fabricatio-milvus

MIT

Milvus vector database integration for Fabricatio — store, search, and retrieve document embeddings with the Fabricatio RAG framework.

Installation

pip install fabricatio[milvus]

Or install all Fabricatio components:

pip install fabricatio[full]

Overview

fabricatio-milvus extends Fabricatio's RAG (Retrieval-Augmented Generation) system with a Milvus vector database backend. It provides:

• Pydantic-based document models that auto-generate Milvus collection schemas from field type annotations.
• A MilvusRAG capability class implementing the add_document/afetch_document/aretrieve contract backed by pymilvus.
• Ready-to-use Action subclasses (InjectToDB, MilvusRAGTalk) for building agent pipelines.

Configuration

Milvus settings are loaded from the Fabricatio config system under the milvus namespace:

from fabricatio_milvus.config import milvus_config

# milvus_config.milvus_uri       → str | None
# milvus_config.milvus_timeout   → float | None
# milvus_config.milvus_token     → SecretStr | None
# milvus_config.milvus_dimensions → int | None

These can also be set per-instance via MilvusScopedConfig (pydantic model with the same fields).

Key Components

Models — fabricatio_milvus.models

Capabilities — fabricatio_milvus.capabilities

MilvusRAG methods:

• client — property returning the MilvusClient, eagerly created on construction from config.
• add_document(data, config) — vectorize documents and insert into a collection. Creates the collection if it doesn't exist.
• afetch_document(query, config) — vectorize query strings, search Milvus, deduplicate by ID, sort by distance descending, and deserialize into typed document models.
• aretrieve(query, document_model, ...) — convenience wrapper that builds a FetchConfig and calls afetch_document.

Actions — fabricatio_milvus.actions

Usage Example

from fabricatio_milvus.actions.rag import InjectToDB
from fabricatio_milvus.models.milvus import MilvusClassicModel

# Create document models
docs = [
    MilvusClassicModel(text="Fabricatio is a Python library for building LLM applications."),
    MilvusClassicModel(text="Milvus is an open-source vector database."),
]

# Inject into Milvus
action = InjectToDB(
    collection_name="my_knowledge_base",
    milvus_uri="http://localhost:19530",
)
await action.execute(to_inject=docs)

Dependencies

• fabricatio-core — core interfaces, config, and utilities
• fabricatio-rag — RAG abstractions (RAG, document models)
• pymilvus (≥2.5.4) — Milvus Python SDK
• pydantic (≥2.7.1) — data validation and schema generation
• more-itertools (≥10.8.0) — additional itertools

License

MIT