2.6.1111 fixes1 feature
Mar 27, 2026

This release introduces support for Python 3.14 and resolves numerous bugs related to connection handling, search operations, and dependency compatibility, including pinning setuptools_scm.

v2.6.106 fixes2 features
Mar 13, 2026

This release introduces global cluster client support and a new ConnectionManager for MilvusClient, alongside numerous bug fixes related to primary keys, vector handling, and connection timeouts.

v2.6.92 fixes2 features
Feb 10, 2026

This release enhances gRPC configuration and adds support for warmup parameters in FieldSchema. It also fixes critical bugs related to database context handling in SearchIteratorV2.

v2.6.81 fix1 feature
Jan 29, 2026

This release focuses on bug fixes, performance optimizations for search results, and improved gRPC error diagnostics. It also removes the dependency on grpcio-testing.

v2.6.711 fixes2 features
Jan 26, 2026

This release introduces support for semantic-highlighter and async list_persistent_segments, alongside numerous bug fixes addressing data type handling, memory usage, and database context management.

v2.6.67 fixes2 features
Dec 30, 2025

This release introduces a new benchmarking framework and Function SDK support, alongside numerous bug fixes related to boolean serialization, JSON parsing, and iterator handling.

v2.6.5Breaking4 fixes3 features
Dec 5, 2025

PyMilvus 2.6.5 introduces significant enhancements to AsyncMilvusClient, adds highlighter support for search, and fixes several bugs related to index creation and data handling. This release also includes a breaking change renaming 'Stage' to 'Volume'.

v2.5.18Breaking
Dec 2, 2025

This release introduces a breaking change by renaming 'Stage' to 'Volume' to improve clarity and consistency within the library.

v2.6.49 fixes7 features
Nov 26, 2025

This release introduces support for the TIMESTAMPTZ data type and significantly enhances BulkWriter capabilities for STRUCT and Geometry types. Numerous bug fixes address issues in connection handling, MilvusClient argument passing, and async operations.

v2.5.171 fix3 features
Nov 10, 2025

This release introduces enhancements for primary key insertion with autoID, support for serialized JSON inserts, and geometry support in the bulk writer. It also includes a fix for GCP stage configuration.

v2.6.3Breaking1 fix6 features
Oct 31, 2025

This release introduces major features including support for Array of Structs and Geometry data types, alongside manual L0 compaction control. A critical breaking change mandates JSON serialization for string data during insertion.

v2.5.163 fixes2 features
Sep 19, 2025

This release introduces support for the geo data type in core operations and includes several minor bug fixes and documentation improvements related to connection types and bulk import.

v2.6.22 fixes2 features
Sep 18, 2025

PyMilvus v2.6.2 introduces support for grouping by JSON fields and fixes several issues related to partial upsert operations. It also includes enhancements for OSS bucket connections.

v2.6.13 fixes2 features
Aug 29, 2025

This patch release (v2.6.1) introduces new features like enhanced bulk import capabilities and stage management support, alongside several bug fixes and performance enhancements like adding unixmsec to RPC calls.

v2.5.152 fixes2 features
Aug 21, 2025

This release introduces support for stageManager/stageFileManager and enhances bulkImport functionality. It also resolves a bug related to returning primary keys during upsert operations.

v2.6.0Breaking39 features
Aug 6, 2025

This release introduces comprehensive asynchronous client support via `AsyncMilvusClient` and significantly expands the API surface of `MilvusClient` with numerous database, partition, index, and search management features. The long-deprecated `Milvus` class has been removed.

Common Errors

MilvusException4 reports

"MilvusException" usually arises from invalid input parameters or schema definitions during operations like index creation, data insertion, or query execution. To fix it, carefully validate all input data types and values against the Milvus schema, and thoroughly inspect the error message for specific constraints violated, adjusting the code accordingly before retrying the operation. Consult Milvus documentation for appropriate data type and range specifications.

ConnectionConfigException3 reports

The "ConnectionConfigException" in Milvus usually arises from incorrect or missing connection parameters such as host, port, secure, or URI when initializing the Milvus client; or shared library loading errors, like missing `libknowhere.so`. Double-check your connection parameters in your MilvusClient instantiation and ensure that the necessary shared libraries like `libknowhere.so` are installed and accessible in your environment or that Milvus dependencies are correctly set up using `pip install pymilvus[default]`. If using Milvus Lite, verify you're using compatible pymilvus version per the docs.

DataNotMatchException3 reports

DataNotMatchException in Milvus usually arises from a discrepancy between the data provided during insertion/upsertion and the schema defined for the collection (e.g., incorrect data type, missing required fields, or extra unexpected fields). To resolve this, carefully review your data against the collection schema – ensure data types align, all non-nullable fields have values (or default values are set), and no unexpected fields are being inserted.

ParamError3 reports
ConnectionNotExistException2 reports

The "ConnectionNotExistException" often arises from attempting to use a Milvus connection after it has been explicitly closed or garbage collected while other clients are still relying on it. Ensure each MilvusClient uses its own dedicated connection or manages a connection pool properly. Avoid prematurely closing the underlying connection object if other MilvusClient instances might still need it.

SchemaNotReadyException2 reports

The SchemaNotReadyException often arises when Milvus attempts to access collection metadata before it's fully initialized, especially after collection creation or server restarts. To fix this, ensure that the collection creation process is complete before performing further operations and introduce a retry mechanism with a short delay to allow the schema to become available. You can also explicitly call `collection.load()` before other operations.

Related AI & LLMs Packages

Subscribe to Updates

Get notified when new versions are released

RSS Feed