README

English ｜简体中文

A lightweight PHP library for interacting with the Milvus vector database via RESTful API — manage collections, insert data, and perform vector searches, all in one place.

This project has been parsed by Zread. Click here for a quick overview of the project structure.

Requirements

PHP 8.0+
ext-curl
ext-json

Installation

composer require hejunjie/milvus

Quick Start

use Hejunjie\Milvus;

// Initialize client (get the URL and API Key from Zilliz Cloud console)
$client = new Milvus\Client('https://xxx.api.zillizcloud.com', 'your-api-key');

// Create a collection
$client->collections()->create('my_collection', [
    'autoID' => true,
    'fields' => [
        ['fieldName' => 'id', 'dataType' => 'Int64', 'isPrimary' => true],
        ['fieldName' => 'vector', 'dataType' => 'FloatVector', 'elementTypeParams' => ['dim' => 128]],
        ['fieldName' => 'title', 'dataType' => 'VarChar', 'elementTypeParams' => ['max_length' => 256]],
    ]
]);

// Create an index (vector fields require an index before searching)
$client->indexes()->create('my_collection', [[
    'fieldName' => 'vector',
    'indexName' => 'vector_idx',
    'metricType' => 'COSINE',
    'indexConfig' => ['index_type' => 'AUTOINDEX'],
]]);

// Load the collection into memory
$client->collections()->load('my_collection');

// Insert data
$client->entities()->insert('my_collection', [
    ['vector' => array_fill(0, 128, 0.1), 'title' => 'First record'],
    ['vector' => array_fill(0, 128, 0.2), 'title' => 'Second record'],
]);

// Vector search
$results = $client->entities()->search('my_collection', [[
    'vector' => array_fill(0, 128, 0.15),
]], 'vector', limit: 10);

API Overview

All methods are accessed via chained calls, organized by module:

Client

Method	Description
`customize($url, $params)`	Custom request for calling APIs not yet wrapped

Collections

Method	Description
`create($name, $schema, $indexParams)`	Create a collection
`describe($name)`	Get collection details
`drop($name)`	Drop a collection
`get_load_state($name)`	Get collection load state
`get_stats($name)`	Get collection statistics
`has($name)`	Check if a collection exists
`list()`	List all collections
`load($name)`	Load a collection into memory
`release($name)`	Release a collection from memory
`rename($name, $newName)`	Rename a collection
`alter_properties($name, $properties)`	Set collection properties (e.g., TTL, mmap)
`drop_properties($name, $keys)`	Remove collection properties

Partitions

Method	Description
`create($collectionName, $partitionName)`	Create a partition
`drop($collectionName, $partitionName)`	Drop a partition
`get_stats($collectionName, $partitionName)`	Get partition load state
`has($collectionName, $partitionName)`	Check if a partition exists
`list($collectionName)`	List all partitions
`load($collectionName, $partitionNames)`	Load partitions into memory
`release($collectionName, $partitionNames)`	Release partitions from memory

Indexes

Method	Description
`create($collectionName, $indexParams)`	Create an index
`describe($collectionName, $indexName)`	Get index details
`drop($collectionName, $indexName)`	Drop an index
`list($collectionName)`	List all indexes

Aliases

Method	Description
`alter($collectionName, $aliasName)`	Alter an alias
`create($collectionName, $aliasName)`	Create an alias
`describe($collectionName, $aliasName)`	Get alias details
`drop($collectionName, $aliasName)`	Drop an alias
`list($collectionName)`	List all aliases

Entities

Method	Description
`insert($collectionName, $data, $partitionName)`	Insert data
`upsert($collectionName, $data, $partitionName)`	Insert or update data
`delete($collectionName, $filter, $partitionName)`	Delete data by filter
`get($collectionName, $ids, $outputFields, $partitionNames)`	Get data by primary key
`query($collectionName, $filter, $limit, $offset, $outputFields, $partitionNames)`	Query data by filter
`search($collectionName, $data, $annsField, $limit, $offset, $partitionNames)`	Vector similarity search

Note

All methods have full PHPDoc comments with parameter details. For more information on Milvus RESTful API parameters, refer to the Zilliz official documentation.

Notes

This library is built on Milvus RESTful API v2. Make sure your Milvus / Zilliz Cloud instance supports v2 endpoints.
Before performing vector search, the collection must have an index created and be loaded into memory (via load()), otherwise the search will fail.
Default request timeout is 10 seconds. Large data inserts may require batching.
API Key can be null (for self-hosted Milvus without authentication), but is required for Zilliz Cloud managed service.

Motivation

I was reading short stories on a novel site that didn't support search and had daily reading limits. Frustrated, I wrote a crawler to download all the stories — but then I couldn't find anything in them locally.

My tiny 2-core 2GB server couldn't run a Milvus instance, but Zilliz Cloud offers a free tier. I run vectorization on my Mac, and use SiliconFlow's free quota for vector queries during search. I just needed a PHP library to tie it all together — and that's how this library came to be.

Contributing

If you find this library useful, give it a ⭐ Star, or submit an Issue / PR to share your ideas.

hejunjie / milvus

Maintainers

Package info

Statistics

Security