graphiti-core 0.4.3__py3-none-any.whl → 0.5.0__py3-none-any.whl

graphiti_core/llm_client/openai_generic_client.py

@@ -0,0 +1,163 @@
+"""
+Copyright 2024, Zep Software, Inc.
+
+Licensed under the Apache License, Version 2.0 (the "License");
+you may not use this file except in compliance with the License.
+You may obtain a copy of the License at
+
+    http://www.apache.org/licenses/LICENSE-2.0
+
+Unless required by applicable law or agreed to in writing, software
+distributed under the License is distributed on an "AS IS" BASIS,
+WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+See the License for the specific language governing permissions and
+limitations under the License.
+"""
+
+import json
+import logging
+import typing
+from typing import ClassVar
+
+import openai
+from openai import AsyncOpenAI
+from openai.types.chat import ChatCompletionMessageParam
+from pydantic import BaseModel
+
+from ..prompts.models import Message
+from .client import LLMClient
+from .config import LLMConfig
+from .errors import RateLimitError, RefusalError
+
+logger = logging.getLogger(__name__)
+
+DEFAULT_MODEL = 'gpt-4o-mini'
+
+
+class OpenAIGenericClient(LLMClient):
+    """
+    OpenAIClient is a client class for interacting with OpenAI's language models.
+
+    This class extends the LLMClient and provides methods to initialize the client,
+    get an embedder, and generate responses from the language model.
+
+    Attributes:
+        client (AsyncOpenAI): The OpenAI client used to interact with the API.
+        model (str): The model name to use for generating responses.
+        temperature (float): The temperature to use for generating responses.
+        max_tokens (int): The maximum number of tokens to generate in a response.
+
+    Methods:
+        __init__(config: LLMConfig | None = None, cache: bool = False, client: typing.Any = None):
+            Initializes the OpenAIClient with the provided configuration, cache setting, and client.
+
+        _generate_response(messages: list[Message]) -> dict[str, typing.Any]:
+            Generates a response from the language model based on the provided messages.
+    """
+
+    # Class-level constants
+    MAX_RETRIES: ClassVar[int] = 2
+
+    def __init__(
+        self, config: LLMConfig | None = None, cache: bool = False, client: typing.Any = None
+    ):
+        """
+        Initialize the OpenAIClient with the provided configuration, cache setting, and client.
+
+        Args:
+            config (LLMConfig | None): The configuration for the LLM client, including API key, model, base URL, temperature, and max tokens.
+            cache (bool): Whether to use caching for responses. Defaults to False.
+            client (Any | None): An optional async client instance to use. If not provided, a new AsyncOpenAI client is created.
+
+        """
+        # removed caching to simplify the `generate_response` override
+        if cache:
+            raise NotImplementedError('Caching is not implemented for OpenAI')
+
+        if config is None:
+            config = LLMConfig()
+
+        super().__init__(config, cache)
+
+        if client is None:
+            self.client = AsyncOpenAI(api_key=config.api_key, base_url=config.base_url)
+        else:
+            self.client = client
+
+    async def _generate_response(
+        self, messages: list[Message], response_model: type[BaseModel] | None = None
+    ) -> dict[str, typing.Any]:
+        openai_messages: list[ChatCompletionMessageParam] = []
+        for m in messages:
+            m.content = self._clean_input(m.content)
+            if m.role == 'user':
+                openai_messages.append({'role': 'user', 'content': m.content})
+            elif m.role == 'system':
+                openai_messages.append({'role': 'system', 'content': m.content})
+        try:
+            response = await self.client.chat.completions.create(
+                model=self.model or DEFAULT_MODEL,
+                messages=openai_messages,
+                temperature=self.temperature,
+                max_tokens=self.max_tokens,
+                response_format={'type': 'json_object'},
+            )
+            result = response.choices[0].message.content or ''
+            return json.loads(result)
+        except openai.RateLimitError as e:
+            raise RateLimitError from e
+        except Exception as e:
+            logger.error(f'Error in generating LLM response: {e}')
+            raise
+
+    async def generate_response(
+        self, messages: list[Message], response_model: type[BaseModel] | None = None
+    ) -> dict[str, typing.Any]:
+        retry_count = 0
+        last_error = None
+
+        if response_model is not None:
+            serialized_model = json.dumps(response_model.model_json_schema())
+            messages[
+                -1
+            ].content += (
+                f'\n\nRespond with a JSON object in the following format:\n\n{serialized_model}'
+            )
+
+        while retry_count <= self.MAX_RETRIES:
+            try:
+                response = await self._generate_response(messages, response_model)
+                return response
+            except (RateLimitError, RefusalError):
+                # These errors should not trigger retries
+                raise
+            except (openai.APITimeoutError, openai.APIConnectionError, openai.InternalServerError):
+                # Let OpenAI's client handle these retries
+                raise
+            except Exception as e:
+                last_error = e
+
+                # Don't retry if we've hit the max retries
+                if retry_count >= self.MAX_RETRIES:
+                    logger.error(f'Max retries ({self.MAX_RETRIES}) exceeded. Last error: {e}')
+                    raise
+
+                retry_count += 1
+
+                # Construct a detailed error message for the LLM
+                error_context = (
+                    f'The previous response attempt was invalid. '
+                    f'Error type: {e.__class__.__name__}. '
+                    f'Error details: {str(e)}. '
+                    f'Please try again with a valid response, ensuring the output matches '
+                    f'the expected format and constraints.'
+                )
+
+                error_message = Message(role='user', content=error_context)
+                messages.append(error_message)
+                logger.warning(
+                    f'Retrying after application error (attempt {retry_count}/{self.MAX_RETRIES}): {e}'
+                )
+
+        # If we somehow get here, raise the last error
+        raise last_error or Exception('Max retries exceeded with no specific error')

graphiti_core/nodes.py

@@ -16,7 +16,7 @@ limitations under the License.
 
 import logging
 from abc import ABC, abstractmethod
-from datetime import datetime, timezone
+from datetime import datetime
 from enum import Enum
 from time import time
 from typing import Any
@@ -28,12 +28,13 @@ from typing_extensions import LiteralString
 
 from graphiti_core.embedder import EmbedderClient
 from graphiti_core.errors import NodeNotFoundError
-from graphiti_core.helpers import DEFAULT_DATABASE, DEFAULT_PAGE_LIMIT
+from graphiti_core.helpers import DEFAULT_DATABASE
 from graphiti_core.models.nodes.node_db_queries import (
     COMMUNITY_NODE_SAVE,
     ENTITY_NODE_SAVE,
     EPISODIC_NODE_SAVE,
 )
+from graphiti_core.utils.datetime_utils import utc_now
 
 logger = logging.getLogger(__name__)
 
@@ -79,7 +80,7 @@ class Node(BaseModel, ABC):
     name: str = Field(description='name of the node')
     group_id: str = Field(description='partition of the graph')
     labels: list[str] = Field(default_factory=list)
-    created_at: datetime = Field(default_factory=lambda: datetime.now(timezone.utc))
+    created_at: datetime = Field(default_factory=lambda: utc_now())
 
     @abstractmethod
     async def save(self, driver: AsyncDriver): ...
@@ -212,10 +213,11 @@ class EpisodicNode(Node):
         cls,
         driver: AsyncDriver,
         group_ids: list[str],
-        limit: int = DEFAULT_PAGE_LIMIT,
+        limit: int | None = None,
         created_at: datetime | None = None,
     ):
         cursor_query: LiteralString = 'AND e.created_at < $created_at' if created_at else ''
+        limit_query: LiteralString = 'LIMIT $limit' if limit is not None else ''
 
         records, _, _ = await driver.execute_query(
             """
@@ -233,8 +235,8 @@ class EpisodicNode(Node):
             e.source_description AS source_description,
             e.source AS source
         ORDER BY e.uuid DESC
-        LIMIT $limit
-        """,
+        """
+            + limit_query,
             group_ids=group_ids,
             created_at=created_at,
             limit=limit,
@@ -328,10 +330,11 @@ class EntityNode(Node):
         cls,
         driver: AsyncDriver,
         group_ids: list[str],
-        limit: int = DEFAULT_PAGE_LIMIT,
+        limit: int | None = None,
         created_at: datetime | None = None,
     ):
         cursor_query: LiteralString = 'AND n.created_at < $created_at' if created_at else ''
+        limit_query: LiteralString = 'LIMIT $limit' if limit is not None else ''
 
         records, _, _ = await driver.execute_query(
             """
@@ -347,8 +350,8 @@ class EntityNode(Node):
             n.created_at AS created_at, 
             n.summary AS summary
         ORDER BY n.uuid DESC
-        LIMIT $limit
-        """,
+        """
+            + limit_query,
             group_ids=group_ids,
             created_at=created_at,
             limit=limit,
@@ -442,10 +445,11 @@ class CommunityNode(Node):
         cls,
         driver: AsyncDriver,
         group_ids: list[str],
-        limit: int = DEFAULT_PAGE_LIMIT,
+        limit: int | None = None,
         created_at: datetime | None = None,
     ):
         cursor_query: LiteralString = 'AND n.created_at < $created_at' if created_at else ''
+        limit_query: LiteralString = 'LIMIT $limit' if limit is not None else ''
 
         records, _, _ = await driver.execute_query(
             """
@@ -461,8 +465,8 @@ class CommunityNode(Node):
             n.created_at AS created_at, 
             n.summary AS summary
         ORDER BY n.uuid DESC
-        LIMIT $limit
-        """,
+        """
+            + limit_query,
             group_ids=group_ids,
             created_at=created_at,
             limit=limit,

graphiti_core/prompts/dedupe_edges.py

@@ -15,11 +15,30 @@ limitations under the License.
 """
 
 import json
-from typing import Any, Protocol, TypedDict
+from typing import Any, Optional, Protocol, TypedDict
+
+from pydantic import BaseModel, Field
 
 from .models import Message, PromptFunction, PromptVersion
 
 
+class EdgeDuplicate(BaseModel):
+    is_duplicate: bool = Field(..., description='true or false')
+    uuid: Optional[str] = Field(
+        None,
+        description="uuid of the existing edge like '5d643020624c42fa9de13f97b1b3fa39' or null",
+    )
+
+
+class UniqueFact(BaseModel):
+    uuid: str = Field(..., description='unique identifier of the fact')
+    fact: str = Field(..., description='fact of a unique edge')
+
+
+class UniqueFacts(BaseModel):
+    unique_facts: list[UniqueFact]
+
+
 class Prompt(Protocol):
     edge: PromptVersion
     edge_list: PromptVersion
@@ -56,12 +75,6 @@ def edge(context: dict[str, Any]) -> list[Message]:
 
         Guidelines:
         1. The facts do not need to be completely identical to be duplicates, they just need to express the same information.
-
-        Respond with a JSON object in the following format:
-            {{
-                "is_duplicate": true or false,
-                "uuid": uuid of the existing edge like "5d643020624c42fa9de13f97b1b3fa39" or null,
-            }}
         """,
         ),
     ]
@@ -90,16 +103,6 @@ def edge_list(context: dict[str, Any]) -> list[Message]:
         3. Facts will often discuss the same or similar relation between identical entities
         4. The final list should have only unique facts. If 3 facts are all duplicates of each other, only one of their
             facts should be in the response
-
-        Respond with a JSON object in the following format:
-        {{
-            "unique_facts": [
-                {{
-                    "uuid": "unique identifier of the fact",
-                    "fact": "fact of a unique edge"
-                }}
-            ]
-        }}
         """,
         ),
     ]

graphiti_core/prompts/dedupe_nodes.py

@@ -15,11 +15,25 @@ limitations under the License.
 """
 
 import json
-from typing import Any, Protocol, TypedDict
+from typing import Any, Optional, Protocol, TypedDict
+
+from pydantic import BaseModel, Field
 
 from .models import Message, PromptFunction, PromptVersion
 
 
+class NodeDuplicate(BaseModel):
+    is_duplicate: bool = Field(..., description='true or false')
+    uuid: Optional[str] = Field(
+        None,
+        description="uuid of the existing node like '5d643020624c42fa9de13f97b1b3fa39' or null",
+    )
+    name: str = Field(
+        ...,
+        description="Updated name of the new node (use the best name between the new node's name, an existing duplicate name, or a combination of both)",
+    )
+
+
 class Prompt(Protocol):
     node: PromptVersion
     node_list: PromptVersion

graphiti_core/prompts/eval.py

@@ -17,9 +17,26 @@ limitations under the License.
 import json
 from typing import Any, Protocol, TypedDict
 
+from pydantic import BaseModel, Field
+
 from .models import Message, PromptFunction, PromptVersion
 
 
+class QueryExpansion(BaseModel):
+    query: str = Field(..., description='query optimized for database search')
+
+
+class QAResponse(BaseModel):
+    ANSWER: str = Field(..., description='how Alice would answer the question')
+
+
+class EvalResponse(BaseModel):
+    is_correct: bool = Field(..., description='boolean if the answer is correct or incorrect')
+    reasoning: str = Field(
+        ..., description='why you determined the response was correct or incorrect'
+    )
+
+
 class Prompt(Protocol):
     qa_prompt: PromptVersion
     eval_prompt: PromptVersion
@@ -41,10 +58,6 @@ def query_expansion(context: dict[str, Any]) -> list[Message]:
     <QUESTION>
     {json.dumps(context['query'])}
     </QUESTION>
-    respond with a JSON object in the following format:
-    {{
-        "query": "query optimized for database search"
-    }}
     """
     return [
         Message(role='system', content=sys_prompt),
@@ -67,10 +80,6 @@ def qa_prompt(context: dict[str, Any]) -> list[Message]:
     <QUESTION>
     {context['query']}
     </QUESTION>
-    respond with a JSON object in the following format:
-    {{
-        "ANSWER": "how Alice would answer the question"
-    }}
     """
     return [
         Message(role='system', content=sys_prompt),
@@ -96,12 +105,6 @@ def eval_prompt(context: dict[str, Any]) -> list[Message]:
     <RESPONSE>
     {context['response']}
     </RESPONSE>
-    
-    respond with a JSON object in the following format:
-    {{
-        "is_correct": "boolean if the answer is correct or incorrect"
-        "reasoning": "why you determined the response was correct or incorrect"
-    }}
     """
     return [
         Message(role='system', content=sys_prompt),

graphiti_core/prompts/extract_edge_dates.py

@@ -14,11 +14,24 @@ See the License for the specific language governing permissions and
 limitations under the License.
 """
 
-from typing import Any, Protocol, TypedDict
+from typing import Any, Optional, Protocol, TypedDict
+
+from pydantic import BaseModel, Field
 
 from .models import Message, PromptFunction, PromptVersion
 
 
+class EdgeDates(BaseModel):
+    valid_at: Optional[str] = Field(
+        None,
+        description='The date and time when the relationship described by the edge fact became true or was established. YYYY-MM-DDTHH:MM:SS.SSSSSSZ or null.',
+    )
+    invalid_at: Optional[str] = Field(
+        None,
+        description='The date and time when the relationship described by the edge fact stopped being true or ended. YYYY-MM-DDTHH:MM:SS.SSSSSSZ or null.',
+    )
+
+
 class Prompt(Protocol):
     v1: PromptVersion
 
@@ -60,7 +73,7 @@ def v1(context: dict[str, Any]) -> list[Message]:
             Analyze the conversation and determine if there are dates that are part of the edge fact. Only set dates if they explicitly relate to the formation or alteration of the relationship itself.
 
             Guidelines:
-            1. Use ISO 8601 format (YYYY-MM-DDTHH:MM:SSZ) for datetimes.
+            1. Use ISO 8601 format (YYYY-MM-DDTHH:MM:SS.SSSSSSZ) for datetimes.
             2. Use the reference timestamp as the current time when determining the valid_at and invalid_at dates.
             3. If the fact is written in the present tense, use the Reference Timestamp for the valid_at date
             4. If no temporal information is found that establishes or changes the relationship, leave the fields as null.
@@ -69,11 +82,6 @@ def v1(context: dict[str, Any]) -> list[Message]:
             7. If only a date is mentioned without a specific time, use 00:00:00 (midnight) for that date.
             8. If only year is mentioned, use January 1st of that year at 00:00:00.
             9. Always include the time zone offset (use Z for UTC if no specific time zone is mentioned).
-            Respond with a JSON object:
-            {{
-                "valid_at": "YYYY-MM-DDTHH:MM:SS.SSSSSSZ or null",
-                "invalid_at": "YYYY-MM-DDTHH:MM:SS.SSSSSSZ or null",
-            }}
             """,
         ),
     ]

graphiti_core/prompts/extract_edges.py

@@ -17,9 +17,26 @@ limitations under the License.
 import json
 from typing import Any, Protocol, TypedDict
 
+from pydantic import BaseModel, Field
+
 from .models import Message, PromptFunction, PromptVersion
 
 
+class Edge(BaseModel):
+    relation_type: str = Field(..., description='RELATION_TYPE_IN_CAPS')
+    source_entity_name: str = Field(..., description='name of the source entity')
+    target_entity_name: str = Field(..., description='name of the target entity')
+    fact: str = Field(..., description='extracted factual information')
+
+
+class ExtractedEdges(BaseModel):
+    edges: list[Edge]
+
+
+class MissingFacts(BaseModel):
+    missing_facts: list[str] = Field(..., description="facts that weren't extracted")
+
+
 class Prompt(Protocol):
     edge: PromptVersion
     reflexion: PromptVersion
@@ -54,25 +71,12 @@ def edge(context: dict[str, Any]) -> list[Message]:
 
         Given the above MESSAGES and ENTITIES, extract all facts pertaining to the listed ENTITIES from the CURRENT MESSAGE. 
         
-
         Guidelines:
         1. Extract facts only between the provided entities.
         2. Each fact should represent a clear relationship between two DISTINCT nodes.
         3. The relation_type should be a concise, all-caps description of the fact (e.g., LOVES, IS_FRIENDS_WITH, WORKS_FOR).
         4. Provide a more detailed fact containing all relevant information.
         5. Consider temporal aspects of relationships when relevant.
-
-        Respond with a JSON object in the following format:
-        {{
-            "edges": [
-                {{
-                    "relation_type": "RELATION_TYPE_IN_CAPS",
-                    "source_entity_name": "name of the source entity",
-                    "target_entity_name": "name of the target entity",
-                    "fact": "extracted factual information",
-                }}
-            ]
-        }}
         """,
         ),
     ]
@@ -98,12 +102,7 @@ def reflexion(context: dict[str, Any]) -> list[Message]:
 </EXTRACTED FACTS>
 
 Given the above MESSAGES, list of EXTRACTED ENTITIES entities, and list of EXTRACTED FACTS; 
-determine if any facts haven't been extracted:
-
-Respond with a JSON object in the following format:
-{{
-    "missing_facts": [ "facts that weren't extracted", ...]
-}}
+determine if any facts haven't been extracted.
 """
     return [
         Message(role='system', content=sys_prompt),

graphiti_core/prompts/extract_nodes.py

@@ -17,9 +17,19 @@ limitations under the License.
 import json
 from typing import Any, Protocol, TypedDict
 
+from pydantic import BaseModel, Field
+
 from .models import Message, PromptFunction, PromptVersion
 
 
+class ExtractedNodes(BaseModel):
+    extracted_node_names: list[str] = Field(..., description='Name of the extracted entity')
+
+
+class MissedEntities(BaseModel):
+    missed_entities: list[str] = Field(..., description="Names of entities that weren't extracted")
+
+
 class Prompt(Protocol):
     extract_message: PromptVersion
     extract_json: PromptVersion
@@ -56,11 +66,6 @@ Guidelines:
 4. DO NOT create nodes for temporal information like dates, times or years (these will be added to edges later).
 5. Be as explicit as possible in your node names, using full names.
 6. DO NOT extract entities mentioned only in PREVIOUS MESSAGES, those messages are only to provide context.
-
-Respond with a JSON object in the following format:
-{{
-    "extracted_node_names": ["Name of the extracted entity", ...],
-}}
 """
     return [
         Message(role='system', content=sys_prompt),
@@ -87,11 +92,6 @@ Given the above source description and JSON, extract relevant entity nodes from
 Guidelines:
 1. Always try to extract an entities that the JSON represents. This will often be something like a "name" or "user field
 2. Do NOT extract any properties that contain dates
-
-Respond with a JSON object in the following format:
-{{
-    "extracted_node_names": ["Name of the extracted entity", ...],
-}}
 """
     return [
         Message(role='system', content=sys_prompt),
@@ -116,11 +116,6 @@ Guidelines:
 2. Avoid creating nodes for relationships or actions.
 3. Avoid creating nodes for temporal information like dates, times or years (these will be added to edges later).
 4. Be as explicit as possible in your node names, using full names and avoiding abbreviations.
-
-Respond with a JSON object in the following format:
-{{
-    "extracted_node_names": ["Name of the extracted entity", ...],
-}}
 """
     return [
         Message(role='system', content=sys_prompt),
@@ -144,12 +139,7 @@ def reflexion(context: dict[str, Any]) -> list[Message]:
 </EXTRACTED ENTITIES>
 
 Given the above previous messages, current message, and list of extracted entities; determine if any entities haven't been
-extracted:
-
-Respond with a JSON object in the following format:
-{{
-    "missed_entities": [ "name of entity that wasn't extracted", ...]
-}}
+extracted.
 """
     return [
         Message(role='system', content=sys_prompt),

graphiti_core/prompts/invalidate_edges.py

@@ -16,9 +16,22 @@ limitations under the License.
 
 from typing import Any, Protocol, TypedDict
 
+from pydantic import BaseModel, Field
+
 from .models import Message, PromptFunction, PromptVersion
 
 
+class InvalidatedEdge(BaseModel):
+    uuid: str = Field(..., description='The UUID of the edge to be invalidated')
+    fact: str = Field(..., description='Updated fact of the edge')
+
+
+class InvalidatedEdges(BaseModel):
+    invalidated_edges: list[InvalidatedEdge] = Field(
+        ..., description='List of edges that should be invalidated'
+    )
+
+
 class Prompt(Protocol):
     v1: PromptVersion
     v2: PromptVersion
@@ -56,18 +69,6 @@ def v1(context: dict[str, Any]) -> list[Message]:
                 {context['new_edges']}
 
                 Each edge is formatted as: "UUID | SOURCE_NODE - EDGE_NAME - TARGET_NODE (fact: EDGE_FACT), START_DATE (END_DATE, optional))"
-
-                For each existing edge that should be invalidated, respond with a JSON object in the following format:
-                {{
-                    "invalidated_edges": [
-                        {{
-                            "edge_uuid": "The UUID of the edge to be invalidated (the part before the | character)",
-                            "fact": "Updated fact of the edge"
-                        }}
-                    ]
-                }}
-
-                If no relationships need to be invalidated based on these strict criteria, return an empty list for "invalidated_edges".
             """,
         ),
     ]
@@ -89,19 +90,6 @@ def v2(context: dict[str, Any]) -> list[Message]:
 
                 New Edge:
                 {context['new_edge']}
-
-
-                For each existing edge that should be invalidated, respond with a JSON object in the following format:
-                {{
-                    "invalidated_edges": [
-                        {{
-                            "uuid": "The UUID of the edge to be invalidated",
-                            "fact": "Updated fact of the edge"
-                        }}
-                    ]
-                }}
-
-                If no relationships need to be invalidated based on these strict criteria, return an empty list for "invalidated_edges".
             """,
         ),
     ]