feat: add support for Weaviate vector database (#14747)

backend/open_webui/config.py

@@ -2180,6 +2180,11 @@ ENABLE_QDRANT_MULTITENANCY_MODE = (
 )
 QDRANT_COLLECTION_PREFIX = os.environ.get("QDRANT_COLLECTION_PREFIX", "open-webui")
 
+WEAVIATE_HTTP_HOST = os.environ.get("WEAVIATE_HTTP_HOST", "")
+WEAVIATE_HTTP_PORT = int(os.environ.get("WEAVIATE_HTTP_PORT", "8080"))
+WEAVIATE_GRPC_PORT = int(os.environ.get("WEAVIATE_GRPC_PORT", "50051"))
+WEAVIATE_API_KEY = os.environ.get("WEAVIATE_API_KEY")
+
 # OpenSearch
 OPENSEARCH_URI = os.environ.get("OPENSEARCH_URI", "https://localhost:9200")
 OPENSEARCH_SSL = os.environ.get("OPENSEARCH_SSL", "true").lower() == "true"

backend/open_webui/retrieval/vector/dbs/weaviate.py

@@ -0,0 +1,301 @@
+import weaviate
+import re
+import uuid
+from typing import Any, Dict, List, Optional, Union
+
+from open_webui.retrieval.vector.main import (
+    VectorDBBase,
+    VectorItem,
+    SearchResult,
+    GetResult,
+)
+from open_webui.retrieval.vector.utils import process_metadata
+from open_webui.config import WEAVIATE_HTTP_HOST, WEAVIATE_HTTP_PORT, WEAVIATE_GRPC_PORT, WEAVIATE_API_KEY
+
+
+def _convert_uuids_to_strings(obj: Any) -> Any:
+    """
+    Recursively convert UUID objects to strings in nested data structures.
+    
+    This function handles:
+    - UUID objects -> string
+    - Dictionaries with UUID values
+    - Lists/Tuples with UUID values
+    - Nested combinations of the above
+    
+    Args:
+        obj: Any object that might contain UUIDs
+        
+    Returns:
+        The same object structure with UUIDs converted to strings
+    """
+    if isinstance(obj, uuid.UUID):
+        return str(obj)
+    elif isinstance(obj, dict):
+        return {key: _convert_uuids_to_strings(value) for key, value in obj.items()}
+    elif isinstance(obj, (list, tuple)):
+        return type(obj)(_convert_uuids_to_strings(item) for item in obj)
+    elif isinstance(obj, (str, int, float, bool, type(None))):
+        return obj
+    else:
+        return obj
+
+
+
+
+class WeaviateClient(VectorDBBase):
+    def __init__(self):
+        self.url = WEAVIATE_HTTP_HOST
+        try:
+           # Build connection parameters
+            connection_params = {
+                "host": WEAVIATE_HTTP_HOST,
+                "port": WEAVIATE_HTTP_PORT,
+                "grpc_port": WEAVIATE_GRPC_PORT,
+            }
+            
+            # Only add auth_credentials if WEAVIATE_API_KEY exists and is not empty
+            if WEAVIATE_API_KEY:
+                connection_params["auth_credentials"] = weaviate.classes.init.Auth.api_key(WEAVIATE_API_KEY)
+            
+            self.client = weaviate.connect_to_local(**connection_params)
+            self.client.connect()
+        except Exception as e:
+            raise ConnectionError(f"Failed to connect to Weaviate: {e}") from e
+
+    def _sanitize_collection_name(self, collection_name: str) -> str:
+        """Sanitize collection name to be a valid Weaviate class name."""
+        if not isinstance(collection_name, str) or not collection_name.strip():
+            raise ValueError("Collection name must be a non-empty string")
+
+        # Requirements for a valid Weaviate class name:
+        # The collection name must begin with a capital letter.
+        # The name can only contain letters, numbers, and the underscore (_) character. Spaces are not allowed.
+
+        # Replace hyphens with underscores and keep only alphanumeric characters
+        name = re.sub(r'[^a-zA-Z0-9_]', '', collection_name.replace("-", "_"))
+        name = name.strip("_")
+
+        if not name:
+            raise ValueError("Could not sanitize collection name to be a valid Weaviate class name")
+
+        # Ensure it starts with a letter and is capitalized
+        if not name[0].isalpha():
+            name = "C" + name
+        
+        return name[0].upper() + name[1:]
+
+    def has_collection(self, collection_name: str) -> bool:
+        sane_collection_name = self._sanitize_collection_name(collection_name)
+        return self.client.collections.exists(sane_collection_name)
+
+    def delete_collection(self, collection_name: str) -> None:
+        sane_collection_name = self._sanitize_collection_name(collection_name)
+        if self.client.collections.exists(sane_collection_name):
+            self.client.collections.delete(sane_collection_name)
+
+    def _create_collection(self, collection_name: str) -> None:
+        self.client.collections.create(
+            name=collection_name,
+            vector_config=weaviate.classes.config.Configure.Vectors.self_provided(),
+            properties=[
+                weaviate.classes.config.Property(name="text", data_type=weaviate.classes.config.DataType.TEXT),
+            ]
+        )
+
+    def insert(self, collection_name: str, items: List[VectorItem]) -> None:
+        sane_collection_name = self._sanitize_collection_name(collection_name)
+        if not self.client.collections.exists(sane_collection_name):
+            self._create_collection(sane_collection_name)
+
+        collection = self.client.collections.get(sane_collection_name)
+        
+        with collection.batch.fixed_size(batch_size=100) as batch:
+            for item in items:
+                item_uuid = str(uuid.uuid4()) if not item["id"] else str(item["id"])
+                
+                properties = {"text": item["text"]}
+                if item["metadata"]:
+                    clean_metadata = _convert_uuids_to_strings(process_metadata(item["metadata"]))
+                    clean_metadata.pop("text", None)
+                    properties.update(clean_metadata)
+                
+                batch.add_object(
+                    properties=properties,
+                    uuid=item_uuid,
+                    vector=item["vector"]
+                )
+
+    def upsert(self, collection_name: str, items: List[VectorItem]) -> None:
+        sane_collection_name = self._sanitize_collection_name(collection_name)
+        if not self.client.collections.exists(sane_collection_name):
+            self._create_collection(sane_collection_name)
+
+        collection = self.client.collections.get(sane_collection_name)
+        
+        with collection.batch.fixed_size(batch_size=100) as batch:
+            for item in items:
+                item_uuid = str(item["id"]) if item["id"] else None
+                
+                properties = {"text": item["text"]}
+                if item["metadata"]:
+                    clean_metadata = _convert_uuids_to_strings(process_metadata(item["metadata"]))
+                    clean_metadata.pop("text", None)
+                    properties.update(clean_metadata)
+                
+                batch.add_object(
+                    properties=properties,
+                    uuid=item_uuid,
+                    vector=item["vector"]
+                )
+
+    def search(
+        self, collection_name: str, vectors: List[List[Union[float, int]]], limit: int
+    ) -> Optional[SearchResult]:
+        sane_collection_name = self._sanitize_collection_name(collection_name)
+        if not self.client.collections.exists(sane_collection_name):
+            return None
+
+        collection = self.client.collections.get(sane_collection_name)
+        
+        result_ids, result_documents, result_metadatas, result_distances = [], [], [], []
+        
+        for vector_embedding in vectors:
+            try:
+                response = collection.query.near_vector(
+                    near_vector=vector_embedding,
+                    limit=limit,
+                    return_metadata=weaviate.classes.query.MetadataQuery(distance=True),
+                )
+                
+                ids = [str(obj.uuid) for obj in response.objects]
+                documents = []
+                metadatas = []
+                distances = []
+                
+                for obj in response.objects:
+                    properties = dict(obj.properties) if obj.properties else {}
+                    documents.append(properties.pop("text", ""))
+                    metadatas.append(_convert_uuids_to_strings(properties))
+                    
+                # Weaviate has cosine distance, 2 (worst) -> 0 (best). Re-ordering to 0 -> 1
+                raw_distances = [obj.metadata.distance if obj.metadata and obj.metadata.distance else 2.0 for obj in response.objects]
+                distances = [(2 - dist) / 2 for dist in raw_distances]
+                
+                result_ids.append(ids)
+                result_documents.append(documents)
+                result_metadatas.append(metadatas)
+                result_distances.append(distances)
+            except Exception:
+                result_ids.append([])
+                result_documents.append([])
+                result_metadatas.append([])
+                result_distances.append([])
+        
+        return SearchResult(
+            **{
+                "ids": result_ids,
+                "documents": result_documents,
+                "metadatas": result_metadatas,
+                "distances": result_distances,
+            }
+        )
+
+    def query(
+        self, collection_name: str, filter: Dict, limit: Optional[int] = None
+    ) -> Optional[GetResult]:
+        sane_collection_name = self._sanitize_collection_name(collection_name)
+        if not self.client.collections.exists(sane_collection_name):
+            return None
+
+        collection = self.client.collections.get(sane_collection_name)
+        
+        weaviate_filter = None
+        if filter:
+            for key, value in filter.items():
+                prop_filter = weaviate.classes.query.Filter.by_property(name=key).equal(value)
+                weaviate_filter = prop_filter if weaviate_filter is None else weaviate.classes.query.Filter.all_of([weaviate_filter, prop_filter])
+        
+        try:
+            response = collection.query.fetch_objects(filters=weaviate_filter, limit=limit)
+            
+            ids = [str(obj.uuid) for obj in response.objects]
+            documents = []
+            metadatas = []
+
+            for obj in response.objects:
+                properties = dict(obj.properties) if obj.properties else {}
+                documents.append(properties.pop("text", ""))
+                metadatas.append(_convert_uuids_to_strings(properties))
+
+            return GetResult(
+                **{
+                    "ids": [ids],
+                    "documents": [documents],
+                    "metadatas": [metadatas],
+                }
+            )
+        except Exception:
+            return None
+
+    def get(self, collection_name: str) -> Optional[GetResult]:
+        sane_collection_name = self._sanitize_collection_name(collection_name)
+        if not self.client.collections.exists(sane_collection_name):
+            return None
+
+        collection = self.client.collections.get(sane_collection_name)
+        ids, documents, metadatas = [], [], []
+
+        try:
+            for item in collection.iterator():
+                ids.append(str(item.uuid))
+                properties = dict(item.properties) if item.properties else {}
+                documents.append(properties.pop("text", ""))
+                metadatas.append(_convert_uuids_to_strings(properties))
+            
+            if not ids:
+                return None
+                
+            return GetResult(
+                **{
+                    "ids": [ids],
+                    "documents": [documents],
+                    "metadatas": [metadatas],
+                }
+            )
+        except Exception:
+            return None
+
+    def delete(
+        self,
+        collection_name: str,
+        ids: Optional[List[str]] = None,
+        filter: Optional[Dict] = None,
+    ) -> None:
+        sane_collection_name = self._sanitize_collection_name(collection_name)
+        if not self.client.collections.exists(sane_collection_name):
+            return
+
+        collection = self.client.collections.get(sane_collection_name)
+
+        try:
+            if ids:
+                for item_id in ids:
+                    collection.data.delete_by_id(uuid=item_id)
+            elif filter:
+                weaviate_filter = None
+                for key, value in filter.items():
+                    prop_filter = weaviate.classes.query.Filter.by_property(name=key).equal(value)
+                    weaviate_filter = prop_filter if weaviate_filter is None else weaviate.classes.query.Filter.all_of([weaviate_filter, prop_filter])
+                    
+                if weaviate_filter:
+                    collection.data.delete_many(where=weaviate_filter)
+        except Exception:
+            pass
+
+    def reset(self) -> None:
+        try:
+            for collection_name in self.client.collections.list_all().keys():
+                self.client.collections.delete(collection_name)
+        except Exception:
+            pass

backend/open_webui/retrieval/vector/factory.py

@@ -67,6 +67,10 @@ class Vector:
                 from open_webui.retrieval.vector.dbs.oracle23ai import Oracle23aiClient
 
                 return Oracle23aiClient()
+            case VectorType.WEAVIATE:
+                from open_webui.retrieval.vector.dbs.weaviate import WeaviateClient
+
+                return WeaviateClient()
             case _:
                 raise ValueError(f"Unsupported vector type: {vector_type}")
 

backend/open_webui/retrieval/vector/type.py

@@ -11,3 +11,4 @@ class VectorType(StrEnum):
     PGVECTOR = "pgvector"
     ORACLE23AI = "oracle23ai"
     S3VECTOR = "s3vector"
+    WEAVIATE = "weaviate"

backend/requirements.txt

@@ -49,6 +49,7 @@ langchain-community==0.3.29
 
 fake-useragent==2.2.0
 chromadb==1.1.0
+weaviate-client==4.17.0
 opensearch-py==2.8.0
 
 transformers