Milvus Lite

2.5.1 · active · verified Fri Apr 10

Milvus Lite is a lightweight, embedded version of Milvus, a high-performance vector database, designed for rapid prototyping, local development, and edge devices. It provides core vector search functionalities and shares the same API as Milvus Standalone and Distributed deployments, ensuring a consistent development experience across various scales. Data is persisted locally in an SQLite file. The library is actively maintained as part of the `pymilvus` ecosystem, with the current version being 2.5.1.

Warnings

gotcha Milvus Lite is designed for small-scale prototyping (typically less than a million vectors) or edge devices. It is not recommended for large-scale production deployments where Milvus Standalone, Distributed, or Zilliz Cloud should be used instead.
Fix: Evaluate your scale requirements early. For larger datasets or production, plan to migrate to other Milvus deployments. Milvus Lite provides a command-line tool for data migration.
breaking Prior to Milvus Lite version 2.4.11, only the FLAT index type was supported, regardless of any other index type specified during collection creation. For versions 2.4.11 and later, both FLAT and IVF_FLAT are supported, with automatic internal switching (FLAT for <100,000 vectors, IVF_FLAT for >=100,000 vectors).
Fix: If precise control over index types or more advanced indexing is critical, ensure you are on version 2.4.11 or later for IVF_FLAT support. For full index flexibility, use a non-Lite Milvus deployment.
gotcha Milvus Lite does not support advanced Milvus features such as partitions, users/roles/RBAC (Role-Based Access Control), or aliases. Attempting to use these features will result in errors or unexpected behavior.
Fix: If your application requires partitions, user management, or aliases, you must use Milvus Standalone, Milvus Distributed, or Zilliz Cloud.
gotcha While Milvus Lite uses the same Python client API (`pymilvus`) as other Milvus deployments, the `uri` parameter for `MilvusClient` differs. For Milvus Lite, it's a local file path (e.g., './milvus_demo.db'). For other deployments, it's a network endpoint (e.g., 'http://localhost:19530' or a Zilliz Cloud endpoint).
Fix: Be mindful of the `uri` parameter when transitioning an application developed with Milvus Lite to a different Milvus deployment. The rest of the client-side code should largely remain compatible.

Install

pip install -U pymilvus[milvus-lite] Recommended Installation
pip install -U milvus-lite Install milvus-lite package directly

Imports

MilvusClient
```
from pymilvus import MilvusClient
```
Milvus Lite is accessed via the MilvusClient from the pymilvus library by specifying a local file URI.
default_server
```
from milvus import default_server
```
For explicit server control (e.g., setting base directory), default_server is imported from 'milvus', not 'milvus_lite'.

Quickstart

This quickstart demonstrates how to initialize Milvus Lite using `MilvusClient` with a local database file, define a collection schema, insert data (vectors and scalar fields), perform a vector similarity search, and query data. The `uri` parameter for `MilvusClient` points to a local file, which acts as the Milvus Lite database.

from pymilvus import MilvusClient, DataType
import os

# Ensure a clean slate for demonstration
if os.path.exists("milvus_demo.db"): 
    os.remove("milvus_demo.db")

# 1. Set up a Milvus client with a local file for persistence
client = MilvusClient(uri="./milvus_demo.db")

# 2. Create schema
schema = MilvusClient.create_schema(
    auto_id=False,
    enable_dynamic_field=True,
)
schema.add_field(field_name="id", datatype=DataType.INT64, is_primary=True)
schema.add_field(field_name="vector", datatype=DataType.FLOAT_VECTOR, dim=5)
schema.add_field(field_name="text", datatype=DataType.VARCHAR, max_length=256)

# 3. Create a collection
collection_name = "demo_collection"
client.create_collection(
    collection_name=collection_name,
    schema=schema,
    # You can specify an index here, Milvus Lite will optimize internally
    # index_params=MilvusClient.prepare_index_params(metric_type="L2")
)

# 4. Insert data
data = [
    {"id": 1, "vector": [0.1, 0.2, 0.3, 0.4, 0.5], "text": "The quick brown fox"},
    {"id": 2, "vector": [0.5, 0.4, 0.3, 0.2, 0.1], "text": "Jumps over the lazy dog"},
    {"id": 3, "vector": [0.8, 0.7, 0.6, 0.5, 0.4], "text": "A fast animal"},
]
client.insert(collection_name=collection_name, data=data)
client.flush(collection_name=collection_name)

# 5. Search for similar vectors
search_vectors = [[0.15, 0.25, 0.35, 0.45, 0.55]]
res = client.search(
    collection_name=collection_name,
    data=search_vectors,
    limit=2,
    output_fields=["text"],
)
print("Search Results:", res)

# 6. Query data by ID
query_res = client.query(collection_name=collection_name, filter="id in", output_fields=["text"])
print("Query Results:", query_res)

# Clean up (optional for next run)
client.drop_collection(collection_name=collection_name)
client.close() # Important to close client to ensure data is written to disk

print("Milvus Lite demo finished successfully!")

view raw JSON →