cognee 0.2.1.dev7__py3-none-any.whl → 0.2.2.dev1__py3-none-any.whl

cognee/infrastructure/databases/vector/lancedb/LanceDBAdapter.py

@@ -51,6 +51,7 @@ class LanceDBAdapter(VectorDBInterface):
         self.url = url
         self.api_key = api_key
         self.embedding_engine = embedding_engine
+        self.VECTOR_DB_LOCK = asyncio.Lock()
 
     async def get_connection(self):
         """
@@ -127,12 +128,14 @@ class LanceDBAdapter(VectorDBInterface):
             payload: payload_schema
 
         if not await self.has_collection(collection_name):
-            connection = await self.get_connection()
-            return await connection.create_table(
-                name=collection_name,
-                schema=LanceDataPoint,
-                exist_ok=True,
-            )
+            async with self.VECTOR_DB_LOCK:
+                if not await self.has_collection(collection_name):
+                    connection = await self.get_connection()
+                    return await connection.create_table(
+                        name=collection_name,
+                        schema=LanceDataPoint,
+                        exist_ok=True,
+                    )
 
     async def get_collection(self, collection_name: str):
         if not await self.has_collection(collection_name):
@@ -145,10 +148,12 @@ class LanceDBAdapter(VectorDBInterface):
         payload_schema = type(data_points[0])
 
         if not await self.has_collection(collection_name):
-            await self.create_collection(
-                collection_name,
-                payload_schema,
-            )
+            async with self.VECTOR_DB_LOCK:
+                if not await self.has_collection(collection_name):
+                    await self.create_collection(
+                        collection_name,
+                        payload_schema,
+                    )
 
         collection = await self.get_collection(collection_name)
 
@@ -188,12 +193,13 @@ class LanceDBAdapter(VectorDBInterface):
             for (data_point_index, data_point) in enumerate(data_points)
         ]
 
-        await (
-            collection.merge_insert("id")
-            .when_matched_update_all()
-            .when_not_matched_insert_all()
-            .execute(lance_data_points)
-        )
+        async with self.VECTOR_DB_LOCK:
+            await (
+                collection.merge_insert("id")
+                .when_matched_update_all()
+                .when_not_matched_insert_all()
+                .execute(lance_data_points)
+            )
 
     async def retrieve(self, collection_name: str, data_point_ids: list[str]):
         collection = await self.get_collection(collection_name)

cognee/infrastructure/databases/vector/pgvector/PGVectorAdapter.py

@@ -54,6 +54,7 @@ class PGVectorAdapter(SQLAlchemyAdapter, VectorDBInterface):
         self.api_key = api_key
         self.embedding_engine = embedding_engine
         self.db_uri: str = connection_string
+        self.VECTOR_DB_LOCK = asyncio.Lock()
 
         relational_db = get_relational_engine()
 
@@ -124,40 +125,41 @@ class PGVectorAdapter(SQLAlchemyAdapter, VectorDBInterface):
         data_point_types = get_type_hints(DataPoint)
         vector_size = self.embedding_engine.get_vector_size()
 
-        if not await self.has_collection(collection_name):
-
-            class PGVectorDataPoint(Base):
-                """
-                Represent a point in a vector data space with associated data and vector representation.
-
-                This class inherits from Base and is associated with a database table defined by
-                __tablename__. It maintains the following public methods and instance variables:
-
-                - __init__(self, id, payload, vector): Initializes a new PGVectorDataPoint instance.
-
-                Instance variables:
-                - id: Identifier for the data point, defined by data_point_types.
-                - payload: JSON data associated with the data point.
-                - vector: Vector representation of the data point, with size defined by vector_size.
-                """
-
-                __tablename__ = collection_name
-                __table_args__ = {"extend_existing": True}
-                # PGVector requires one column to be the primary key
-                id: Mapped[data_point_types["id"]] = mapped_column(primary_key=True)
-                payload = Column(JSON)
-                vector = Column(self.Vector(vector_size))
-
-                def __init__(self, id, payload, vector):
-                    self.id = id
-                    self.payload = payload
-                    self.vector = vector
-
-            async with self.engine.begin() as connection:
-                if len(Base.metadata.tables.keys()) > 0:
-                    await connection.run_sync(
-                        Base.metadata.create_all, tables=[PGVectorDataPoint.__table__]
-                    )
+        async with self.VECTOR_DB_LOCK:
+            if not await self.has_collection(collection_name):
+
+                class PGVectorDataPoint(Base):
+                    """
+                    Represent a point in a vector data space with associated data and vector representation.
+
+                    This class inherits from Base and is associated with a database table defined by
+                    __tablename__. It maintains the following public methods and instance variables:
+
+                    - __init__(self, id, payload, vector): Initializes a new PGVectorDataPoint instance.
+
+                    Instance variables:
+                    - id: Identifier for the data point, defined by data_point_types.
+                    - payload: JSON data associated with the data point.
+                    - vector: Vector representation of the data point, with size defined by vector_size.
+                    """
+
+                    __tablename__ = collection_name
+                    __table_args__ = {"extend_existing": True}
+                    # PGVector requires one column to be the primary key
+                    id: Mapped[data_point_types["id"]] = mapped_column(primary_key=True)
+                    payload = Column(JSON)
+                    vector = Column(self.Vector(vector_size))
+
+                    def __init__(self, id, payload, vector):
+                        self.id = id
+                        self.payload = payload
+                        self.vector = vector
+
+                async with self.engine.begin() as connection:
+                    if len(Base.metadata.tables.keys()) > 0:
+                        await connection.run_sync(
+                            Base.metadata.create_all, tables=[PGVectorDataPoint.__table__]
+                        )
 
     @retry(
         retry=retry_if_exception_type(DeadlockDetectedError),

cognee/infrastructure/databases/vector/vector_db_interface.py

@@ -1,4 +1,4 @@
-from typing import List, Protocol, Optional
+from typing import List, Protocol, Optional, Union, Any
 from abc import abstractmethod
 from cognee.infrastructure.engine import DataPoint
 from .models.PayloadSchema import PayloadSchema
@@ -31,7 +31,7 @@ class VectorDBInterface(Protocol):
     async def create_collection(
         self,
         collection_name: str,
-        payload_schema: Optional[PayloadSchema] = None,
+        payload_schema: Optional[Any] = None,
     ):
         """
         Create a new collection with an optional payload schema.
@@ -40,8 +40,8 @@ class VectorDBInterface(Protocol):
         -----------
 
             - collection_name (str): The name of the new collection to create.
-            - payload_schema (Optional[PayloadSchema]): An optional schema for the payloads
-              within this collection. (default None)
+            - payload_schema (Optional[Any]): An optional schema for the payloads
+              within this collection. Can be PayloadSchema, BaseModel, or other schema types. (default None)
         """
         raise NotImplementedError
 
@@ -71,7 +71,7 @@ class VectorDBInterface(Protocol):
 
             - collection_name (str): The name of the collection from which to retrieve data
               points.
-            - data_point_ids (list[str]): A list of IDs of the data points to retrieve.
+            - data_point_ids (Union[List[str], list[str]]): A list of IDs of the data points to retrieve.
         """
         raise NotImplementedError
 
@@ -123,7 +123,9 @@ class VectorDBInterface(Protocol):
         raise NotImplementedError
 
     @abstractmethod
-    async def delete_data_points(self, collection_name: str, data_point_ids: list[str]):
+    async def delete_data_points(
+        self, collection_name: str, data_point_ids: Union[List[str], list[str]]
+    ):
         """
         Delete specified data points from a collection.
 
@@ -132,7 +134,7 @@ class VectorDBInterface(Protocol):
 
             - collection_name (str): The name of the collection from which to delete data
               points.
-            - data_point_ids (list[str]): A list of IDs of the data points to delete.
+            - data_point_ids (Union[List[str], list[str]]): A list of IDs of the data points to delete.
         """
         raise NotImplementedError
 
@@ -142,3 +144,72 @@ class VectorDBInterface(Protocol):
         Remove obsolete or unnecessary data from the database.
         """
         raise NotImplementedError
+
+    @abstractmethod
+    async def embed_data(self, data: List[str]) -> List[List[float]]:
+        """
+        Embed textual data into vector representations.
+
+        Parameters:
+        -----------
+
+            - data (List[str]): A list of strings to be embedded.
+
+        Returns:
+        --------
+
+            - List[List[float]]: A list of embedded vectors corresponding to the input data.
+        """
+        raise NotImplementedError
+
+    # Optional methods that may be implemented by adapters
+    async def get_connection(self):
+        """
+        Get a connection to the vector database.
+        This method is optional and may return None for adapters that don't use connections.
+        """
+        return None
+
+    async def get_collection(self, collection_name: str):
+        """
+        Get a collection object from the vector database.
+        This method is optional and may return None for adapters that don't expose collection objects.
+        """
+        return None
+
+    async def create_vector_index(self, index_name: str, index_property_name: str):
+        """
+        Create a vector index for improved search performance.
+        This method is optional and may be a no-op for adapters that don't support indexing.
+        """
+        pass
+
+    async def index_data_points(
+        self, index_name: str, index_property_name: str, data_points: List[DataPoint]
+    ):
+        """
+        Index data points for improved search performance.
+        This method is optional and may be a no-op for adapters that don't support separate indexing.
+
+        Parameters:
+        -----------
+            - index_name (str): Name of the index to create/update
+            - index_property_name (str): Property name to index on
+            - data_points (List[DataPoint]): Data points to index
+        """
+        pass
+
+    def get_data_point_schema(self, model_type: Any) -> Any:
+        """
+        Get or transform a data point schema for the specific vector database.
+        This method is optional and may return the input unchanged for simple adapters.
+
+        Parameters:
+        -----------
+            - model_type (Any): The model type to get schema for
+
+        Returns:
+        --------
+            - Any: The schema object suitable for this vector database
+        """
+        return model_type

cognee/infrastructure/files/utils/get_data_file_path.py

@@ -0,0 +1,39 @@
+import os
+from urllib.parse import urlparse
+
+
+def get_data_file_path(file_path: str):
+    # Check if this is a file URI BEFORE normalizing (which corrupts URIs)
+    if file_path.startswith("file://"):
+        # Normalize the file URI for Windows - replace backslashes with forward slashes
+        normalized_file_uri = os.path.normpath(file_path)
+
+        parsed_url = urlparse(normalized_file_uri)
+
+        # Convert URI path to file system path
+        if os.name == "nt":  # Windows
+            # Handle Windows drive letters correctly
+            fs_path = parsed_url.path
+            if fs_path.startswith("/") and len(fs_path) > 1 and fs_path[2] == ":":
+                fs_path = fs_path[1:]  # Remove leading slash for Windows drive paths
+        else:  # Unix-like systems
+            fs_path = parsed_url.path
+
+        # Now split the actual filesystem path
+        actual_fs_path = os.path.normpath(fs_path)
+        return actual_fs_path
+
+    elif file_path.startswith("s3://"):
+        # Handle S3 URLs without normalization (which corrupts them)
+        parsed_url = urlparse(file_path)
+
+        normalized_url = (
+            f"s3://{parsed_url.netloc}{os.sep}{os.path.normpath(parsed_url.path).lstrip(os.sep)}"
+        )
+
+        return normalized_url
+
+    else:
+        # Regular file path - normalize separators
+        normalized_path = os.path.normpath(file_path)
+        return normalized_path

cognee/infrastructure/files/utils/guess_file_type.py

@@ -109,8 +109,8 @@ def guess_file_type(file: BinaryIO) -> filetype.Type:
     """
     Guess the file type from the given binary file stream.
 
-    If the file type cannot be determined, raise a FileTypeException with an appropriate
-    message.
+    If the file type cannot be determined from content, attempts to infer from extension.
+    If still unable to determine, raise a FileTypeException with an appropriate message.
 
     Parameters:
     -----------

cognee/infrastructure/files/utils/open_data_file.py

@@ -3,6 +3,7 @@ from os import path
 from urllib.parse import urlparse
 from contextlib import asynccontextmanager
 
+from cognee.infrastructure.files.utils.get_data_file_path import get_data_file_path
 from cognee.infrastructure.files.storage.S3FileStorage import S3FileStorage
 from cognee.infrastructure.files.storage.LocalFileStorage import LocalFileStorage
 
@@ -11,22 +12,8 @@ from cognee.infrastructure.files.storage.LocalFileStorage import LocalFileStorag
 async def open_data_file(file_path: str, mode: str = "rb", encoding: str = None, **kwargs):
     # Check if this is a file URI BEFORE normalizing (which corrupts URIs)
     if file_path.startswith("file://"):
-        # Normalize the file URI for Windows - replace backslashes with forward slashes
-        normalized_file_uri = os.path.normpath(file_path)
-
-        parsed_url = urlparse(normalized_file_uri)
-
-        # Convert URI path to file system path
-        if os.name == "nt":  # Windows
-            # Handle Windows drive letters correctly
-            fs_path = parsed_url.path
-            if fs_path.startswith("/") and len(fs_path) > 1 and fs_path[2] == ":":
-                fs_path = fs_path[1:]  # Remove leading slash for Windows drive paths
-        else:  # Unix-like systems
-            fs_path = parsed_url.path
-
         # Now split the actual filesystem path
-        actual_fs_path = os.path.normpath(fs_path)
+        actual_fs_path = get_data_file_path(file_path)
         file_dir_path = path.dirname(actual_fs_path)
         file_name = path.basename(actual_fs_path)
 
@@ -36,13 +23,7 @@ async def open_data_file(file_path: str, mode: str = "rb", encoding: str = None,
             yield file
 
     elif file_path.startswith("s3://"):
-        # Handle S3 URLs without normalization (which corrupts them)
-        parsed_url = urlparse(file_path)
-
-        normalized_url = (
-            f"s3://{parsed_url.netloc}{os.sep}{os.path.normpath(parsed_url.path).lstrip(os.sep)}"
-        )
-
+        normalized_url = get_data_file_path(file_path)
         s3_dir_path = os.path.dirname(normalized_url)
         s3_filename = os.path.basename(normalized_url)
 
@@ -66,7 +47,7 @@ async def open_data_file(file_path: str, mode: str = "rb", encoding: str = None,
 
     else:
         # Regular file path - normalize separators
-        normalized_path = os.path.normpath(file_path)
+        normalized_path = get_data_file_path(file_path)
         file_dir_path = path.dirname(normalized_path)
         file_name = path.basename(normalized_path)
 

cognee/infrastructure/llm/LLMGateway.py

@@ -0,0 +1,137 @@
+from typing import Type
+from pydantic import BaseModel
+from typing import Coroutine
+from cognee.infrastructure.llm import get_llm_config
+
+
+class LLMGateway:
+    """
+    Class handles selection of structured output frameworks and LLM functions.
+    Class used as a namespace for LLM related functions, should not be instantiated, all methods are static.
+    """
+
+    @staticmethod
+    def render_prompt(filename: str, context: dict, base_directory: str = None):
+        from cognee.infrastructure.llm.prompts import render_prompt
+
+        return render_prompt(filename=filename, context=context, base_directory=base_directory)
+
+    @staticmethod
+    def acreate_structured_output(
+        text_input: str, system_prompt: str, response_model: Type[BaseModel]
+    ) -> Coroutine:
+        from cognee.infrastructure.llm.structured_output_framework.litellm_instructor.llm.get_llm_client import (
+            get_llm_client,
+        )
+
+        llm_client = get_llm_client()
+        return llm_client.acreate_structured_output(
+            text_input=text_input, system_prompt=system_prompt, response_model=response_model
+        )
+
+    @staticmethod
+    def create_structured_output(
+        text_input: str, system_prompt: str, response_model: Type[BaseModel]
+    ) -> BaseModel:
+        from cognee.infrastructure.llm.structured_output_framework.litellm_instructor.llm.get_llm_client import (
+            get_llm_client,
+        )
+
+        llm_client = get_llm_client()
+        return llm_client.create_structured_output(
+            text_input=text_input, system_prompt=system_prompt, response_model=response_model
+        )
+
+    @staticmethod
+    def create_transcript(input) -> Coroutine:
+        from cognee.infrastructure.llm.structured_output_framework.litellm_instructor.llm.get_llm_client import (
+            get_llm_client,
+        )
+
+        llm_client = get_llm_client()
+        return llm_client.create_transcript(input=input)
+
+    @staticmethod
+    def transcribe_image(input) -> Coroutine:
+        from cognee.infrastructure.llm.structured_output_framework.litellm_instructor.llm.get_llm_client import (
+            get_llm_client,
+        )
+
+        llm_client = get_llm_client()
+        return llm_client.transcribe_image(input=input)
+
+    @staticmethod
+    def show_prompt(text_input: str, system_prompt: str) -> str:
+        from cognee.infrastructure.llm.structured_output_framework.litellm_instructor.llm.get_llm_client import (
+            get_llm_client,
+        )
+
+        llm_client = get_llm_client()
+        return llm_client.show_prompt(text_input=text_input, system_prompt=system_prompt)
+
+    @staticmethod
+    def read_query_prompt(prompt_file_name: str, base_directory: str = None):
+        from cognee.infrastructure.llm.prompts import (
+            read_query_prompt,
+        )
+
+        return read_query_prompt(prompt_file_name=prompt_file_name, base_directory=base_directory)
+
+    @staticmethod
+    def extract_content_graph(
+        content: str, response_model: Type[BaseModel], mode: str = "simple"
+    ) -> Coroutine:
+        llm_config = get_llm_config()
+        if llm_config.structured_output_framework.upper() == "BAML":
+            from cognee.infrastructure.llm.structured_output_framework.baml.baml_src.extraction import (
+                extract_content_graph,
+            )
+
+            return extract_content_graph(content=content, response_model=response_model, mode=mode)
+        else:
+            from cognee.infrastructure.llm.structured_output_framework.litellm_instructor.extraction import (
+                extract_content_graph,
+            )
+
+            return extract_content_graph(content=content, response_model=response_model)
+
+    @staticmethod
+    def extract_categories(content: str, response_model: Type[BaseModel]) -> Coroutine:
+        # TODO: Add BAML version of category and extraction and update function
+        from cognee.infrastructure.llm.structured_output_framework.litellm_instructor.extraction import (
+            extract_categories,
+        )
+
+        return extract_categories(content=content, response_model=response_model)
+
+    @staticmethod
+    def extract_code_summary(content: str) -> Coroutine:
+        llm_config = get_llm_config()
+        if llm_config.structured_output_framework.upper() == "BAML":
+            from cognee.infrastructure.llm.structured_output_framework.baml.baml_src.extraction import (
+                extract_code_summary,
+            )
+
+            return extract_code_summary(content=content)
+        else:
+            from cognee.infrastructure.llm.structured_output_framework.litellm_instructor.extraction import (
+                extract_code_summary,
+            )
+
+            return extract_code_summary(content=content)
+
+    @staticmethod
+    def extract_summary(content: str, response_model: Type[BaseModel]) -> Coroutine:
+        llm_config = get_llm_config()
+        if llm_config.structured_output_framework.upper() == "BAML":
+            from cognee.infrastructure.llm.structured_output_framework.baml.baml_src.extraction import (
+                extract_summary,
+            )
+
+            return extract_summary(content=content, response_model=response_model)
+        else:
+            from cognee.infrastructure.llm.structured_output_framework.litellm_instructor.extraction import (
+                extract_summary,
+            )
+
+            return extract_summary(content=content, response_model=response_model)

cognee/infrastructure/llm/__init__.py

@@ -1,4 +1,14 @@
-from .config import get_llm_config
-from .utils import get_max_chunk_tokens
-from .utils import test_llm_connection
-from .utils import test_embedding_connection
+from cognee.infrastructure.llm.config import (
+    get_llm_config,
+)
+from cognee.infrastructure.llm.utils import (
+    get_max_chunk_tokens,
+)
+from cognee.infrastructure.llm.utils import (
+    test_llm_connection,
+)
+from cognee.infrastructure.llm.utils import (
+    test_embedding_connection,
+)
+
+from cognee.infrastructure.llm.LLMGateway import LLMGateway

cognee/infrastructure/llm/config.py

@@ -1,8 +1,9 @@
 import os
-from typing import Optional
+from typing import Optional, ClassVar
 from functools import lru_cache
 from pydantic_settings import BaseSettings, SettingsConfigDict
 from pydantic import model_validator
+from baml_py import ClientRegistry
 
 
 class LLMConfig(BaseSettings):
@@ -32,6 +33,7 @@ class LLMConfig(BaseSettings):
     - to_dict
     """
 
+    structured_output_framework: str = "instructor"
     llm_provider: str = "openai"
     llm_model: str = "gpt-4o-mini"
     llm_endpoint: str = ""
@@ -40,6 +42,14 @@ class LLMConfig(BaseSettings):
     llm_temperature: float = 0.0
     llm_streaming: bool = False
     llm_max_tokens: int = 16384
+
+    baml_llm_provider: str = "openai"
+    baml_llm_model: str = "gpt-4o-mini"
+    baml_llm_endpoint: str = ""
+    baml_llm_api_key: Optional[str] = None
+    baml_llm_temperature: float = 0.0
+    baml_llm_api_version: str = ""
+
     transcription_model: str = "whisper-1"
     graph_prompt_path: str = "generate_graph_prompt.txt"
     llm_rate_limit_enabled: bool = False
@@ -53,8 +63,26 @@ class LLMConfig(BaseSettings):
     fallback_endpoint: str = ""
     fallback_model: str = ""
 
+    baml_registry: ClassVar[ClientRegistry] = ClientRegistry()
+
     model_config = SettingsConfigDict(env_file=".env", extra="allow")
 
+    def model_post_init(self, __context) -> None:
+        """Initialize the BAML registry after the model is created."""
+        self.baml_registry.add_llm_client(
+            name=self.baml_llm_provider,
+            provider=self.baml_llm_provider,
+            options={
+                "model": self.baml_llm_model,
+                "temperature": self.baml_llm_temperature,
+                "api_key": self.baml_llm_api_key,
+                "base_url": self.baml_llm_endpoint,
+                "api_version": self.baml_llm_api_version,
+            },
+        )
+        # Sets the primary client
+        self.baml_registry.set_primary(self.baml_llm_provider)
+
     @model_validator(mode="after")
     def ensure_env_vars_for_ollama(self) -> "LLMConfig":
         """

cognee/infrastructure/llm/prompts/answer_hotpot_question.txt

@@ -1,2 +1,2 @@
 Answer the question using the provided context. Be as brief as possible.
-Each entry in the context is a paragraph, which is represented as a list with two elements [title, sentences] and sentences is a list of strings.
+Each entry in the context is a paragraph, which is represented as a list with two elements [title, sentences] and sentences is a list of strings.

cognee/infrastructure/llm/prompts/answer_hotpot_using_cognee_search.txt

@@ -1,2 +1,2 @@
 Answer the question using the provided context. Be as brief as possible.
-Each entry in the context is tuple of length 3, representing an edge of a knowledge graph with its two nodes.
+Each entry in the context is tuple of length 3, representing an edge of a knowledge graph with its two nodes.

cognee/infrastructure/llm/prompts/answer_simple_question.txt

@@ -1 +1 @@
-Answer the question using the provided context. Be as brief as possible.
+Answer the question using the provided context. Be as brief as possible.

cognee/infrastructure/llm/prompts/answer_simple_question_restricted.txt

@@ -1 +1 @@
-Answer the question using the provided context. If the provided context is not connected to the question, just answer "The provided knowledge base does not contain the answer to the question". Be as brief as possible.
+Answer the question using the provided context. If the provided context is not connected to the question, just answer "The provided knowledge base does not contain the answer to the question". Be as brief as possible.

cognee/infrastructure/llm/prompts/categorize_categories.txt

@@ -1,2 +1,2 @@
 Chose the summary that is the most relevant to the query`{{ query }}`
-Here are the categories:`{{ categories }}`
+Here are the categories:`{{ categories }}`

cognee/infrastructure/llm/prompts/classify_content.txt

@@ -174,4 +174,4 @@ The possible classifications are:
             "Recipes and crafting instructions"
         ]
     }
-}
+}

cognee/infrastructure/llm/prompts/context_for_question.txt

@@ -1,2 +1,2 @@
 The question is: `{{ question }}`
-And here is the context: `{{ context }}`
+And here is the context: `{{ context }}`

cognee/infrastructure/llm/prompts/graph_context_for_question.txt

@@ -1,2 +1,2 @@
 The question is: `{{ question }}`
-and here is the context provided with a set of relationships from a knowledge graph separated by \n---\n each represented as node1 -- relation -- node2 triplet: `{{ context }}`
+and here is the context provided with a set of relationships from a knowledge graph separated by \n---\n each represented as node1 -- relation -- node2 triplet: `{{ context }}`

cognee/infrastructure/llm/prompts/natural_language_retriever_system.txt

@@ -63,4 +63,4 @@ This queries doesn't work. Do NOT use them:
 Example 1:
 Get all nodes connected to John
 MATCH (n:Entity {'name': 'John'})--(neighbor)
-RETURN n, neighbor
+RETURN n, neighbor

cognee/infrastructure/llm/prompts/patch_gen_instructions.txt

@@ -1,2 +1,2 @@
 I need you to solve this issue by generating a single patch file that I can apply directly to this repository using git apply.
-Please respond with a single patch file in the following format.
+Please respond with a single patch file in the following format.