feat: add optional Milvus vector backend

README.md

@@ -320,7 +320,7 @@ opencontext/
 
 5. **Storage Layer** (`storage/`)
 
-   - Multi-backend support (SQLite, ChromaDB)
+   - Multi-backend support (SQLite, ChromaDB, Qdrant, and Milvus)
    - Vector storage for similarity search
    - Unified storage interface
 
@@ -379,6 +379,33 @@ capture:
    - `prompts_en.yaml`: English prompt templates
    - `prompts_zh.yaml`: Chinese prompt templates
 
+#### Milvus vector backend
+
+Milvus is available as an optional vector backend. Install the extra before selecting it:
+
+```bash
+uv sync --extra milvus
+```
+
+Then replace the default ChromaDB vector backend in `config/config.yaml`:
+
+```yaml
+- name: "default_vector"
+  storage_type: "vector_db"
+  backend: "milvus"
+  config:
+    vector_size: 2048 # Must match the embedding model output dimension
+    uri: "${MILVUS_URI:./persist/milvus.db}"
+    token: "${MILVUS_TOKEN:}"
+    collection_prefix: "opencontext"
+    consistency_level: "Session"
+```
+
+The same `uri` setting supports Milvus Lite database files, remote Milvus servers such as
+`http://localhost:19530`, and Zilliz Cloud endpoints. Set `MILVUS_TOKEN` for authenticated
+servers. Milvus Lite is not supported on Windows; Windows users must configure an `http://`
+or `https://` remote URI. ChromaDB remains the default backend.
+
 ### Running the Server
 
 ```bash

config/config.yaml

@@ -146,7 +146,7 @@ processing:
 storage:
   enabled: true
   backends:
-    # Vector database configuration - Choose either ChromaDB or Qdrant
+    # Vector database configuration - Choose ChromaDB, Qdrant, or Milvus
     # ChromaDB (default)
     - name: "default_vector"
       storage_type: "vector_db"
@@ -174,6 +174,25 @@ storage:
     #     # https: false
     #     # api_key: "${QDRANT_API_KEY}"  # Optional: API key for authentication
 
+    # Milvus (alternative vector database)
+    # Install first with: uv sync --extra milvus
+    # Uncomment to use Milvus instead of ChromaDB:
+    # - name: "default_vector"
+    #   storage_type: "vector_db"
+    #   backend: "milvus"
+    #   config:
+    #     # Vector size MUST match your embedding model's output dimension
+    #     vector_size: 2048
+    #     collection_prefix: "opencontext"
+    #     consistency_level: "Session"
+    #
+    #     # Milvus Lite (local file; not supported on Windows):
+    #     uri: "${MILVUS_URI:./persist/milvus.db}"
+    #
+    #     # Remote Milvus or Zilliz Cloud (also supported on Windows):
+    #     # uri: "${MILVUS_URI:http://localhost:19530}"
+    #     # token: "${MILVUS_TOKEN:}"  # Required for authenticated endpoints
+
     - name: "document_store"
       storage_type: "document_db"
       backend: "sqlite"

opencontext/storage/backends/__init__.py

@@ -4,14 +4,23 @@
 # Copyright (c) 2025 Beijing Volcano Engine Technology Co., Ltd.
 # SPDX-License-Identifier: Apache-2.0
 
-"""
-Storage backend package initialization file
-"""
+"""Storage backend package with lazy provider imports."""
 
-from .chromadb_backend import ChromaDBBackend
-from .sqlite_backend import SQLiteBackend
+from importlib import import_module
 
-try:
-    __all__ = ["SQLiteBackend", "ChromaDBBackend"]
-except ImportError:
-    __all__ = ["SQLiteBackend", "ChromaDBBackend"]
+__all__ = ["SQLiteBackend", "ChromaDBBackend", "QdrantBackend", "MilvusBackend"]
+
+_BACKEND_MODULES = {
+    "SQLiteBackend": ".sqlite_backend",
+    "ChromaDBBackend": ".chromadb_backend",
+    "QdrantBackend": ".qdrant_backend",
+    "MilvusBackend": ".milvus_backend",
+}
+
+
+def __getattr__(name: str):
+    if name not in _BACKEND_MODULES:
+        raise AttributeError(f"module {__name__!r} has no attribute {name!r}")
+    backend = getattr(import_module(_BACKEND_MODULES[name], __name__), name)
+    globals()[name] = backend
+    return backend