Skip to main content

Installation

Requirements​

  • Python 3.12 or higher
  • A Milvus instance to connect to — a real server, or Milvus Lite (embedded, no server needed, bundled with pymilvus ≥ 2.4.2)

Core package​

The DBAPI is the only hard requirement. Install it on its own if you only need cursor.execute(...) — a script, a notebook, a raw async client:

pip install milvusql

Or with uv:

uv add milvusql

This pulls in sqlglot-milvus (the MilvusQL parser), pymilvus, and polars (the client-side relational engine used for JOIN/GROUP BY/subqueries) — nothing else.

SQLAlchemy​

pip install milvusql-sqlalchemy

Installing it registers the milvusql dialect under SQLAlchemy's entry points — create_engine("milvusql://...") works as soon as the package is importable, no explicit registration call needed.

Django​

pip install milvusql-django

Point DATABASES["default"]["ENGINE"] at "milvusql_django" (see Django → Overview).

Everything​

Each package depends on milvusql, so installing all three doesn't duplicate anything:

pip install milvusql-sqlalchemy milvusql-django

A note on sqlglot[c]/sqlglot[rs]​

sqlglot-milvus (and therefore milvusql) needs the pure-Python build of sqlglot. The mypyc-compiled variants (sqlglot[c], sqlglot[rs]) disable runtime subclassing of the classes every third-party dialect is built on. sqlglot-milvus detects this at import time and raises ImportError with the fix, rather than letting it fail confusingly deep inside sqlglot on your first parse_one() call.