Merge branch 'dev' into multi-tenancy

.env.template

@@ -28,11 +28,10 @@ EMBEDDING_ENDPOINT=""
 EMBEDDING_API_VERSION=""
 EMBEDDING_DIMENSIONS=3072
 EMBEDDING_MAX_TOKENS=8191
+EMBEDDING_BATCH_SIZE=36
 # If embedding key is not provided same key set for LLM_API_KEY will be used
 #EMBEDDING_API_KEY="your_api_key"
-# Note: OpenAI support up to 2048 elements and Gemini supports a maximum of 100 elements in an embedding batch,
+
-#       Cognee sets the optimal batch size for OpenAI and Gemini, but a custom size can be defined if necessary for other models
-#EMBEDDING_BATCH_SIZE=2048
 
 # If using BAML structured output these env variables will be used
 BAML_LLM_PROVIDER=openai
@@ -248,10 +247,10 @@ LITELLM_LOG="ERROR"
 #LLM_PROVIDER="ollama"
 #LLM_ENDPOINT="http://localhost:11434/v1"
 #EMBEDDING_PROVIDER="ollama"
-#EMBEDDING_MODEL="avr/sfr-embedding-mistral:latest"
+#EMBEDDING_MODEL="nomic-embed-text:latest"
 #EMBEDDING_ENDPOINT="http://localhost:11434/api/embeddings"
-#EMBEDDING_DIMENSIONS=4096
+#EMBEDDING_DIMENSIONS=768
-#HUGGINGFACE_TOKENIZER="Salesforce/SFR-Embedding-Mistral"
+#HUGGINGFACE_TOKENIZER="nomic-ai/nomic-embed-text-v1.5"
 
 ########## OpenRouter (also free) #########################################################
 

.github/workflows/e2e_tests.yml

@@ -1,6 +1,4 @@
 name: Reusable Integration Tests
-permissions:
-  contents: read
 
 on:
   workflow_call:
@@ -267,8 +265,6 @@ jobs:
           EMBEDDING_API_VERSION: ${{ secrets.EMBEDDING_API_VERSION }}
         run: uv run python ./cognee/tests/test_edge_ingestion.py
 
-
-
   run_concurrent_subprocess_access_test:
     name: Concurrent Subprocess access test
     runs-on: ubuntu-latest
@@ -331,3 +327,31 @@ jobs:
           DB_USERNAME: cognee
           DB_PASSWORD: cognee
         run: uv run python ./cognee/tests/test_concurrent_subprocess_access.py
+
+  test-entity-extraction:
+    name: Test Entity Extraction
+    runs-on: ubuntu-22.04
+    steps:
+      - name: Check out repository
+        uses: actions/checkout@v4
+
+      - name: Cognee Setup
+        uses: ./.github/actions/cognee_setup
+        with:
+          python-version: '3.11.x'
+
+      - name: Dependencies already installed
+        run: echo "Dependencies already installed in setup"
+
+      - name: Run Entity Extraction Test
+        env:
+          ENV: 'dev'
+          LLM_MODEL: ${{ secrets.LLM_MODEL }}
+          LLM_ENDPOINT: ${{ secrets.LLM_ENDPOINT }}
+          LLM_API_KEY: ${{ secrets.LLM_API_KEY }}
+          LLM_API_VERSION: ${{ secrets.LLM_API_VERSION }}
+          EMBEDDING_MODEL: ${{ secrets.EMBEDDING_MODEL }}
+          EMBEDDING_ENDPOINT: ${{ secrets.EMBEDDING_ENDPOINT }}
+          EMBEDDING_API_KEY: ${{ secrets.EMBEDDING_API_KEY }}
+          EMBEDDING_API_VERSION: ${{ secrets.EMBEDDING_API_VERSION }}
+        run: uv run python ./cognee/tests/tasks/entity_extraction/entity_extraction_test.py

cognee/api/v1/cognify/cognify.py

@@ -44,6 +44,7 @@ async def cognify(
     graph_model: BaseModel = KnowledgeGraph,
     chunker=TextChunker,
     chunk_size: int = None,
+    chunks_per_batch: int = None,
     config: Config = None,
     vector_db_config: dict = None,
     graph_db_config: dict = None,
@@ -106,6 +107,7 @@ async def cognify(
                    Formula: min(embedding_max_completion_tokens, llm_max_completion_tokens // 2)
                    Default limits: ~512-8192 tokens depending on models.
                    Smaller chunks = more granular but potentially fragmented knowledge.
+        chunks_per_batch: Number of chunks to be processed in a single batch in Cognify tasks.
         vector_db_config: Custom vector database configuration for embeddings storage.
         graph_db_config: Custom graph database configuration for relationship storage.
         run_in_background: If True, starts processing asynchronously and returns immediately.
@@ -210,10 +212,18 @@ async def cognify(
             }
 
     if temporal_cognify:
-        tasks = await get_temporal_tasks(user, chunker, chunk_size)
+        tasks = await get_temporal_tasks(
+            user=user, chunker=chunker, chunk_size=chunk_size, chunks_per_batch=chunks_per_batch
+        )
     else:
         tasks = await get_default_tasks(
-            user, graph_model, chunker, chunk_size, config, custom_prompt
+            user=user,
+            graph_model=graph_model,
+            chunker=chunker,
+            chunk_size=chunk_size,
+            config=config,
+            custom_prompt=custom_prompt,
+            chunks_per_batch=chunks_per_batch,
         )
 
     # By calling get pipeline executor we get a function that will have the run_pipeline run in the background or a function that we will need to wait for
@@ -240,6 +250,7 @@ async def get_default_tasks(  # TODO: Find out a better way to do this (Boris's
     chunk_size: int = None,
     config: Config = None,
     custom_prompt: Optional[str] = None,
+    chunks_per_batch: int = 100,
 ) -> list[Task]:
     if config is None:
         ontology_config = get_ontology_env_config()
@@ -258,6 +269,9 @@ async def get_default_tasks(  # TODO: Find out a better way to do this (Boris's
                 "ontology_config": {"ontology_resolver": get_default_ontology_resolver()}
             }
 
+    if chunks_per_batch is None:
+        chunks_per_batch = 100
+
     default_tasks = [
         Task(classify_documents),
         Task(check_permissions_on_dataset, user=user, permissions=["write"]),
@@ -271,20 +285,20 @@ async def get_default_tasks(  # TODO: Find out a better way to do this (Boris's
             graph_model=graph_model,
             config=config,
             custom_prompt=custom_prompt,
-            task_config={"batch_size": 10},
+            task_config={"batch_size": chunks_per_batch},
         ),  # Generate knowledge graphs from the document chunks.
         Task(
             summarize_text,
-            task_config={"batch_size": 10},
+            task_config={"batch_size": chunks_per_batch},
         ),
-        Task(add_data_points, task_config={"batch_size": 10}),
+        Task(add_data_points, task_config={"batch_size": chunks_per_batch}),
     ]
 
     return default_tasks
 
 
 async def get_temporal_tasks(
-    user: User = None, chunker=TextChunker, chunk_size: int = None
+    user: User = None, chunker=TextChunker, chunk_size: int = None, chunks_per_batch: int = 10
 ) -> list[Task]:
     """
     Builds and returns a list of temporal processing tasks to be executed in sequence.
@@ -301,10 +315,14 @@ async def get_temporal_tasks(
         user (User, optional): The user requesting task execution, used for permission checks.
         chunker (Callable, optional): A text chunking function/class to split documents. Defaults to TextChunker.
         chunk_size (int, optional): Maximum token size per chunk. If not provided, uses system default.
+        chunks_per_batch (int, optional): Number of chunks to process in a single batch in Cognify
 
     Returns:
         list[Task]: A list of Task objects representing the temporal processing pipeline.
     """
+    if chunks_per_batch is None:
+        chunks_per_batch = 10
+
     temporal_tasks = [
         Task(classify_documents),
         Task(check_permissions_on_dataset, user=user, permissions=["write"]),
@@ -313,9 +331,9 @@ async def get_temporal_tasks(
             max_chunk_size=chunk_size or get_max_chunk_tokens(),
             chunker=chunker,
         ),
-        Task(extract_events_and_timestamps, task_config={"chunk_size": 10}),
+        Task(extract_events_and_timestamps, task_config={"batch_size": chunks_per_batch}),
         Task(extract_knowledge_graph_from_events),
-        Task(add_data_points, task_config={"batch_size": 10}),
+        Task(add_data_points, task_config={"batch_size": chunks_per_batch}),
     ]
 
     return temporal_tasks

cognee/infrastructure/databases/graph/neo4j_driver/adapter.py

@@ -1067,7 +1067,7 @@ class Neo4jAdapter(GraphDBInterface):
         query_nodes = f"""
         MATCH (n)
         WHERE {where_clause}
-        RETURN ID(n) AS id, labels(n) AS labels, properties(n) AS properties
+        RETURN n.id AS id, labels(n) AS labels, properties(n) AS properties
         """
         result_nodes = await self.query(query_nodes)
 
@@ -1082,7 +1082,7 @@ class Neo4jAdapter(GraphDBInterface):
         query_edges = f"""
         MATCH (n)-[r]->(m)
         WHERE {where_clause} AND {where_clause.replace("n.", "m.")}
-        RETURN ID(n) AS source, ID(m) AS target, TYPE(r) AS type, properties(r) AS properties
+        RETURN n.id AS source, n.id AS target, TYPE(r) AS type, properties(r) AS properties
         """
         result_edges = await self.query(query_edges)
 

cognee/infrastructure/databases/vector/embeddings/FastembedEmbeddingEngine.py

@@ -1,8 +1,17 @@
-from cognee.shared.logging_utils import get_logger
+import os
+import logging
 from typing import List, Optional
 from fastembed import TextEmbedding
 import litellm
-import os
+from tenacity import (
+    retry,
+    stop_after_delay,
+    wait_exponential_jitter,
+    retry_if_not_exception_type,
+    before_sleep_log,
+)
+
+from cognee.shared.logging_utils import get_logger
 from cognee.infrastructure.databases.vector.embeddings.EmbeddingEngine import EmbeddingEngine
 from cognee.infrastructure.databases.exceptions import EmbeddingException
 from cognee.infrastructure.llm.tokenizer.TikToken import (
@@ -57,6 +66,13 @@ class FastembedEmbeddingEngine(EmbeddingEngine):
             enable_mocking = str(enable_mocking).lower()
         self.mock = enable_mocking in ("true", "1", "yes")
 
+    @retry(
+        stop=stop_after_delay(128),
+        wait=wait_exponential_jitter(2, 128),
+        retry=retry_if_not_exception_type(litellm.exceptions.NotFoundError),
+        before_sleep=before_sleep_log(logger, logging.DEBUG),
+        reraise=True,
+    )
     async def embed_text(self, text: List[str]) -> List[List[float]]:
         """
         Embed the given text into numerical vectors.

cognee/infrastructure/databases/vector/embeddings/LiteLLMEmbeddingEngine.py

@@ -1,15 +1,21 @@
 import asyncio
+import logging
+
 from cognee.shared.logging_utils import get_logger
 from typing import List, Optional
 import numpy as np
 import math
+from tenacity import (
+    retry,
+    stop_after_delay,
+    wait_exponential_jitter,
+    retry_if_not_exception_type,
+    before_sleep_log,
+)
 import litellm
 import os
 from cognee.infrastructure.databases.vector.embeddings.EmbeddingEngine import EmbeddingEngine
 from cognee.infrastructure.databases.exceptions import EmbeddingException
-from cognee.infrastructure.llm.tokenizer.Gemini import (
-    GeminiTokenizer,
-)
 from cognee.infrastructure.llm.tokenizer.HuggingFace import (
     HuggingFaceTokenizer,
 )
@@ -19,10 +25,6 @@ from cognee.infrastructure.llm.tokenizer.Mistral import (
 from cognee.infrastructure.llm.tokenizer.TikToken import (
     TikTokenTokenizer,
 )
-from cognee.infrastructure.databases.vector.embeddings.embedding_rate_limiter import (
-    embedding_rate_limit_async,
-    embedding_sleep_and_retry_async,
-)
 
 litellm.set_verbose = False
 logger = get_logger("LiteLLMEmbeddingEngine")
@@ -76,8 +78,13 @@ class LiteLLMEmbeddingEngine(EmbeddingEngine):
             enable_mocking = str(enable_mocking).lower()
         self.mock = enable_mocking in ("true", "1", "yes")
 
-    @embedding_sleep_and_retry_async()
+    @retry(
-    @embedding_rate_limit_async
+        stop=stop_after_delay(128),
+        wait=wait_exponential_jitter(2, 128),
+        retry=retry_if_not_exception_type(litellm.exceptions.NotFoundError),
+        before_sleep=before_sleep_log(logger, logging.DEBUG),
+        reraise=True,
+    )
     async def embed_text(self, text: List[str]) -> List[List[float]]:
         """
         Embed a list of text strings into vector representations.

cognee/infrastructure/databases/vector/embeddings/OllamaEmbeddingEngine.py

@@ -3,8 +3,16 @@ from cognee.shared.logging_utils import get_logger
 import aiohttp
 from typing import List, Optional
 import os
-
+import litellm
+import logging
 import aiohttp.http_exceptions
+from tenacity import (
+    retry,
+    stop_after_delay,
+    wait_exponential_jitter,
+    retry_if_not_exception_type,
+    before_sleep_log,
+)
 
 from cognee.infrastructure.databases.vector.embeddings.EmbeddingEngine import EmbeddingEngine
 from cognee.infrastructure.llm.tokenizer.HuggingFace import (
@@ -69,7 +77,6 @@ class OllamaEmbeddingEngine(EmbeddingEngine):
             enable_mocking = str(enable_mocking).lower()
         self.mock = enable_mocking in ("true", "1", "yes")
 
-    @embedding_rate_limit_async
     async def embed_text(self, text: List[str]) -> List[List[float]]:
         """
         Generate embedding vectors for a list of text prompts.
@@ -92,7 +99,13 @@ class OllamaEmbeddingEngine(EmbeddingEngine):
         embeddings = await asyncio.gather(*[self._get_embedding(prompt) for prompt in text])
         return embeddings
 
-    @embedding_sleep_and_retry_async()
+    @retry(
+        stop=stop_after_delay(128),
+        wait=wait_exponential_jitter(2, 128),
+        retry=retry_if_not_exception_type(litellm.exceptions.NotFoundError),
+        before_sleep=before_sleep_log(logger, logging.DEBUG),
+        reraise=True,
+    )
     async def _get_embedding(self, prompt: str) -> List[float]:
         """
         Internal method to call the Ollama embeddings endpoint for a single prompt.

cognee/infrastructure/databases/vector/embeddings/config.py

@@ -24,11 +24,10 @@ class EmbeddingConfig(BaseSettings):
     model_config = SettingsConfigDict(env_file=".env", extra="allow")
 
     def model_post_init(self, __context) -> None:
-        # If embedding batch size is not defined use 2048 as default for OpenAI and 100 for all other embedding models
         if not self.embedding_batch_size and self.embedding_provider.lower() == "openai":
-            self.embedding_batch_size = 2048
+            self.embedding_batch_size = 36
         elif not self.embedding_batch_size:
-            self.embedding_batch_size = 100
+            self.embedding_batch_size = 36
 
     def to_dict(self) -> dict:
         """

cognee/infrastructure/files/utils/guess_file_type.py

@@ -124,6 +124,12 @@ def guess_file_type(file: BinaryIO) -> filetype.Type:
     """
     file_type = filetype.guess(file)
 
+    # If file type could not be determined consider it a plain text file as they don't have magic number encoding
+    if file_type is None:
+        from filetype.types.base import Type
+
+        file_type = Type("text/plain", "txt")
+
     if file_type is None:
         raise FileTypeException(f"Unknown file detected: {file.name}.")
 

cognee/infrastructure/llm/structured_output_framework/litellm_instructor/llm/anthropic/adapter.py

@@ -1,19 +1,24 @@
+import logging
 from typing import Type
 from pydantic import BaseModel
+import litellm
 import instructor
+from cognee.shared.logging_utils import get_logger
+from tenacity import (
+    retry,
+    stop_after_delay,
+    wait_exponential_jitter,
+    retry_if_not_exception_type,
+    before_sleep_log,
+)
 
-from cognee.infrastructure.llm.exceptions import MissingSystemPromptPathError
 from cognee.infrastructure.llm.structured_output_framework.litellm_instructor.llm.llm_interface import (
     LLMInterface,
 )
-from cognee.infrastructure.llm.structured_output_framework.litellm_instructor.llm.rate_limiter import (
-    rate_limit_async,
-    sleep_and_retry_async,
-)
-
-from cognee.infrastructure.llm.LLMGateway import LLMGateway
 from cognee.infrastructure.llm.config import get_llm_config
 
+logger = get_logger()
+
 
 class AnthropicAdapter(LLMInterface):
     """
@@ -35,8 +40,13 @@ class AnthropicAdapter(LLMInterface):
         self.model = model
         self.max_completion_tokens = max_completion_tokens
 
-    @sleep_and_retry_async()
+    @retry(
-    @rate_limit_async
+        stop=stop_after_delay(128),
+        wait=wait_exponential_jitter(2, 128),
+        retry=retry_if_not_exception_type(litellm.exceptions.NotFoundError),
+        before_sleep=before_sleep_log(logger, logging.DEBUG),
+        reraise=True,
+    )
     async def acreate_structured_output(
         self, text_input: str, system_prompt: str, response_model: Type[BaseModel]
     ) -> BaseModel:

cognee/infrastructure/llm/structured_output_framework/litellm_instructor/llm/gemini/adapter.py

@@ -12,11 +12,18 @@ from cognee.infrastructure.llm.exceptions import ContentPolicyFilterError
 from cognee.infrastructure.llm.structured_output_framework.litellm_instructor.llm.llm_interface import (
     LLMInterface,
 )
-from cognee.infrastructure.llm.structured_output_framework.litellm_instructor.llm.rate_limiter import (
+import logging
-    rate_limit_async,
+from cognee.shared.logging_utils import get_logger
-    sleep_and_retry_async,
+from tenacity import (
+    retry,
+    stop_after_delay,
+    wait_exponential_jitter,
+    retry_if_not_exception_type,
+    before_sleep_log,
 )
 
+logger = get_logger()
+
 
 class GeminiAdapter(LLMInterface):
     """
@@ -58,8 +65,13 @@ class GeminiAdapter(LLMInterface):
 
         self.aclient = instructor.from_litellm(litellm.acompletion, mode=instructor.Mode.JSON)
 
-    @sleep_and_retry_async()
+    @retry(
-    @rate_limit_async
+        stop=stop_after_delay(128),
+        wait=wait_exponential_jitter(2, 128),
+        retry=retry_if_not_exception_type(litellm.exceptions.NotFoundError),
+        before_sleep=before_sleep_log(logger, logging.DEBUG),
+        reraise=True,
+    )
     async def acreate_structured_output(
         self, text_input: str, system_prompt: str, response_model: Type[BaseModel]
     ) -> BaseModel:

cognee/infrastructure/llm/structured_output_framework/litellm_instructor/llm/generic_llm_api/adapter.py

@@ -12,11 +12,18 @@ from cognee.infrastructure.llm.exceptions import ContentPolicyFilterError
 from cognee.infrastructure.llm.structured_output_framework.litellm_instructor.llm.llm_interface import (
     LLMInterface,
 )
-from cognee.infrastructure.llm.structured_output_framework.litellm_instructor.llm.rate_limiter import (
+import logging
-    rate_limit_async,
+from cognee.shared.logging_utils import get_logger
-    sleep_and_retry_async,
+from tenacity import (
+    retry,
+    stop_after_delay,
+    wait_exponential_jitter,
+    retry_if_not_exception_type,
+    before_sleep_log,
 )
 
+logger = get_logger()
+
 
 class GenericAPIAdapter(LLMInterface):
     """
@@ -58,8 +65,13 @@ class GenericAPIAdapter(LLMInterface):
 
         self.aclient = instructor.from_litellm(litellm.acompletion, mode=instructor.Mode.JSON)
 
-    @sleep_and_retry_async()
+    @retry(
-    @rate_limit_async
+        stop=stop_after_delay(128),
+        wait=wait_exponential_jitter(2, 128),
+        retry=retry_if_not_exception_type(litellm.exceptions.NotFoundError),
+        before_sleep=before_sleep_log(logger, logging.DEBUG),
+        reraise=True,
+    )
     async def acreate_structured_output(
         self, text_input: str, system_prompt: str, response_model: Type[BaseModel]
     ) -> BaseModel:

cognee/infrastructure/llm/structured_output_framework/litellm_instructor/llm/mistral/adapter.py

@@ -1,20 +1,23 @@
 import litellm
 import instructor
 from pydantic import BaseModel
-from typing import Type, Optional
+from typing import Type
-from litellm import acompletion, JSONSchemaValidationError
+from litellm import JSONSchemaValidationError
 
 from cognee.shared.logging_utils import get_logger
 from cognee.modules.observability.get_observe import get_observe
-from cognee.infrastructure.llm.exceptions import MissingSystemPromptPathError
 from cognee.infrastructure.llm.structured_output_framework.litellm_instructor.llm.llm_interface import (
     LLMInterface,
 )
-from cognee.infrastructure.llm.LLMGateway import LLMGateway
 from cognee.infrastructure.llm.config import get_llm_config
-from cognee.infrastructure.llm.structured_output_framework.litellm_instructor.llm.rate_limiter import (
+
-    rate_limit_async,
+import logging
-    sleep_and_retry_async,
+from tenacity import (
+    retry,
+    stop_after_delay,
+    wait_exponential_jitter,
+    retry_if_not_exception_type,
+    before_sleep_log,
 )
 
 logger = get_logger()
@@ -47,8 +50,13 @@ class MistralAdapter(LLMInterface):
             api_key=get_llm_config().llm_api_key,
         )
 
-    @sleep_and_retry_async()
+    @retry(
-    @rate_limit_async
+        stop=stop_after_delay(128),
+        wait=wait_exponential_jitter(2, 128),
+        retry=retry_if_not_exception_type(litellm.exceptions.NotFoundError),
+        before_sleep=before_sleep_log(logger, logging.DEBUG),
+        reraise=True,
+    )
     async def acreate_structured_output(
         self, text_input: str, system_prompt: str, response_model: Type[BaseModel]
     ) -> BaseModel:
@@ -99,31 +107,3 @@ class MistralAdapter(LLMInterface):
             logger.error(f"Schema validation failed: {str(e)}")
             logger.debug(f"Raw response: {e.raw_response}")
             raise ValueError(f"Response failed schema validation: {str(e)}")
-
-    def show_prompt(self, text_input: str, system_prompt: str) -> str:
-        """
-        Format and display the prompt for a user query.
-
-        Parameters:
-        -----------
-            - text_input (str): Input text from the user to be included in the prompt.
-            - system_prompt (str): The system prompt that will be shown alongside the user input.
-
-        Returns:
-        --------
-            - str: The formatted prompt string combining system prompt and user input.
-        """
-        if not text_input:
-            text_input = "No user input provided."
-        if not system_prompt:
-            raise MissingSystemPromptPathError()
-
-        system_prompt = LLMGateway.read_query_prompt(system_prompt)
-
-        formatted_prompt = (
-            f"""System Prompt:\n{system_prompt}\n\nUser Input:\n{text_input}\n"""
-            if system_prompt
-            else None
-        )
-
-        return formatted_prompt

cognee/infrastructure/llm/structured_output_framework/litellm_instructor/llm/ollama/adapter.py

@@ -1,4 +1,6 @@
 import base64
+import litellm
+import logging
 import instructor
 from typing import Type
 from openai import OpenAI
@@ -7,11 +9,17 @@ from pydantic import BaseModel
 from cognee.infrastructure.llm.structured_output_framework.litellm_instructor.llm.llm_interface import (
     LLMInterface,
 )
-from cognee.infrastructure.llm.structured_output_framework.litellm_instructor.llm.rate_limiter import (
-    rate_limit_async,
-    sleep_and_retry_async,
-)
 from cognee.infrastructure.files.utils.open_data_file import open_data_file
+from cognee.shared.logging_utils import get_logger
+from tenacity import (
+    retry,
+    stop_after_delay,
+    wait_exponential_jitter,
+    retry_if_not_exception_type,
+    before_sleep_log,
+)
+
+logger = get_logger()
 
 
 class OllamaAPIAdapter(LLMInterface):
@@ -47,8 +55,13 @@ class OllamaAPIAdapter(LLMInterface):
             OpenAI(base_url=self.endpoint, api_key=self.api_key), mode=instructor.Mode.JSON
         )
 
-    @sleep_and_retry_async()
+    @retry(
-    @rate_limit_async
+        stop=stop_after_delay(128),
+        wait=wait_exponential_jitter(2, 128),
+        retry=retry_if_not_exception_type(litellm.exceptions.NotFoundError),
+        before_sleep=before_sleep_log(logger, logging.DEBUG),
+        reraise=True,
+    )
     async def acreate_structured_output(
         self, text_input: str, system_prompt: str, response_model: Type[BaseModel]
     ) -> BaseModel:
@@ -90,7 +103,13 @@ class OllamaAPIAdapter(LLMInterface):
 
         return response
 
-    @rate_limit_async
+    @retry(
+        stop=stop_after_delay(128),
+        wait=wait_exponential_jitter(2, 128),
+        retry=retry_if_not_exception_type(litellm.exceptions.NotFoundError),
+        before_sleep=before_sleep_log(logger, logging.DEBUG),
+        reraise=True,
+    )
     async def create_transcript(self, input_file: str) -> str:
         """
         Generate an audio transcript from a user query.
@@ -123,7 +142,13 @@ class OllamaAPIAdapter(LLMInterface):
 
         return transcription.text
 
-    @rate_limit_async
+    @retry(
+        stop=stop_after_delay(128),
+        wait=wait_exponential_jitter(2, 128),
+        retry=retry_if_not_exception_type(litellm.exceptions.NotFoundError),
+        before_sleep=before_sleep_log(logger, logging.DEBUG),
+        reraise=True,
+    )
     async def transcribe_image(self, input_file: str) -> str:
         """
         Transcribe content from an image using base64 encoding.

cognee/infrastructure/llm/structured_output_framework/litellm_instructor/llm/openai/adapter.py

@@ -7,6 +7,15 @@ from openai import ContentFilterFinishReasonError
 from litellm.exceptions import ContentPolicyViolationError
 from instructor.core import InstructorRetryException
 
+import logging
+from tenacity import (
+    retry,
+    stop_after_delay,
+    wait_exponential_jitter,
+    retry_if_not_exception_type,
+    before_sleep_log,
+)
+
 from cognee.infrastructure.llm.structured_output_framework.litellm_instructor.llm.llm_interface import (
     LLMInterface,
 )
@@ -14,19 +23,13 @@ from cognee.infrastructure.llm.exceptions import (
     ContentPolicyFilterError,
 )
 from cognee.infrastructure.files.utils.open_data_file import open_data_file
-from cognee.infrastructure.llm.structured_output_framework.litellm_instructor.llm.rate_limiter import (
-    rate_limit_async,
-    rate_limit_sync,
-    sleep_and_retry_async,
-    sleep_and_retry_sync,
-)
 from cognee.modules.observability.get_observe import get_observe
 from cognee.shared.logging_utils import get_logger
 
-observe = get_observe()
-
 logger = get_logger()
 
+observe = get_observe()
+
 
 class OpenAIAdapter(LLMInterface):
     """
@@ -97,8 +100,13 @@ class OpenAIAdapter(LLMInterface):
         self.fallback_endpoint = fallback_endpoint
 
     @observe(as_type="generation")
-    @sleep_and_retry_async()
+    @retry(
-    @rate_limit_async
+        stop=stop_after_delay(128),
+        wait=wait_exponential_jitter(2, 128),
+        retry=retry_if_not_exception_type(litellm.exceptions.NotFoundError),
+        before_sleep=before_sleep_log(logger, logging.DEBUG),
+        reraise=True,
+    )
     async def acreate_structured_output(
         self, text_input: str, system_prompt: str, response_model: Type[BaseModel]
     ) -> BaseModel:
@@ -148,10 +156,7 @@ class OpenAIAdapter(LLMInterface):
             InstructorRetryException,
         ) as e:
             if not (self.fallback_model and self.fallback_api_key):
-                raise ContentPolicyFilterError(
+                raise e
-                    f"The provided input contains content that is not aligned with our content policy: {text_input}"
-                ) from e
-
             try:
                 return await self.aclient.chat.completions.create(
                     model=self.fallback_model,
@@ -186,8 +191,13 @@ class OpenAIAdapter(LLMInterface):
                     ) from error
 
     @observe
-    @sleep_and_retry_sync()
+    @retry(
-    @rate_limit_sync
+        stop=stop_after_delay(128),
+        wait=wait_exponential_jitter(2, 128),
+        retry=retry_if_not_exception_type(litellm.exceptions.NotFoundError),
+        before_sleep=before_sleep_log(logger, logging.DEBUG),
+        reraise=True,
+    )
     def create_structured_output(
         self, text_input: str, system_prompt: str, response_model: Type[BaseModel]
     ) -> BaseModel:
@@ -231,7 +241,13 @@ class OpenAIAdapter(LLMInterface):
             max_retries=self.MAX_RETRIES,
         )
 
-    @rate_limit_async
+    @retry(
+        stop=stop_after_delay(128),
+        wait=wait_exponential_jitter(2, 128),
+        retry=retry_if_not_exception_type(litellm.exceptions.NotFoundError),
+        before_sleep=before_sleep_log(logger, logging.DEBUG),
+        reraise=True,
+    )
     async def create_transcript(self, input):
         """
         Generate an audio transcript from a user query.
@@ -263,7 +279,13 @@ class OpenAIAdapter(LLMInterface):
 
         return transcription
 
-    @rate_limit_async
+    @retry(
+        stop=stop_after_delay(128),
+        wait=wait_exponential_jitter(2, 128),
+        retry=retry_if_not_exception_type(litellm.exceptions.NotFoundError),
+        before_sleep=before_sleep_log(logger, logging.DEBUG),
+        reraise=True,
+    )
     async def transcribe_image(self, input) -> BaseModel:
         """
         Generate a transcription of an image from a user query.

cognee/infrastructure/loaders/external/advanced_pdf_loader.py

@@ -14,14 +14,6 @@ from cognee.infrastructure.loaders.external.pypdf_loader import PyPdfLoader
 
 logger = get_logger(__name__)
 
-try:
-    from unstructured.partition.pdf import partition_pdf
-except ImportError as e:
-    logger.info(
-        "unstructured[pdf] not installed, can't use AdvancedPdfLoader, will use PyPdfLoader instead."
-    )
-    raise ImportError from e
-
 
 @dataclass
 class _PageBuffer:
@@ -88,6 +80,8 @@ class AdvancedPdfLoader(LoaderInterface):
                 **kwargs,
             }
             # Use partition to extract elements
+            from unstructured.partition.pdf import partition_pdf
+
             elements = partition_pdf(**partition_kwargs)
 
             # Process elements into text content

cognee/tasks/storage/index_data_points.py

@@ -1,6 +1,6 @@
-from cognee.shared.logging_utils import get_logger
+import asyncio
 
-from cognee.infrastructure.databases.exceptions import EmbeddingException
+from cognee.shared.logging_utils import get_logger
 from cognee.infrastructure.databases.vector import get_vector_engine
 from cognee.infrastructure.engine import DataPoint
 
@@ -33,18 +33,23 @@ async def index_data_points(data_points: list[DataPoint]):
             indexed_data_point.metadata["index_fields"] = [field_name]
             index_points[index_name].append(indexed_data_point)
 
-    for index_name_and_field, indexable_points in index_points.items():
+    tasks: list[asyncio.Task] = []
-        first_occurence = index_name_and_field.index("_")
+    batch_size = vector_engine.embedding_engine.get_batch_size()
-        index_name = index_name_and_field[:first_occurence]
+
-        field_name = index_name_and_field[first_occurence + 1 :]
+    for index_name_and_field, points in index_points.items():
-        try:
+        first = index_name_and_field.index("_")
-            # In case the amount of indexable points is too large we need to send them in batches
+        index_name = index_name_and_field[:first]
-            batch_size = vector_engine.embedding_engine.get_batch_size()
+        field_name = index_name_and_field[first + 1 :]
-            for i in range(0, len(indexable_points), batch_size):
+
-                batch = indexable_points[i : i + batch_size]
+        # Create embedding requests per batch to run in parallel later
-                await vector_engine.index_data_points(index_name, field_name, batch)
+        for i in range(0, len(points), batch_size):
-        except EmbeddingException as e:
+            batch = points[i : i + batch_size]
-            logger.warning(f"Failed to index data points for {index_name}.{field_name}: {e}")
+            tasks.append(
+                asyncio.create_task(vector_engine.index_data_points(index_name, field_name, batch))
+            )
+
+    # Run all embedding requests in parallel
+    await asyncio.gather(*tasks)
 
     return data_points
 

cognee/tasks/storage/index_graph_edges.py

@@ -1,3 +1,5 @@
+import asyncio
+
 from cognee.modules.engine.utils.generate_edge_id import generate_edge_id
 from cognee.shared.logging_utils import get_logger
 from collections import Counter
@@ -76,15 +78,20 @@ async def index_graph_edges(
             indexed_data_point.metadata["index_fields"] = [field_name]
             index_points[index_name].append(indexed_data_point)
 
+    # Get maximum batch size for embedding model
+    batch_size = vector_engine.embedding_engine.get_batch_size()
+    tasks: list[asyncio.Task] = []
+
     for index_name, indexable_points in index_points.items():
         index_name, field_name = index_name.split(".")
 
-        # Get maximum batch size for embedding model
+        # Create embedding tasks to run in parallel later
-        batch_size = vector_engine.embedding_engine.get_batch_size()
-        # We save the data in batches of {batch_size} to not put a lot of pressure on the database
         for start in range(0, len(indexable_points), batch_size):
             batch = indexable_points[start : start + batch_size]
 
-            await vector_engine.index_data_points(index_name, field_name, batch)
+            tasks.append(vector_engine.index_data_points(index_name, field_name, batch))
+
+    # Start all embedding tasks and wait for completion
+    await asyncio.gather(*tasks)
 
     return None

cognee/tests/tasks/descriptive_metrics/metrics_test_utils.py

@@ -1,7 +1,6 @@
 from typing import List
 from cognee.infrastructure.engine import DataPoint
 from cognee.tasks.storage.add_data_points import add_data_points
-from cognee.infrastructure.databases.graph.get_graph_engine import create_graph_engine
 import cognee
 from cognee.infrastructure.databases.graph import get_graph_engine
 import json
@@ -64,7 +63,6 @@ async def create_connected_test_graph():
 
 
 async def get_metrics(provider: str, include_optional=True):
-    create_graph_engine.cache_clear()
     cognee.config.set_graph_database_provider(provider)
     graph_engine = await get_graph_engine()
     await graph_engine.delete_graph()

cognee/tests/tasks/descriptive_metrics/neo4j_metrics_test.py

@@ -1,7 +1,12 @@
-from cognee.tests.tasks.descriptive_metrics.metrics_test_utils import assert_metrics
 import asyncio
 
 
+async def main():
+    from cognee.tests.tasks.descriptive_metrics.metrics_test_utils import assert_metrics
+
+    await assert_metrics(provider="neo4j", include_optional=False)
+    await assert_metrics(provider="neo4j", include_optional=True)
+
+
 if __name__ == "__main__":
-    asyncio.run(assert_metrics(provider="neo4j", include_optional=False))
+    asyncio.run(main())
-    asyncio.run(assert_metrics(provider="neo4j", include_optional=True))

cognee/tests/tasks/entity_extraction/entity_extraction_test.py

@@ -0,0 +1,89 @@
+import os
+import pathlib
+import asyncio
+
+import cognee
+import cognee.modules.ingestion as ingestion
+from cognee.infrastructure.llm import get_max_chunk_tokens
+from cognee.infrastructure.llm.extraction import extract_content_graph
+from cognee.modules.chunking.TextChunker import TextChunker
+from cognee.modules.data.processing.document_types import TextDocument
+from cognee.modules.users.methods import get_default_user
+from cognee.shared.data_models import KnowledgeGraph
+from cognee.tasks.documents import extract_chunks_from_documents
+from cognee.tasks.ingestion import save_data_item_to_storage
+from cognee.infrastructure.files.utils.open_data_file import open_data_file
+
+
+async def extract_graphs(document_chunks):
+    """
+    Extract graph, and check if entities are present
+    """
+
+    extraction_results = await asyncio.gather(
+        *[extract_content_graph(chunk.text, KnowledgeGraph) for chunk in document_chunks]
+    )
+
+    return all(
+        any(
+            term in node.name.lower()
+            for extraction_result in extraction_results
+            for node in extraction_result.nodes
+        )
+        for term in ("qubit", "algorithm", "superposition")
+    )
+
+
+async def main():
+    """
+    Test how well the entity extraction works. Repeat graph generation a few times.
+    If 80% or more graphs are correctly generated, the test passes.
+    """
+
+    file_path = os.path.join(
+        pathlib.Path(__file__).parent.parent.parent, "test_data/Quantum_computers.txt"
+    )
+
+    await cognee.prune.prune_data()
+    await cognee.prune.prune_system(metadata=True)
+
+    await cognee.add("NLP is a subfield of computer science.")
+
+    original_file_path = await save_data_item_to_storage(file_path)
+
+    async with open_data_file(original_file_path) as file:
+        classified_data = ingestion.classify(file)
+
+        # data_id is the hash of original file contents + owner id to avoid duplicate data
+        data_id = ingestion.identify(classified_data, await get_default_user())
+
+    await cognee.add(file_path)
+
+    text_document = TextDocument(
+        id=data_id,
+        type="text",
+        mime_type="text/plain",
+        name="quantum_text",
+        raw_data_location=file_path,
+        external_metadata=None,
+    )
+
+    document_chunks = []
+    async for chunk in extract_chunks_from_documents(
+        [text_document], max_chunk_size=get_max_chunk_tokens(), chunker=TextChunker
+    ):
+        document_chunks.append(chunk)
+
+    number_of_reps = 5
+
+    graph_results = await asyncio.gather(
+        *[extract_graphs(document_chunks) for _ in range(number_of_reps)]
+    )
+
+    correct_graphs = [result for result in graph_results if result]
+
+    assert len(correct_graphs) >= 0.8 * number_of_reps
+
+
+if __name__ == "__main__":
+    asyncio.run(main())

pyproject.toml

@@ -56,6 +56,7 @@ dependencies = [
     "gunicorn>=20.1.0,<24",
     "websockets>=15.0.1,<16.0.0",
     "mistralai>=1.9.10",
+    "tenacity>=9.0.0",
 ]
 
 [project.optional-dependencies]
@@ -64,6 +65,7 @@ api=[]
 distributed = [
     "modal>=1.0.5,<2.0.0",
 ]
+
 scraping = [
     "tavily-python>=0.7.12",
     "beautifulsoup4>=4.13.1",
@@ -72,6 +74,7 @@ scraping = [
     "protego>=0.1",
     "APScheduler>=3.10.0,<=3.11.0"
 ]
+
 neo4j = ["neo4j>=5.28.0,<6"]
 neptune = ["langchain_aws>=0.2.22"]
 postgres = [