Use Milvus storage - AgentKit Documents

The milvus backend stores knowledge-chunk embeddings in a Milvus collection. It can connect to Milvus or Zilliz Cloud, or use a local Milvus Lite file. This backend is available in VeADK 1.0.3 and later.

Dependencies

pip install "veadk-python[extensions]"

Use Milvus Lite

For local validation, point DATABASE_MILVUS_URI to a Milvus Lite data file and configure an embedding model:

export DATABASE_MILVUS_URI="./data/product_docs.db"
export MODEL_EMBEDDING_API_KEY="your-ark-api-key"

The first run creates the data file. Later runs can retrieve existing content by using the same file and collection name.

Example

from veadk import Agent
from veadk.knowledgebase import KnowledgeBase

kb = KnowledgeBase(
    backend="milvus",
    index="product_docs",
    backend_config={
        "index": "product_docs",
        "milvus_config": {
            "uri": "https://milvus.example.com",
            "token": "your-milvus-token",
            "db_name": "default",
        },
    },
)
kb.add_from_directory("./docs")

agent = Agent(knowledgebase=kb)

Parameters

Parameter	Type	Default	Description
`index`	`str`	None	Milvus collection name. It must start with a letter or underscore, contain only letters, digits, and underscores, and be at most 255 characters.
`milvus_config`	`MilvusConfig`	Read from `DATABASE_MILVUS_*`	Milvus connection and collection settings.
`embedding_config`	`EmbeddingModelConfig`	Read from `MODEL_EMBEDDING_*`	Embedding model used to vectorize knowledge chunks.

Milvus configuration

Setting	Environment variable	Type	Default	Description
`uri`	`DATABASE_MILVUS_URI`	`str`	`""`	Remote Milvus URI or local Milvus Lite file path. Required.
`token`	`DATABASE_MILVUS_TOKEN`	`str`	`""`	Milvus token; takes precedence over username and password.
`user`	`DATABASE_MILVUS_USER`	`str`	`""`	Username, used with `password`.
`password`	`DATABASE_MILVUS_PASSWORD`	`str`	`""`	Password.
`db_name`	`DATABASE_MILVUS_DB_NAME`	`str`	`default`	Database name.
`overwrite`	`DATABASE_MILVUS_OVERWRITE`	`bool`	`false`	Whether to replace an existing collection during initialization.
`timeout`	`DATABASE_MILVUS_TIMEOUT`	`float \| None`	`None`	Request timeout in seconds.
`output_fields`	`DATABASE_MILVUS_OUTPUT_FIELDS`	`list[str] \| str`	Empty	Fields returned with search results; use a comma-separated value or JSON array in the environment variable.

Enabling overwrite can remove existing data from a collection with the same name. Use it only when the knowledge base can be rebuilt.

Last modified on September 19, 2026