Skip to content

Pinecone vector store driver

PineconeVectorStoreDriver

Bases: BaseVectorStoreDriver

Source code in griptape/drivers/vector/pinecone_vector_store_driver.py 
 11
 12
 13
 14
 15
 16
 17
 18
 19
 20
 21
 22
 23
 24
 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
 56
 57
 58
 59
 60
 61
 62
 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
 88
 89
 90
 91
 92
 93
 94
 95
 96
 97
 98
 99
100
@define
class PineconeVectorStoreDriver(BaseVectorStoreDriver):
    api_key: str = field(kw_only=True, metadata={"serializable": True})
    index_name: str = field(kw_only=True, metadata={"serializable": True})
    environment: str = field(kw_only=True, metadata={"serializable": True})
    project_name: Optional[str] = field(default=None, kw_only=True, metadata={"serializable": True})
    index: pinecone.Index = field(init=False)

    def __attrs_post_init__(self) -> None:
        pinecone = import_optional_dependency("pinecone").Pinecone(
            api_key=self.api_key, environment=self.environment, project_name=self.project_name
        )

        self.index = pinecone.Index(self.index_name)

    def upsert_vector(
        self,
        vector: list[float],
        vector_id: Optional[str] = None,
        namespace: Optional[str] = None,
        meta: Optional[dict] = None,
        **kwargs,
    ) -> str:
        vector_id = vector_id if vector_id else str_to_hash(str(vector))

        params: dict[str, Any] = {"namespace": namespace} | kwargs

        self.index.upsert(vectors=[(vector_id, vector, meta)], **params)

        return vector_id

    def load_entry(self, vector_id: str, namespace: Optional[str] = None) -> Optional[BaseVectorStoreDriver.Entry]:
        result = self.index.fetch(ids=[vector_id], namespace=namespace).to_dict()
        vectors = list(result["vectors"].values())

        if len(vectors) > 0:
            vector = vectors[0]

            return BaseVectorStoreDriver.Entry(
                id=vector["id"], meta=vector["metadata"], vector=vector["values"], namespace=result["namespace"]
            )
        else:
            return None

    def load_entries(self, namespace: Optional[str] = None) -> list[BaseVectorStoreDriver.Entry]:
        # This is a hacky way to query up to 10,000 values from Pinecone. Waiting on an official API for fetching
        # all values from a namespace:
        # https://community.pinecone.io/t/is-there-a-way-to-query-all-the-vectors-and-or-metadata-from-a-namespace/797/5

        results = self.index.query(
            vector=self.embedding_driver.embed_string(""), top_k=10000, include_metadata=True, namespace=namespace
        )

        return [
            BaseVectorStoreDriver.Entry(
                id=r["id"], vector=r["values"], meta=r["metadata"], namespace=results["namespace"]
            )
            for r in results["matches"]
        ]

    def query(
        self,
        query: str,
        count: Optional[int] = None,
        namespace: Optional[str] = None,
        include_vectors: bool = False,
        # PineconeVectorStorageDriver-specific params:
        include_metadata=True,
        **kwargs,
    ) -> list[BaseVectorStoreDriver.QueryResult]:
        vector = self.embedding_driver.embed_string(query)

        params = {
            "top_k": count if count else BaseVectorStoreDriver.DEFAULT_QUERY_COUNT,
            "namespace": namespace,
            "include_values": include_vectors,
            "include_metadata": include_metadata,
        } | kwargs

        results = self.index.query(vector=vector, **params)

        return [
            BaseVectorStoreDriver.QueryResult(
                id=r["id"], vector=r["values"], score=r["score"], meta=r["metadata"], namespace=results["namespace"]
            )
            for r in results["matches"]
        ]

    def delete_vector(self, vector_id: str):
        raise NotImplementedError(f"{self.__class__.__name__} does not support deletion.")

api_key: str = field(kw_only=True, metadata={'serializable': True}) class-attribute instance-attribute

environment: str = field(kw_only=True, metadata={'serializable': True}) class-attribute instance-attribute

index: pinecone.Index = field(init=False) class-attribute instance-attribute

index_name: str = field(kw_only=True, metadata={'serializable': True}) class-attribute instance-attribute

project_name: Optional[str] = field(default=None, kw_only=True, metadata={'serializable': True}) class-attribute instance-attribute

__attrs_post_init__()

Source code in griptape/drivers/vector/pinecone_vector_store_driver.py 
19
20
21
22
23
24
def __attrs_post_init__(self) -> None:
    pinecone = import_optional_dependency("pinecone").Pinecone(
        api_key=self.api_key, environment=self.environment, project_name=self.project_name
    )

    self.index = pinecone.Index(self.index_name)

delete_vector(vector_id)

Source code in griptape/drivers/vector/pinecone_vector_store_driver.py 
 99
100
def delete_vector(self, vector_id: str):
    raise NotImplementedError(f"{self.__class__.__name__} does not support deletion.")

load_entries(namespace=None)

Source code in griptape/drivers/vector/pinecone_vector_store_driver.py 
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
def load_entries(self, namespace: Optional[str] = None) -> list[BaseVectorStoreDriver.Entry]:
    # This is a hacky way to query up to 10,000 values from Pinecone. Waiting on an official API for fetching
    # all values from a namespace:
    # https://community.pinecone.io/t/is-there-a-way-to-query-all-the-vectors-and-or-metadata-from-a-namespace/797/5

    results = self.index.query(
        vector=self.embedding_driver.embed_string(""), top_k=10000, include_metadata=True, namespace=namespace
    )

    return [
        BaseVectorStoreDriver.Entry(
            id=r["id"], vector=r["values"], meta=r["metadata"], namespace=results["namespace"]
        )
        for r in results["matches"]
    ]

load_entry(vector_id, namespace=None)

Source code in griptape/drivers/vector/pinecone_vector_store_driver.py 
42
43
44
45
46
47
48
49
50
51
52
53
def load_entry(self, vector_id: str, namespace: Optional[str] = None) -> Optional[BaseVectorStoreDriver.Entry]:
    result = self.index.fetch(ids=[vector_id], namespace=namespace).to_dict()
    vectors = list(result["vectors"].values())

    if len(vectors) > 0:
        vector = vectors[0]

        return BaseVectorStoreDriver.Entry(
            id=vector["id"], meta=vector["metadata"], vector=vector["values"], namespace=result["namespace"]
        )
    else:
        return None

query(query, count=None, namespace=None, include_vectors=False, include_metadata=True, **kwargs)

Source code in griptape/drivers/vector/pinecone_vector_store_driver.py 
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
def query(
    self,
    query: str,
    count: Optional[int] = None,
    namespace: Optional[str] = None,
    include_vectors: bool = False,
    # PineconeVectorStorageDriver-specific params:
    include_metadata=True,
    **kwargs,
) -> list[BaseVectorStoreDriver.QueryResult]:
    vector = self.embedding_driver.embed_string(query)

    params = {
        "top_k": count if count else BaseVectorStoreDriver.DEFAULT_QUERY_COUNT,
        "namespace": namespace,
        "include_values": include_vectors,
        "include_metadata": include_metadata,
    } | kwargs

    results = self.index.query(vector=vector, **params)

    return [
        BaseVectorStoreDriver.QueryResult(
            id=r["id"], vector=r["values"], score=r["score"], meta=r["metadata"], namespace=results["namespace"]
        )
        for r in results["matches"]
    ]

upsert_vector(vector, vector_id=None, namespace=None, meta=None, **kwargs)

Source code in griptape/drivers/vector/pinecone_vector_store_driver.py 
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
def upsert_vector(
    self,
    vector: list[float],
    vector_id: Optional[str] = None,
    namespace: Optional[str] = None,
    meta: Optional[dict] = None,
    **kwargs,
) -> str:
    vector_id = vector_id if vector_id else str_to_hash(str(vector))

    params: dict[str, Any] = {"namespace": namespace} | kwargs

    self.index.upsert(vectors=[(vector_id, vector, meta)], **params)

    return vector_id