# pymilvus

> Python SDK for Milvus Vector Database

- **URL**: https://www.freshcrate.ai/projects/pymilvus
- **Author**: milvus-io
- **Category**: Databases
- **Latest version**: `v2.6.15` (2026-06-01)
- **License**: Apache-2.0
- **Source**: https://github.com/milvus-io/pymilvus
- **Language**: Python
- **GitHub**: 1,368 stars, 428 forks
- **Registry**: github
- **Tags**: `anns`, `faiss`, `faiss-vector-database`, `milvus`, `milvus-lite`, `milvus-sdk`, `python`, `python-sdk`

## Description

Python SDK for Milvus Vector Database

## Recent releases

| Version | Date | Urgency | Changes |
| --- | --- | --- | --- |
| `v2.6.15` | 2026-06-01 | High | ## What's Changed  * Add `get_replicate_info()` to `MilvusClient` and `AsyncMilvusClient` for reading replication checkpoint state. #3509 by @bigsheeper, backported in #3515. * Move Search logical-type handling to the shared type registry. #3482, backported in #3498. * Add the internal `TypeCapability` contract for shared type-specific operation bindings. #3531, backported in #3532. * Use the logical type registry in batch insert and upsert paths. #3517, backported in #3540. * Migrate deve |
| `v2.6.14` | 2026-05-15 | High | ## What's Changed * refactor: consolidate duplicated array element type mapping (#3401) by @pymilvus-bot in https://github.com/milvus-io/pymilvus/pull/3478 * feat(client): add internal data type metadata registry (#3475) by @pymilvus-bot in https://github.com/milvus-io/pymilvus/pull/3480 * feat: [2.6] add FieldOp helpers for array partial update operators by @weiliu1031 in https://github.com/milvus-io/pymilvus/pull/3492 * fix(client): reject oversized compaction target sizes (#3494) by @pymi |
| `v3.0.0` | 2026-05-07 | High | ## PyMilvus 3.0.0 for Milvus 3.0.0beta |
| `v2.6.12` | 2026-04-09 | High | ## What's Changed  ### Bug Fixes * **fix: default to port 443 for https:// URIs in MilvusClient** ([#3372](https://github.com/milvus-io/pymilvus/pull/3372), [#3374](https://github.com/milvus-io/pymilvus/pull/3374)) by @pymilvus-bot — reported by @XuanYang-cn in [#3369](https://github.com/milvus-io/pymilvus/issues/3369) * **fix: remove duplicate timeout kwarg in Collection.\_\_init\_\_** ([#3371](https://github.com/milvus-io/pymilvus/pull/3371), [#3378](https://github.com/milvus-io/pymilvus/p |
| `v2.6.11` | 2026-03-27 | Medium | ## What's Changed    ### Bug Fixes   * **fix: pass address to GrpcHandler in ConnectionManager strategies** — fixed Milvus Lite local path broken since 2.6.10 ([#3321](https://github.com/milvus-io/pymilvus/pull/3321),  [#3323](https://github.com/milvus-io/pymilvus/pull/3323)) by @XuanYang-cn   * **fix: prevent "Cannot invoke RPC on closed channel" after connection recovery** ([#3326](https://github.com/milvus-io/pymilvus/pull/3326), [#3345](https://github.com/milvus-io/pymilvus/pull/3345)) b |
| `v2.6.10` | 2026-03-13 | Low | ## What's Changed                                                                                                                ### New Features                                                                                                - feat: add global cluster client support (#3264)   - feat: add ConnectionManager to replace connections singleton for MilvusClient (#3307)    ### Bug Fixes    - fix: is_legal_ids() to accept negative int64 primary keys (#3281)   - fix: avoid call |
| `v2.6.9` | 2026-02-10 | Low | ### Bug Fixes  - **fix: pass context to `describe_collection` in `SearchIteratorV2`** ([#3271](https://github.com/milvus-io/pymilvus/pull/3271)) — Forward `kwargs` (including `context` carrying `db_name`) from `SearchIteratorV2.__init__` to `describe_collection`, matching the existing `QueryIterator` pattern. Without this fix, `SearchIteratorV2` always queries the `default` database, causing "can't find collection" errors for non-default databases.  - **fix: unpack `_get_schema` tuple return |
| `v2.6.8` | 2026-01-29 | Low | ## 🐛 Bug Fixes  - **Parse dbname from uri** ([#3247](https://github.com/milvus-io/pymilvus/pull/3247))  ## 🔧 Enhancements  - **Remove the usage of grpcio-testing** ([#3225](https://github.com/milvus-io/pymilvus/pull/3225)) - **Exclude data generation time from benchmark** ([#3246](https://github.com/milvus-io/pymilvus/pull/3246)) - **Optimize search result performance (Hit class, HybridHits init)** ([#3248](https://github.com/milvus-io/pymilvus/pull/3248)) - **Add channel_state and |
| `v2.6.7` | 2026-01-26 | Low | ## 🎉 New Features  - **Support semantic-highlighter** ([#3180](https://github.com/milvus-io/pymilvus/pull/3180)) - **Support async client list_persistent_segments** ([#3207](https://github.com/milvus-io/pymilvus/pull/3207))  ## 🐛 Bug Fixes  - **Pass parameters as keyword arguments to avoid incompatible issue of minio-py 7.2.19** ([#3171](https://github.com/milvus-io/pymilvus/pull/3171)) - **Float scalar datatype supports floating-point numbers** ([#3173](https://github.com/milvus-io/py |
| `v2.6.6` | 2025-12-30 | Low | ## Features  **1. Add comprehensive benchmarking framework (#3151)**  **2. Add Function SDK support (#3080)**  ## Enhancements  **1. Change highlight query keyword (#3143)**  ## Bug Fixes  1. **fix: Keep boolean serialization as uppercase for e2e compatibility** (#3168) 2. **fix: Preserve False boolean values in SCANN index parameters** (#3165) 3. **fix: Use json to parse deep recursive json when orjson failed** (#3156) 4. **fix: Fix search params in orm search iterator** (#3162 |

## Citation

- HTML: https://www.freshcrate.ai/projects/pymilvus
- Markdown: https://www.freshcrate.ai/projects/pymilvus.md
- Dependencies JSON: https://www.freshcrate.ai/api/projects/pymilvus/deps

_Generated by freshcrate.ai. Indexes github releases for AI-agent ecosystem packages._