graphiti-core 0.11.5__tar.gz → 0.11.6rc1__tar.gz

{graphiti_core-0.11.5 → graphiti_core-0.11.6rc1}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.3
 Name: graphiti-core
-Version: 0.11.5
+Version: 0.11.6rc1
 Summary: A temporal graph building library
 License: Apache-2.0
 Author: Paul Paliychuk
@@ -18,7 +18,6 @@ Provides-Extra: groq
 Requires-Dist: anthropic (>=0.49.0) ; extra == "anthropic"
 Requires-Dist: diskcache (>=5.6.3)
 Requires-Dist: google-genai (>=1.8.0) ; extra == "google-genai"
-Requires-Dist: graph-service (>=1.0.0.7,<2.0.0.0)
 Requires-Dist: groq (>=0.2.0) ; extra == "groq"
 Requires-Dist: neo4j (>=5.23.0)
 Requires-Dist: numpy (>=1.0.0)

{graphiti_core-0.11.5 → graphiti_core-0.11.6rc1}/graphiti_core/edges.py

@@ -46,7 +46,6 @@ ENTITY_EDGE_RETURN: LiteralString = """
             e.name AS name,
             e.group_id AS group_id,
             e.fact AS fact,
-            e.fact_embedding AS fact_embedding,
             e.episodes AS episodes,
             e.expired_at AS expired_at,
             e.valid_at AS valid_at,
@@ -222,6 +221,20 @@ class EntityEdge(Edge):
 
         return self.fact_embedding
 
+    async def load_fact_embedding(self, driver: AsyncDriver):
+        query: LiteralString = """
+            MATCH (n:Entity)-[e:RELATES_TO {uuid: $uuid}]->(m:Entity)
+            RETURN e.fact_embedding AS fact_embedding
+        """
+        records, _, _ = await driver.execute_query(
+            query, uuid=self.uuid, database_=DEFAULT_DATABASE, routing_='r'
+        )
+
+        if len(records) == 0:
+            raise EdgeNotFoundError(self.uuid)
+
+        self.fact_embedding = records[0]['fact_embedding']
+
     async def save(self, driver: AsyncDriver):
         result = await driver.execute_query(
             ENTITY_EDGE_SAVE,
@@ -321,8 +334,8 @@ class EntityEdge(Edge):
     async def get_by_node_uuid(cls, driver: AsyncDriver, node_uuid: str):
         query: LiteralString = (
             """
-                        MATCH (n:Entity {uuid: $node_uuid})-[e:RELATES_TO]-(m:Entity)
-                        """
+                                        MATCH (n:Entity {uuid: $node_uuid})-[e:RELATES_TO]-(m:Entity)
+                                        """
             + ENTITY_EDGE_RETURN
         )
         records, _, _ = await driver.execute_query(
@@ -452,7 +465,6 @@ def get_entity_edge_from_record(record: Any) -> EntityEdge:
         name=record['name'],
         group_id=record['group_id'],
         episodes=record['episodes'],
-        fact_embedding=record['fact_embedding'],
         created_at=record['created_at'].to_native(),
         expired_at=parse_db_date(record['expired_at']),
         valid_at=parse_db_date(record['valid_at']),
@@ -471,6 +483,8 @@ def get_community_edge_from_record(record: Any):
 
 
 async def create_entity_edge_embeddings(embedder: EmbedderClient, edges: list[EntityEdge]):
+    if len(edges) == 0:
+        return
     fact_embeddings = await embedder.create_batch([edge.fact for edge in edges])
     for edge, fact_embedding in zip(edges, fact_embeddings, strict=True):
         edge.fact_embedding = fact_embedding

{graphiti_core-0.11.5 → graphiti_core-0.11.6rc1}/graphiti_core/graphiti.py

@@ -380,6 +380,7 @@ class Graphiti:
                 resolve_extracted_edges(
                     self.clients,
                     edges,
+                    episode,
                 ),
                 extract_attributes_from_nodes(
                     self.clients, nodes, episode, previous_episodes, entity_types
@@ -682,7 +683,11 @@ class Graphiti:
 
         related_edges = await get_relevant_edges(self.driver, [updated_edge], SearchFilters(), 0.8)
 
-        resolved_edge = await dedupe_extracted_edge(self.llm_client, updated_edge, related_edges[0])
+        resolved_edge = await dedupe_extracted_edge(
+            self.llm_client,
+            updated_edge,
+            related_edges[0],
+        )
 
         contradicting_edges = await get_edge_contradictions(self.llm_client, edge, related_edges[0])
         invalidated_edges = resolve_edge_contradictions(resolved_edge, contradicting_edges)

{graphiti_core-0.11.5 → graphiti_core-0.11.6rc1}/graphiti_core/llm_client/anthropic_client.py

@@ -139,15 +139,11 @@ class AnthropicClient(LLMClient):
             A list containing a single tool definition for use with the Anthropic API.
         """
         if response_model is not None:
-            # temporary debug log
-            logger.info(f'Creating tool for response_model: {response_model}')
             # Use the response_model to define the tool
             model_schema = response_model.model_json_schema()
             tool_name = response_model.__name__
             description = model_schema.get('description', f'Extract {tool_name} information')
         else:
-            # temporary debug log
-            logger.info('Creating generic JSON output tool')
             # Create a generic JSON output tool
             tool_name = 'generic_json_output'
             description = 'Output data in JSON format'
@@ -205,8 +201,6 @@ class AnthropicClient(LLMClient):
         try:
             # Create the appropriate tool based on whether response_model is provided
             tools, tool_choice = self._create_tool(response_model)
-            # temporary debug log
-            logger.info(f'using model: {self.model} with max_tokens: {self.max_tokens}')
             result = await self.client.messages.create(
                 system=system_message.content,
                 max_tokens=max_creation_tokens,
@@ -227,13 +221,6 @@ class AnthropicClient(LLMClient):
                     return tool_args
 
             # If we didn't get a proper tool_use response, try to extract from text
-            # logger.debug(
-            #     f'Did not get a tool_use response, trying to extract json from text. Result: {result.content}'
-            # )
-            # temporary debug log
-            logger.info(
-                f'Did not get a tool_use response, trying to extract json from text. Result: {result.content}'
-            )
             for content_item in result.content:
                 if content_item.type == 'text':
                     return self._extract_json_from_text(content_item.text)

{graphiti_core-0.11.5 → graphiti_core-0.11.6rc1}/graphiti_core/nodes.py

@@ -42,7 +42,6 @@ ENTITY_NODE_RETURN: LiteralString = """
         RETURN
             n.uuid As uuid, 
             n.name AS name,
-            n.name_embedding AS name_embedding,
             n.group_id AS group_id,
             n.created_at AS created_at, 
             n.summary AS summary,
@@ -305,6 +304,20 @@ class EntityNode(Node):
 
         return self.name_embedding
 
+    async def load_name_embedding(self, driver: AsyncDriver):
+        query: LiteralString = """
+            MATCH (n:Entity {uuid: $uuid})
+            RETURN n.name_embedding AS name_embedding
+        """
+        records, _, _ = await driver.execute_query(
+            query, uuid=self.uuid, database_=DEFAULT_DATABASE, routing_='r'
+        )
+
+        if len(records) == 0:
+            raise NodeNotFoundError(self.uuid)
+
+        self.name_embedding = records[0]['name_embedding']
+
     async def save(self, driver: AsyncDriver):
         entity_data: dict[str, Any] = {
             'uuid': self.uuid,
@@ -332,8 +345,8 @@ class EntityNode(Node):
     async def get_by_uuid(cls, driver: AsyncDriver, uuid: str):
         query = (
             """
-                                    MATCH (n:Entity {uuid: $uuid})
-                                    """
+                                                        MATCH (n:Entity {uuid: $uuid})
+                                                        """
             + ENTITY_NODE_RETURN
         )
         records, _, _ = await driver.execute_query(
@@ -428,6 +441,20 @@ class CommunityNode(Node):
 
         return self.name_embedding
 
+    async def load_name_embedding(self, driver: AsyncDriver):
+        query: LiteralString = """
+            MATCH (c:Community {uuid: $uuid})
+            RETURN c.name_embedding AS name_embedding
+        """
+        records, _, _ = await driver.execute_query(
+            query, uuid=self.uuid, database_=DEFAULT_DATABASE, routing_='r'
+        )
+
+        if len(records) == 0:
+            raise NodeNotFoundError(self.uuid)
+
+        self.name_embedding = records[0]['name_embedding']
+
     @classmethod
     async def get_by_uuid(cls, driver: AsyncDriver, uuid: str):
         records, _, _ = await driver.execute_query(
@@ -436,7 +463,6 @@ class CommunityNode(Node):
         RETURN
             n.uuid As uuid, 
             n.name AS name,
-            n.name_embedding AS name_embedding,
             n.group_id AS group_id,
             n.created_at AS created_at, 
             n.summary AS summary
@@ -461,7 +487,6 @@ class CommunityNode(Node):
         RETURN
             n.uuid As uuid, 
             n.name AS name,
-            n.name_embedding AS name_embedding,
             n.group_id AS group_id,
             n.created_at AS created_at, 
             n.summary AS summary
@@ -495,7 +520,6 @@ class CommunityNode(Node):
         RETURN
             n.uuid As uuid, 
             n.name AS name,
-            n.name_embedding AS name_embedding,
             n.group_id AS group_id,
             n.created_at AS created_at, 
             n.summary AS summary
@@ -534,7 +558,6 @@ def get_entity_node_from_record(record: Any) -> EntityNode:
         uuid=record['uuid'],
         name=record['name'],
         group_id=record['group_id'],
-        name_embedding=record['name_embedding'],
         labels=record['labels'],
         created_at=record['created_at'].to_native(),
         summary=record['summary'],

{graphiti_core-0.11.5 → graphiti_core-0.11.6rc1}/graphiti_core/prompts/dedupe_nodes.py

@@ -44,7 +44,7 @@ def node(context: dict[str, Any]) -> list[Message]:
     return [
         Message(
             role='system',
-            content='You are a helpful assistant that de-duplicates entities from entity lists.',
+            content='You are a helpful assistant that determines whether or not a NEW ENTITY is a duplicate of any EXISTING ENTITIES.',
         ),
         Message(
             role='user',
@@ -69,14 +69,21 @@ def node(context: dict[str, Any]) -> list[Message]:
         Given the above EXISTING ENTITIES and their attributes, MESSAGE, and PREVIOUS MESSAGES; Determine if the NEW ENTITY extracted from the conversation
         is a duplicate entity of one of the EXISTING ENTITIES.
         
-        The ENTITY TYPE DESCRIPTION gives more insight into what the entity type means for the NEW ENTITY.
+        Entities should only be considered duplicates if they refer to the *same real-world object or concept*.
+
+        Do NOT mark entities as duplicates if:
+        - They are related but distinct.
+        - They have similar names or purposes but refer to separate instances or concepts.
 
         Task:
         If the NEW ENTITY represents a duplicate entity of any entity in EXISTING ENTITIES, set duplicate_entity_id to the
-        id of the EXISTING ENTITY that is the duplicate. If the NEW ENTITY is not a duplicate of any of the EXISTING ENTITIES,
+        id of the EXISTING ENTITY that is the duplicate. 
+        
+        If the NEW ENTITY is not a duplicate of any of the EXISTING ENTITIES,
         duplicate_entity_id should be set to -1.
         
-        Also return the most complete name for the entity.
+        Also return the name that best describes the NEW ENTITY (whether it is the name of the NEW ENTITY, a node it
+        is a duplicate of, or a combination of the two).
         """,
         ),
     ]

{graphiti_core-0.11.5 → graphiti_core-0.11.6rc1}/graphiti_core/prompts/extract_nodes.py

@@ -256,7 +256,7 @@ def extract_attributes(context: dict[str, Any]) -> list[Message]:
         1. Do not hallucinate entity property values if they cannot be found in the current context.
         2. Only use the provided MESSAGES and ENTITY to set attribute values.
         3. The summary attribute represents a summary of the ENTITY, and should be updated with new information about the Entity from the MESSAGES. 
-            Summaries must be no longer than 500 words.
+            Summaries must be no longer than 250 words.
         
         <ENTITY>
         {context['node']}

{graphiti_core-0.11.5 → graphiti_core-0.11.6rc1}/graphiti_core/prompts/summarize_nodes.py

@@ -25,7 +25,7 @@ from .models import Message, PromptFunction, PromptVersion
 class Summary(BaseModel):
     summary: str = Field(
         ...,
-        description='Summary containing the important information about the entity. Under 500 words',
+        description='Summary containing the important information about the entity. Under 250 words',
     )
 
 
@@ -56,7 +56,7 @@ def summarize_pair(context: dict[str, Any]) -> list[Message]:
             content=f"""
         Synthesize the information from the following two summaries into a single succinct summary.
         
-        Summaries must be under 500 words.
+        Summaries must be under 250 words.
 
         Summaries:
         {json.dumps(context['node_summaries'], indent=2)}
@@ -82,7 +82,7 @@ def summarize_context(context: dict[str, Any]) -> list[Message]:
         
         Given the above MESSAGES and the following ENTITY name, create a summary for the ENTITY. Your summary must only use
         information from the provided MESSAGES. Your summary should also only contain information relevant to the
-        provided ENTITY. Summaries must be under 500 words.
+        provided ENTITY. Summaries must be under 250 words.
         
         In addition, extract any values for the provided entity properties based on their descriptions.
         If the value of the entity property cannot be found in the current context, set the value of the property to the Python value None.
@@ -117,7 +117,7 @@ def summary_description(context: dict[str, Any]) -> list[Message]:
             role='user',
             content=f"""
         Create a short one sentence description of the summary that explains what kind of information is summarized.
-        Summaries must be under 500 words.
+        Summaries must be under 250 words.
 
         Summary:
         {json.dumps(context['summary'], indent=2)}

{graphiti_core-0.11.5 → graphiti_core-0.11.6rc1}/graphiti_core/search/search.py

@@ -209,6 +209,9 @@ async def edge_search(
 
         reranked_uuids = rrf(search_result_uuids, min_score=reranker_min_score)
     elif config.reranker == EdgeReranker.mmr:
+        await semaphore_gather(
+            *[edge.load_fact_embedding(driver) for result in search_results for edge in result]
+        )
         search_result_uuids_and_vectors = [
             (edge.uuid, edge.fact_embedding if edge.fact_embedding is not None else [0.0] * 1024)
             for result in search_results
@@ -308,6 +311,9 @@ async def node_search(
     if config.reranker == NodeReranker.rrf:
         reranked_uuids = rrf(search_result_uuids, min_score=reranker_min_score)
     elif config.reranker == NodeReranker.mmr:
+        await semaphore_gather(
+            *[node.load_name_embedding(driver) for result in search_results for node in result]
+        )
         search_result_uuids_and_vectors = [
             (node.uuid, node.name_embedding if node.name_embedding is not None else [0.0] * 1024)
             for result in search_results
@@ -431,6 +437,13 @@ async def community_search(
     if config.reranker == CommunityReranker.rrf:
         reranked_uuids = rrf(search_result_uuids, min_score=reranker_min_score)
     elif config.reranker == CommunityReranker.mmr:
+        await semaphore_gather(
+            *[
+                community.load_name_embedding(driver)
+                for result in search_results
+                for community in result
+            ]
+        )
         search_result_uuids_and_vectors = [
             (
                 community.uuid,

{graphiti_core-0.11.5 → graphiti_core-0.11.6rc1}/graphiti_core/search/search_utils.py

@@ -101,7 +101,6 @@ async def get_mentioned_nodes(
             n.uuid As uuid, 
             n.group_id AS group_id,
             n.name AS name,
-            n.name_embedding AS name_embedding,
             n.created_at AS created_at, 
             n.summary AS summary,
             labels(n) AS labels,
@@ -128,7 +127,6 @@ async def get_communities_by_nodes(
         c.uuid As uuid, 
         c.group_id AS group_id,
         c.name AS name,
-        c.name_embedding AS name_embedding
         c.created_at AS created_at, 
         c.summary AS summary
     """,
@@ -172,7 +170,6 @@ async def edge_fulltext_search(
                      r.created_at AS created_at,
                      r.name AS name,
                      r.fact AS fact,
-                     r.fact_embedding AS fact_embedding,
                      r.episodes AS episodes,
                      r.expired_at AS expired_at,
                      r.valid_at AS valid_at,
@@ -242,7 +239,6 @@ async def edge_similarity_search(
                     r.created_at AS created_at,
                     r.name AS name,
                     r.fact AS fact,
-                    r.fact_embedding AS fact_embedding,
                     r.episodes AS episodes,
                     r.expired_at AS expired_at,
                     r.valid_at AS valid_at,
@@ -301,7 +297,6 @@ async def edge_bfs_search(
                     r.created_at AS created_at,
                     r.name AS name,
                     r.fact AS fact,
-                    r.fact_embedding AS fact_embedding,
                     r.episodes AS episodes,
                     r.expired_at AS expired_at,
                     r.valid_at AS valid_at,
@@ -341,10 +336,10 @@ async def node_fulltext_search(
 
     query = (
         """
-                                        CALL db.index.fulltext.queryNodes("node_name_and_summary", $query, {limit: $limit}) 
-                                        YIELD node AS n, score
-                                        WHERE n:Entity
-                                        """
+                                                CALL db.index.fulltext.queryNodes("node_name_and_summary", $query, {limit: $limit}) 
+                                                YIELD node AS n, score
+                                                WHERE n:Entity
+                                                """
         + filter_query
         + ENTITY_NODE_RETURN
         + """
@@ -510,7 +505,6 @@ async def community_fulltext_search(
             comm.uuid AS uuid,
             comm.group_id AS group_id, 
             comm.name AS name, 
-            comm.name_embedding AS name_embedding,
             comm.created_at AS created_at, 
             comm.summary AS summary
         ORDER BY score DESC
@@ -555,7 +549,6 @@ async def community_similarity_search(
                comm.uuid As uuid,
                comm.group_id AS group_id,
                comm.name AS name, 
-               comm.name_embedding AS name_embedding,
                comm.created_at AS created_at, 
                comm.summary AS summary
            ORDER BY score DESC

{graphiti_core-0.11.5 → graphiti_core-0.11.6rc1}/graphiti_core/utils/maintenance/community_operations.py

@@ -239,7 +239,6 @@ async def determine_entity_community(
     RETURN
         c.uuid As uuid, 
         c.name AS name,
-        c.name_embedding AS name_embedding,
         c.group_id AS group_id,
         c.created_at AS created_at, 
         c.summary AS summary
@@ -258,7 +257,6 @@ async def determine_entity_community(
     RETURN
         c.uuid As uuid, 
         c.name AS name,
-        c.name_embedding AS name_embedding,
         c.group_id AS group_id,
         c.created_at AS created_at, 
         c.summary AS summary

{graphiti_core-0.11.5 → graphiti_core-0.11.6rc1}/graphiti_core/utils/maintenance/edge_operations.py

@@ -91,7 +91,6 @@ async def extract_edges(
 
     extract_edges_max_tokens = 16384
     llm_client = clients.llm_client
-    embedder = clients.embedder
 
     node_uuids_by_name_map = {node.name: node.uuid for node in nodes}
 
@@ -184,8 +183,6 @@ async def extract_edges(
             f'Created new edge: {edge.name} from (UUID: {edge.source_node_uuid}) to (UUID: {edge.target_node_uuid})'
         )
 
-    await create_entity_edge_embeddings(embedder, edges)
-
     logger.debug(f'Extracted edges: {[(e.name, e.uuid) for e in edges]}')
 
     return edges
@@ -238,9 +235,13 @@ async def dedupe_extracted_edges(
 async def resolve_extracted_edges(
     clients: GraphitiClients,
     extracted_edges: list[EntityEdge],
+    episode: EpisodicNode,
 ) -> tuple[list[EntityEdge], list[EntityEdge]]:
     driver = clients.driver
     llm_client = clients.llm_client
+    embedder = clients.embedder
+
+    await create_entity_edge_embeddings(embedder, extracted_edges)
 
     search_results: tuple[list[list[EntityEdge]], list[list[EntityEdge]]] = await semaphore_gather(
         get_relevant_edges(driver, extracted_edges, SearchFilters()),
@@ -258,10 +259,7 @@ async def resolve_extracted_edges(
         await semaphore_gather(
             *[
                 resolve_extracted_edge(
-                    llm_client,
-                    extracted_edge,
-                    related_edges,
-                    existing_edges,
+                    llm_client, extracted_edge, related_edges, existing_edges, episode
                 )
                 for extracted_edge, related_edges, existing_edges in zip(
                     extracted_edges, related_edges_lists, edge_invalidation_candidates, strict=True
@@ -322,9 +320,10 @@ async def resolve_extracted_edge(
     extracted_edge: EntityEdge,
     related_edges: list[EntityEdge],
     existing_edges: list[EntityEdge],
+    episode: EpisodicNode,
 ) -> tuple[EntityEdge, list[EntityEdge]]:
     resolved_edge, invalidation_candidates = await semaphore_gather(
-        dedupe_extracted_edge(llm_client, extracted_edge, related_edges),
+        dedupe_extracted_edge(llm_client, extracted_edge, related_edges, episode),
         get_edge_contradictions(llm_client, extracted_edge, existing_edges),
     )
 
@@ -356,7 +355,10 @@ async def resolve_extracted_edge(
 
 
 async def dedupe_extracted_edge(
-    llm_client: LLMClient, extracted_edge: EntityEdge, related_edges: list[EntityEdge]
+    llm_client: LLMClient,
+    extracted_edge: EntityEdge,
+    related_edges: list[EntityEdge],
+    episode: EpisodicNode | None = None,
 ) -> EntityEdge:
     if len(related_edges) == 0:
         return extracted_edge
@@ -391,6 +393,9 @@ async def dedupe_extracted_edge(
         else extracted_edge
     )
 
+    if duplicate_fact_id >= 0 and episode is not None:
+        edge.episodes += episode.uuid
+
     end = time()
     logger.debug(
         f'Resolved Edge: {extracted_edge.name} is {edge.name}, in {(end - start) * 1000} ms'

{graphiti_core-0.11.5 → graphiti_core-0.11.6rc1}/graphiti_core/utils/maintenance/node_operations.py

@@ -72,7 +72,6 @@ async def extract_nodes(
 ) -> list[EntityNode]:
     start = time()
     llm_client = clients.llm_client
-    embedder = clients.embedder
     llm_response = {}
     custom_prompt = ''
     entities_missed = True
@@ -165,8 +164,6 @@ async def extract_nodes(
         extracted_nodes.append(new_node)
         logger.debug(f'Created new node: {new_node.name} (UUID: {new_node.uuid})')
 
-    await create_entity_node_embeddings(embedder, extracted_nodes)
-
     logger.debug(f'Extracted nodes: {[(n.name, n.uuid) for n in extracted_nodes]}')
     return extracted_nodes
 
@@ -235,7 +232,6 @@ async def resolve_extracted_nodes(
             search(
                 clients=clients,
                 query=node.name,
-                query_vector=node.name_embedding,
                 group_ids=[node.group_id],
                 search_filter=SearchFilters(),
                 config=NODE_HYBRID_SEARCH_RRF,
@@ -294,7 +290,6 @@ async def resolve_extracted_node(
                 'id': i,
                 'name': node.name,
                 'entity_types': node.labels,
-                'summary': node.summary,
             },
             **node.attributes,
         }
@@ -388,7 +383,7 @@ async def extract_attributes_from_node(
         'summary': (
             str,
             Field(
-                description='Summary containing the important information about the entity. Under 500 words',
+                description='Summary containing the important information about the entity. Under 250 words',
             ),
         )
     }
@@ -440,10 +435,7 @@ async def dedupe_node_list(
         node_map[node.uuid] = node
 
     # Prepare context for LLM
-    nodes_context = [
-        {'uuid': node.uuid, 'name': node.name, 'summary': node.summary}.update(node.attributes)
-        for node in nodes
-    ]
+    nodes_context = [{'uuid': node.uuid, 'name': node.name, **node.attributes} for node in nodes]
 
     context = {
         'nodes': nodes_context,

{graphiti_core-0.11.5 → graphiti_core-0.11.6rc1}/pyproject.toml

@@ -1,7 +1,7 @@
 [project]
 name = "graphiti-core"
 description = "A temporal graph building library"
-version = "0.11.5"
+version = "0.11.6pre1"
 authors = [
     { "name" = "Paul Paliychuk", "email" = "paul@getzep.com" },
     { "name" = "Preston Rasmussen", "email" = "preston@getzep.com" },
@@ -19,7 +19,6 @@ dependencies = [
     "tenacity>=9.0.0",
     "numpy>=1.0.0",
     "python-dotenv>=1.0.1",
-    "graph-service (>=1.0.0.7,<2.0.0.0)",
 ]
 
 [project.urls]