Skip to content

Class swarmauri_vectorstore_duckdb.DuckDBVectorStore.DuckDBVectorStore

swarmauri_vectorstore_duckdb.DuckDBVectorStore.DuckDBVectorStore

DuckDBVectorStore(**data)

Bases: VectorStoreSaveLoadMixin, VectorStoreRetrieveMixin, VectorStoreBase

A vector store implementation using DuckDB as the backend.

Source code in swarmauri_vectorstore_duckdb/DuckDBVectorStore.py
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
def __init__(self, **data):
    super().__init__(**data)
    self._embedder = Doc2VecEmbedding()
    self._distance = CosineDistance()

    if not os.path.exists(self.persist_dir):
        os.makedirs(self.persist_dir)

    if self.database_name == ":memory:":
        self._conn = duckdb.connect(self.database_name)
        self._setup_extensions(self._conn)
        self._initialize_table(self._conn)
    else:
        self._database_path = os.path.join(self.persist_dir, self.database_name)
        with duckdb.connect(self._database_path) as conn:
            self._setup_extensions(conn)
            self._initialize_table(conn)

type class-attribute instance-attribute

type = 'DuckDBVectorStore'

database_name class-attribute instance-attribute

database_name = Field(
    default=":memory:",
    description="Name of the DuckDB database",
)

table_name class-attribute instance-attribute

table_name = Field(
    default="documents",
    description="Name of the table to store documents",
)

embed_dim class-attribute instance-attribute

embed_dim = Field(
    default=None,
    description="Dimension of the embedding vectors",
)

persist_dir class-attribute instance-attribute

persist_dir = Field(
    default="./storage",
    description="Directory to persist the database",
)

model_config class-attribute instance-attribute

model_config = ConfigDict(
    extra="allow", arbitrary_types_allowed=True
)

id class-attribute instance-attribute

id = Field(default_factory=generate_id)

members class-attribute instance-attribute

members = None

owners class-attribute instance-attribute

owners = None

host class-attribute instance-attribute

host = None

default_logger class-attribute

default_logger = None

logger class-attribute instance-attribute

logger = None

name class-attribute instance-attribute

name = None

resource class-attribute instance-attribute

resource = Field(default=VECTOR_STORE.value)

version class-attribute instance-attribute

version = '0.1.0'

documents class-attribute instance-attribute

documents = []

embedder property

embedder

connect

connect()

Connect to the DuckDB database and initialize if necessary.

Source code in swarmauri_vectorstore_duckdb/DuckDBVectorStore.py
79
80
81
82
83
84
85
86
87
88
def connect(self) -> None:
    """Connect to the DuckDB database and initialize if necessary."""
    if not self._is_initialized:
        if self.database_name == ":memory:":
            self._initialize_table(self._conn)
        else:
            with duckdb.connect(self._database_path) as conn:
                self._setup_extensions(conn)
                self._initialize_table(conn)
        self._is_initialized = True

disconnect

disconnect()

Disconnect from the DuckDB database.

Source code in swarmauri_vectorstore_duckdb/DuckDBVectorStore.py
102
103
104
105
106
107
def disconnect(self) -> None:
    """Disconnect from the DuckDB database."""
    if self._conn and self.database_name == ":memory:":
        self._conn.close()
        self._conn = None
    self._is_initialized = False

add_document

add_document(document)
Source code in swarmauri_vectorstore_duckdb/DuckDBVectorStore.py
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
def add_document(self, document: Document) -> None:
    # Ensure the document is properly prepared before insertion
    data = self._prepare_document(document)
    query = f"""
        INSERT OR REPLACE INTO {self.table_name} (id, content, embedding, metadata)
        VALUES (?, ?, ?, ?)
    """
    if self.database_name == ":memory:":
        self._conn.execute(
            query,
            [data["id"], data["content"], data["embedding"], data["metadata"]],
        )
    else:
        with duckdb.connect(self._database_path) as conn:
            conn.execute(
                query,
                [data["id"], data["content"], data["embedding"], data["metadata"]],
            )

add_documents

add_documents(documents)
Source code in swarmauri_vectorstore_duckdb/DuckDBVectorStore.py
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
def add_documents(self, documents: List[Document]) -> None:
    ids = [doc.id for doc in documents]
    contents = [doc.content for doc in documents]

    self._embedder.fit(contents)  # Fit the embedder once with all contents

    embeddings = [
        self._embedder.transform([doc.content])[0].to_numpy().tolist()
        for doc in documents
    ]
    metadatas = [json.dumps(doc.metadata or {}) for doc in documents]

    data_list = list(zip(ids, contents, embeddings, metadatas))

    query = f"""
        INSERT OR REPLACE INTO {self.table_name} (id, content, embedding, metadata) 
        VALUES (?, ?, ?, ?)
    """

    if self.database_name == ":memory:":
        self._conn.executemany(query, data_list)
    else:
        with duckdb.connect(self._database_path) as conn:
            conn.executemany(query, data_list)

get_document

get_document(id)
Source code in swarmauri_vectorstore_duckdb/DuckDBVectorStore.py
173
174
175
176
177
178
179
180
181
182
183
184
185
def get_document(self, id: str) -> Optional[Document]:
    query = f"SELECT id, content, metadata FROM {self.table_name} WHERE id = ?"
    if self.database_name == ":memory:":
        result = self._conn.execute(query, [id]).fetchone()
    else:
        with duckdb.connect(self._database_path) as conn:
            result = conn.execute(query, [id]).fetchone()

    if result:
        return Document(
            id=result[0], content=result[1], metadata=json.loads(result[2])
        )
    return None

retrieve

retrieve(query, top_k=5)
Source code in swarmauri_vectorstore_duckdb/DuckDBVectorStore.py
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
def retrieve(self, query: str, top_k: int = 5) -> List[Document]:
    query_embedding = self._embedder.transform([query])[0].to_numpy().tolist()
    select_query = f"""
        SELECT id, content, metadata, embedding
        FROM {self.table_name}
    """

    if self.database_name == ":memory:":
        results = self._conn.execute(select_query).fetchall()
    else:
        with duckdb.connect(self._database_path) as conn:
            results = conn.execute(select_query).fetchall()

    # Calculate cosine similarities
    similarities = [
        (
            row[0],
            row[1],
            row[2],
            row[3],
            self._cosine_similarity(query_embedding, row[3]),
        )
        for row in results
    ]

    # Get top-k results sorted by similarity
    top_results = sorted(similarities, key=lambda x: x[4], reverse=True)[:top_k]

    return [
        Document(
            id=row[0],
            content=row[1],
            metadata=json.loads(row[2]),
            embedding=Vector(value=row[3]),
        )
        for row in top_results
    ]

delete_document

delete_document(id)
Source code in swarmauri_vectorstore_duckdb/DuckDBVectorStore.py
225
226
227
228
229
230
231
def delete_document(self, id: str) -> None:
    query = f"DELETE FROM {self.table_name} WHERE id = ?"
    if self.database_name == ":memory:":
        self._conn.execute(query, [id])
    else:
        with duckdb.connect(self._database_path) as conn:
            conn.execute(query, [id])

get_all_documents

get_all_documents()
Source code in swarmauri_vectorstore_duckdb/DuckDBVectorStore.py
233
234
235
236
237
238
239
240
241
242
243
244
def get_all_documents(self) -> List[Document]:
    query = f"SELECT id, content, metadata FROM {self.table_name}"
    if self.database_name == ":memory:":
        results = self._conn.execute(query).fetchall()
    else:
        with duckdb.connect(self._database_path) as conn:
            results = conn.execute(query).fetchall()

    return [
        Document(id=row[0], content=row[1], metadata=json.loads(row[2]))
        for row in results
    ]

update_document

update_document(id, new_document)

Update an existing document in the DuckDB store.

Source code in swarmauri_vectorstore_duckdb/DuckDBVectorStore.py
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
def update_document(self, id: str, new_document: Document) -> None:
    """Update an existing document in the DuckDB store."""
    try:
        data = self._prepare_document(new_document)
        query = f"""
            UPDATE {self.table_name}
            SET content = ?, embedding = ?, metadata = ?
            WHERE id = ?
        """
        if self.database_name == ":memory:":
            self._conn.execute(
                query, [data["content"], data["embedding"], data["metadata"], id]
            )
        else:
            with duckdb.connect(self._database_path) as conn:
                conn.execute(
                    query,
                    [data["content"], data["embedding"], data["metadata"], id],
                )
    except Exception as e:
        raise RuntimeError(f"Failed to update document {id}: {str(e)}")

from_local classmethod

from_local(database_path, table_name='documents', **kwargs)

Load a DuckDBVectorStore from a local file.

Source code in swarmauri_vectorstore_duckdb/DuckDBVectorStore.py
268
269
270
271
272
273
274
275
276
277
278
@classmethod
def from_local(cls, database_path: str, table_name: str = "documents", **kwargs):
    """Load a DuckDBVectorStore from a local file."""
    database_name = os.path.basename(database_path)
    persist_dir = os.path.dirname(database_path)
    return cls(
        database_name=database_name,
        table_name=table_name,
        persist_dir=persist_dir,
        **kwargs,
    )

model_dump_json

model_dump_json(*args, **kwargs)

Override model_dump_json to ensure connection is closed before serialization.

Source code in swarmauri_vectorstore_duckdb/DuckDBVectorStore.py
280
281
282
283
def model_dump_json(self, *args, **kwargs) -> str:
    """Override model_dump_json to ensure connection is closed before serialization."""
    self.disconnect()
    return super().model_dump_json(*args, **kwargs)

register_model classmethod

register_model()

Decorator to register a base model in the unified registry.

RETURNS DESCRIPTION
Callable

A decorator function that registers the model class.

TYPE: Callable[[Type[BaseModel]], Type[BaseModel]]

Source code in swarmauri_base/DynamicBase.py
562
563
564
565
566
567
568
569
570
571
572
573
574
575
576
577
578
579
580
581
582
583
584
585
@classmethod
def register_model(cls) -> Callable[[Type[BaseModel]], Type[BaseModel]]:
    """
    Decorator to register a base model in the unified registry.

    Returns:
        Callable: A decorator function that registers the model class.
    """

    def decorator(model_cls: Type[BaseModel]):
        """Register ``model_cls`` as a base model."""
        model_name = model_cls.__name__
        if model_name in cls._registry:
            glogger.warning(
                "Model '%s' is already registered; skipping duplicate.", model_name
            )
            return model_cls

        cls._registry[model_name] = {"model_cls": model_cls, "subtypes": {}}
        glogger.debug("Registered base model '%s'.", model_name)
        DynamicBase._recreate_models()
        return model_cls

    return decorator

register_type classmethod

register_type(resource_type=None, type_name=None)

Decorator to register a subtype under one or more base models in the unified registry.

PARAMETER DESCRIPTION
resource_type

The base model(s) under which to register the subtype. If None, all direct base classes (except DynamicBase) are used.

TYPE: Optional[Union[Type[T], List[Type[T]]]] DEFAULT: None

type_name

An optional custom type name for the subtype.

TYPE: Optional[str] DEFAULT: None

RETURNS DESCRIPTION
Callable

A decorator function that registers the subtype.

TYPE: Callable[[Type[DynamicBase]], Type[DynamicBase]]

Source code in swarmauri_base/DynamicBase.py
587
588
589
590
591
592
593
594
595
596
597
598
599
600
601
602
603
604
605
606
607
608
609
610
611
612
613
614
615
616
617
618
619
620
621
622
623
624
625
626
627
628
629
630
631
632
633
634
635
636
637
638
639
640
641
642
643
644
645
646
647
648
649
650
651
652
653
654
@classmethod
def register_type(
    cls,
    resource_type: Optional[Union[Type[T], List[Type[T]]]] = None,
    type_name: Optional[str] = None,
) -> Callable[[Type["DynamicBase"]], Type["DynamicBase"]]:
    """
    Decorator to register a subtype under one or more base models in the unified registry.

    Parameters:
        resource_type (Optional[Union[Type[T], List[Type[T]]]]):
            The base model(s) under which to register the subtype. If None, all direct base classes (except DynamicBase)
            are used.
        type_name (Optional[str]): An optional custom type name for the subtype.

    Returns:
        Callable: A decorator function that registers the subtype.
    """

    def decorator(subclass: Type["DynamicBase"]):
        """Register ``subclass`` as a subtype."""
        if resource_type is None:
            resource_types = [
                base for base in subclass.__bases__ if base is not cls
            ]
        elif not isinstance(resource_type, list):
            resource_types = [resource_type]
        else:
            resource_types = resource_type

        for rt in resource_types:
            if not issubclass(subclass, rt):
                raise TypeError(
                    f"'{subclass.__name__}' must be a subclass of '{rt.__name__}'."
                )
            final_type_name = type_name or getattr(
                subclass, "_type", subclass.__name__
            )
            base_model_name = rt.__name__

            if base_model_name not in cls._registry:
                cls._registry[base_model_name] = {"model_cls": rt, "subtypes": {}}
                glogger.debug(
                    "Created new registry entry for base model '%s'.",
                    base_model_name,
                )

            subtypes_dict = cls._registry[base_model_name]["subtypes"]
            if final_type_name in subtypes_dict:
                glogger.warning(
                    "Type '%s' already exists under '%s'; skipping duplicate.",
                    final_type_name,
                    base_model_name,
                )
                continue

            subtypes_dict[final_type_name] = subclass
            glogger.debug(
                "Registered '%s' as '%s' under '%s'.",
                subclass.__name__,
                final_type_name,
                base_model_name,
            )

        DynamicBase._recreate_models()
        return subclass

    return decorator

model_validate_toml classmethod

model_validate_toml(toml_data)

Validate a model from a TOML string.

Source code in swarmauri_base/TomlMixin.py
12
13
14
15
16
17
18
19
20
21
22
23
24
@classmethod
def model_validate_toml(cls, toml_data: str):
    """Validate a model from a TOML string."""
    try:
        # Parse TOML into a Python dictionary
        toml_content = tomllib.loads(toml_data)

        # Convert the dictionary to JSON and validate using Pydantic
        return cls.model_validate_json(json.dumps(toml_content))
    except tomllib.TOMLDecodeError as e:
        raise ValueError(f"Invalid TOML data: {e}")
    except ValidationError as e:
        raise ValueError(f"Validation failed: {e}")

model_dump_toml

model_dump_toml(
    fields_to_exclude=None, api_key_placeholder=None
)

Return a TOML representation of the model.

Source code in swarmauri_base/TomlMixin.py
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
def model_dump_toml(self, fields_to_exclude=None, api_key_placeholder=None):
    """Return a TOML representation of the model."""
    if fields_to_exclude is None:
        fields_to_exclude = []

    # Load the JSON string into a Python dictionary
    json_data = json.loads(self.model_dump_json())

    # Function to recursively remove specific keys and handle api_key placeholders
    def process_fields(data, fields_to_exclude):
        """Recursively filter fields and apply placeholders."""
        if isinstance(data, dict):
            return {
                key: (
                    api_key_placeholder
                    if key == "api_key" and api_key_placeholder is not None
                    else process_fields(value, fields_to_exclude)
                )
                for key, value in data.items()
                if key not in fields_to_exclude
            }
        elif isinstance(data, list):
            return [process_fields(item, fields_to_exclude) for item in data]
        else:
            return data

    # Filter the JSON data
    filtered_data = process_fields(json_data, fields_to_exclude)

    # Convert the filtered data into TOML
    return toml.dumps(filtered_data)

model_validate_yaml classmethod

model_validate_yaml(yaml_data)

Validate a model from a YAML string.

Source code in swarmauri_base/YamlMixin.py
11
12
13
14
15
16
17
18
19
20
21
22
23
@classmethod
def model_validate_yaml(cls, yaml_data: str):
    """Validate a model from a YAML string."""
    try:
        # Parse YAML into a Python dictionary
        yaml_content = yaml.safe_load(yaml_data)

        # Convert the dictionary to JSON and validate using Pydantic
        return cls.model_validate_json(json.dumps(yaml_content))
    except yaml.YAMLError as e:
        raise ValueError(f"Invalid YAML data: {e}")
    except ValidationError as e:
        raise ValueError(f"Validation failed: {e}")

model_dump_yaml

model_dump_yaml(
    fields_to_exclude=None, api_key_placeholder=None
)

Return a YAML representation of the model.

Source code in swarmauri_base/YamlMixin.py
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
def model_dump_yaml(self, fields_to_exclude=None, api_key_placeholder=None):
    """Return a YAML representation of the model."""
    if fields_to_exclude is None:
        fields_to_exclude = []

    # Load the JSON string into a Python dictionary
    json_data = json.loads(self.model_dump_json())

    # Function to recursively remove specific keys and handle api_key placeholders
    def process_fields(data, fields_to_exclude):
        """Recursively filter fields and apply placeholders."""
        if isinstance(data, dict):
            return {
                key: (
                    api_key_placeholder
                    if key == "api_key" and api_key_placeholder is not None
                    else process_fields(value, fields_to_exclude)
                )
                for key, value in data.items()
                if key not in fields_to_exclude
            }
        elif isinstance(data, list):
            return [process_fields(item, fields_to_exclude) for item in data]
        else:
            return data

    # Filter the JSON data
    filtered_data = process_fields(json_data, fields_to_exclude)

    # Convert the filtered data into YAML using safe mode
    return yaml.safe_dump(filtered_data, default_flow_style=False)

model_post_init

model_post_init(logger=None)

Assign a logger instance after model initialization.

Source code in swarmauri_base/LoggerMixin.py
23
24
25
26
27
28
def model_post_init(self, logger: Optional[FullUnion[LoggerBase]] = None) -> None:
    """Assign a logger instance after model initialization."""

    # Directly assign the provided FullUnion[LoggerBase] or fallback to the
    # class-level default.
    self.logger = self.logger or logger or self.default_logger

clear_documents

clear_documents()

Deletes all documents from the vector store

Source code in swarmauri_base/vector_stores/VectorStoreBase.py
94
95
96
97
98
99
def clear_documents(self) -> None:
    """
    Deletes all documents from the vector store

    """
    self.documents = []

document_count

document_count()
Source code in swarmauri_base/vector_stores/VectorStoreBase.py
101
102
def document_count(self):
    return len(self.documents)

document_dumps

document_dumps()

Placeholder

Source code in swarmauri_base/vector_stores/VectorStoreBase.py
104
105
106
107
108
def document_dumps(self) -> str:
    """
    Placeholder
    """
    return json.dumps([each.to_dict() for each in self.documents])

document_dump

document_dump(file_path)

Placeholder

Source code in swarmauri_base/vector_stores/VectorStoreBase.py
110
111
112
113
114
115
116
117
118
119
120
def document_dump(self, file_path: str) -> None:
    """
    Placeholder
    """
    with open(file_path, "w", encoding="utf-8") as f:
        json.dump(
            [each.to_dict() for each in self.documents],
            f,
            ensure_ascii=False,
            indent=4,
        )

document_loads

document_loads(json_data)

Placeholder

Source code in swarmauri_base/vector_stores/VectorStoreBase.py
122
123
124
125
126
127
128
def document_loads(self, json_data: str) -> None:
    """
    Placeholder
    """
    self.documents = [
        globals()[each["type"]].from_dict(each) for each in json.loads(json_data)
    ]

document_load

document_load(file_path)

Placeholder

Source code in swarmauri_base/vector_stores/VectorStoreBase.py
130
131
132
133
134
135
136
137
def document_load(self, file_path: str) -> None:
    """
    Placeholder
    """
    with open(file_path, "r", encoding="utf-8"):
        self.documents = [
            globals()[each["type"]].from_dict(each) for each in json.load(file_path)
        ]

save_store

save_store(directory_path)

Saves both the vectorizer's model and the documents.

Source code in swarmauri_base/vector_stores/VectorStoreSaveLoadMixin.py
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
def save_store(self, directory_path: str) -> None:
    """
    Saves both the vectorizer's model and the documents.
    """
    # Ensure the directory exists
    if not os.path.exists(directory_path):
        os.makedirs(directory_path)

    # Save the vectorizer model
    model_path = os.path.join(directory_path, "embedding_model")
    self._vectorizer.save_model(model_path)

    # Save documents
    documents_path = os.path.join(directory_path, "documents.json")
    with open(documents_path, "w", encoding="utf-8") as f:
        json.dump(
            [each.to_dict() for each in self.documents],
            f,
            ensure_ascii=False,
            indent=4,
        )

load_store

load_store(directory_path)

Loads both the vectorizer's model and the documents.

Source code in swarmauri_base/vector_stores/VectorStoreSaveLoadMixin.py
38
39
40
41
42
43
44
45
46
47
48
49
def load_store(self, directory_path: str) -> None:
    """
    Loads both the vectorizer's model and the documents.
    """
    # Load the vectorizer model
    model_path = os.path.join(directory_path, "embedding_model")
    self.vectorizer.load_model(model_path)

    # Load documents
    documents_path = os.path.join(directory_path, "documents.json")
    with open(documents_path, "r", encoding="utf-8") as f:
        self.documents = [self._load_document(each) for each in json.load(f)]

save_parts

save_parts(directory_path, chunk_size=10485760)

Splits the file into parts if it's too large and saves those parts individually.

Source code in swarmauri_base/vector_stores/VectorStoreSaveLoadMixin.py
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
def save_parts(self, directory_path: str, chunk_size: int = 10485760) -> None:
    """
    Splits the file into parts if it's too large and saves those parts individually.
    """
    file_number = 1
    model_path = os.path.join(directory_path, "embedding_model")
    parts_directory = os.path.join(directory_path, "parts")

    if not os.path.exists(parts_directory):
        os.makedirs(parts_directory)

    with open(f"{model_path}/model.safetensors", "rb") as f:
        chunk = f.read(chunk_size)
        while chunk:
            with open(
                f"{parts_directory}/model.safetensors.part{file_number:03}", "wb"
            ) as chunk_file:
                chunk_file.write(chunk)
            file_number += 1
            chunk = f.read(chunk_size)

    # Split the documents into parts and save them
    os.path.join(directory_path, "documents")

    self._split_json_file(directory_path, chunk_size=chunk_size)

load_parts

load_parts(directory_path, file_pattern='*.part*')

Combines file parts from a directory back into a single file and loads it.

Source code in swarmauri_base/vector_stores/VectorStoreSaveLoadMixin.py
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
def load_parts(self, directory_path: str, file_pattern: str = "*.part*") -> None:
    """
    Combines file parts from a directory back into a single file and loads it.
    """
    model_path = os.path.join(directory_path, "embedding_model")
    parts_directory = os.path.join(directory_path, "parts")
    output_file_path = os.path.join(model_path, "model.safetensors")

    parts = sorted(glob.glob(os.path.join(parts_directory, file_pattern)))
    with open(output_file_path, "wb") as output_file:
        for part in parts:
            with open(part, "rb") as file_part:
                output_file.write(file_part.read())

    # Load the combined_model now
    model_path = os.path.join(directory_path, "embedding_model")
    self._vectorizer.load_model(model_path)

    # Load document files
    self._load_documents(directory_path)