Store knowledge in Milvus

The milvus backend embeds knowledge chunks and stores them in a Milvus collection. Use it when you already operate Milvus and need persistent knowledge shared by multiple agent instances. KnowledgeBase.index maps to the Milvus collection name.

Prerequisites

Install the extra that contains the Milvus and LlamaIndex integrations, and prepare a reachable Milvus instance:

pip install "veadk-python[extensions]"

The Milvus backend requires an embedding model. The following environment variables configure both Milvus and the embedding model:

export DATABASE_MILVUS_URI="http://127.0.0.1:19530"
export DATABASE_MILVUS_TOKEN="${MILVUS_TOKEN}"
export MODEL_EMBEDDING_NAME="doubao-embedding-text-240715"
export MODEL_EMBEDDING_DIM="2560"
export MODEL_EMBEDDING_API_KEY="${ARK_API_KEY}"

Instead of DATABASE_MILVUS_TOKEN, you can set DATABASE_MILVUS_USER and DATABASE_MILVUS_PASSWORD. A token takes precedence over the username and password pair.

Example

from veadk.knowledgebase import KnowledgeBase

knowledgebase = KnowledgeBase(
    backend="milvus",
    index="company_faq",
    top_k=3,
)

knowledgebase.add_from_text(
    [
        "The standard annual leave allowance is 15 days.",
        "With manager approval, employees may work remotely up to two days a week.",
    ]
)

for entry in knowledgebase.search("How many remote-work days are allowed?"):
    print(entry.content)

Parameters

Pass backend parameters through KnowledgeBase(backend="milvus", backend_config={...}). Without backend_config, VeADK reads the connection from environment variables and uses the global embedding configuration.

Parameter	Environment variable	Type	Default	Description
`index`	None	`str`	Required	Milvus collection name. It must start with a letter or underscore, contain only letters, digits, and underscores, and be at most 255 characters. Usually passed as `KnowledgeBase.index`.
`milvus_config.uri`	`DATABASE_MILVUS_URI`	`str`	`""`	Milvus service URI, or a local file path supported by Milvus Lite. Required.
`milvus_config.token`	`DATABASE_MILVUS_TOKEN`	`str`	`""`	Milvus token. Takes precedence over username and password.
`milvus_config.user`	`DATABASE_MILVUS_USER`	`str`	`""`	Milvus username, used with `password`.
`milvus_config.password`	`DATABASE_MILVUS_PASSWORD`	`str`	`""`	Milvus password, used with `user`.
`milvus_config.db_name`	`DATABASE_MILVUS_DB_NAME`	`str`	`"default"`	Milvus database name.
`milvus_config.overwrite`	`DATABASE_MILVUS_OVERWRITE`	`bool`	`False`	Whether initialization overwrites a collection with the same name.
`milvus_config.timeout`	`DATABASE_MILVUS_TIMEOUT`	`float \| None`	`None`	Milvus request timeout in seconds.
`milvus_config.output_fields`	`DATABASE_MILVUS_OUTPUT_FIELDS`	`list[str] \| str`	`[]`	Additional fields returned during retrieval. The environment variable accepts a comma-separated string or a JSON array string.
`embedding_config`	`MODEL_EMBEDDING_*`	`EmbeddingModelConfig`	Global embedding config	Embedding model name, dimension, API base, and API key.

DATABASE_MILVUS_OVERWRITE=true overwrites a collection with the same name during initialization and may remove existing knowledge. Enable it only when that data does not need to be retained.

In addition to search(query, top_k), this backend supports add_from_text, add_from_files, and add_from_directory. File and directory imports split content by file type before storing it in the collection.

Overview

Get started

Core components

Observability

Extensions

Frontend

Security

Prerequisites

Example

Parameters

​Prerequisites

​Example

​Parameters

Prerequisites

Example

Parameters